Chris Reynolds reveals how Indeed's search experience, led by AI and user intent, is rewriting the rules of SEO. If you're in SEO, content strategy, or digital growth, this episode is a must-listen.
https://page2pod.com - In this powerhouse episode of the Page 2 Podcast, we sit down with Chris Reynolds, SEO Director at Indeed and former SEO leader at eBay, to unpack how massive enterprise websites manage SEO at scale.
Chris dives into:
⢠Leveraging machine learning for indexability decisions
⢠The trade-offs between internal linking and XML sitemaps
⢠How Googleâs indexing API and Google for Jobs are changing the SEO game
⢠What happens when LLMs become your discovery layer, not just search engines
⢠The challenges of handling billions of URLs, bad listings, and scam job content
⢠How prompt data and LLM visibility tools are shaping modern SEO tactics
⢠The future of SEO in a world led by AI agents and semantic search
Whether youâre an SEO veteran or trying to wrap your head around algorithmic indexing decisions at the highest scale, this episode is a masterclass in modern SEO strategy.
đ In This Episode:
⢠đ§ How machine learning drives SEO strategy at Indeed
⢠đď¸ Differences and similarities between SEO for eBay and Indeed
⢠đ Indexing decisions based on user behavior and job quality
⢠đ Scaling SEO with crawl budget and internal linking strategies
⢠đ¤ The emerging SEO impact of LLMs and AI assistants
⢠đ Using log files and synthetic prompts to track LLM citations
⢠đşď¸ Navigating dual audiences: job seekers vs employers
⢠đ The real influence of brand and PR in SEO success
⢠đŹ How FAQs and structured content affect visibility in LLMs
Chris reveals how a search experience led by AI and user intent is rewriting the rules of SEO. If you're in SEO, content strategy, or digital growth, this episode is a must-listen.
â Subscribe for more expert SEO insights every week!
đŹ Whatâs your take on SEOâs future with LLMs like ChatGPT and AI overviews? Drop your thoughts below!
đ Useful Resources & Links from the Episode
⢠Chris Reynolds on Linkedin
⢠Chris' personal website Q for Query
⢠Indeed Hiring Lab Data Portal
00;00;00;00 - 00;00;21;05
Unknown
What does it take to keep 340 million monthly visitors landing on the right job listings without overwhelming Google's index? Chris Reynolds is the SEO director at Indeed, the world's largest job site. Before that, he spent nearly a decade at eBay. At first glance, those two sites might seem worlds apart. One sells products, the other post jobs. But in terms of search strategy, they're surprisingly similar.
00;00;21;08 - 00;00;47;16
Unknown
And Chris, he's brought a very product-driven, very technical SEO mindset to both. This episode is about how one of the internet's biggest platforms handles SEO at scale. Chris walks us through how Indeed uses machine learning to decide which pages should even be indexable. Think of it as a regression model with SEO traffic as the dependent variable. We get into crawl budgets, the role of internal linking versus XML sitemaps, and how user behavior inside Indeed helps drive what shows up in Google.
00;00;47;18 - 00;01;08;00
Unknown
We also talk about what happens when LLMs, not just search engines, become your discovery layer. Chris is doing SEO in a world where traditional tactics aren't enough, and user expectations are shifting toward AI generated answers, not just build links. Whether you're managing an enterprise site or just trying to turn user behavior into smarter SEO decisions, this episode is a masterclass in scaling with intent.
00;01;08;02 - 00;01;08;27
Unknown
Let's get into it.
00;01;18;14 - 00;01;34;28
Unknown
Welcome to another exciting episode of the Page 2 Podcast, episode 97. I'm your host, Jon Clark, and as always, joined by my partner in crime at Moving Traffic Media, Joe DeVita. - Hey. - And today we're welcoming Chris Reynolds to the show. Welcome, Chris. Yeah, good to be here. I was, I was doing some research on
00;01;34;28 - 00;02;00;16
Unknown
Indeed. And these are similar web numbers, but 340 million monthly visits, ranked number 55 in the world. Number 30 in terms of website size in the US. And then I was going through your your job background and eBay. Right. Even bigger. So I'd love to learn a little bit more about maybe the transition from eBay to indeed, I guess what you could traditionally call like it an e-commerce site.
00;02;00;18 - 00;02;17;21
Unknown
to something like indeed, where more job driven. I'm sure there's a lot of similarities, but sort of take us through what that transition was like. Yeah, I mean, honestly, it was very, very similar. So the structure of a job site or a job aggregator site in many ways is very analogous to like a large e-commerce site. So you have individual jobs and I guess to analyze it for sale.
00;02;17;26 - 00;02;34;13
Unknown
And then you have, let's say a search page and a job site, which is very analogous to, analogous to a category page. So actually a lot of the stuff that I did at eBay was like 100% applicable to Indeed in many ways, I feel like I've been speedrunning everything on under eight years at eBay for the last four years.
00;02;34;13 - 00;02;59;21
Unknown
Indeed. So it was it was a good running start into indeed, let's put it that way. That's amazing. It seems like, on the indeed side, there might be a little bit more sort of programmatic SEO. Is that fair versus like individual like product listings? Yeah. Yes and no. The sort of fundamental difference in indeed and eBay's approach to SEO is that, eBay has a very well built out ontology of items.
00;02;59;21 - 00;03;27;04
Unknown
So effectively a taxonomist defines like what goes into Google's index, whereas eBay, Indeed is much more led by user searches because as you said, it's a huge destination site, the biggest job site in the world by far. People come and SEO is a minority of the traffic, a relatively small ish minority, the traffic. So we have all this insight into what users are searching for, and so we let users more or less determine what is the content that we put into Google, if that makes sense.
00;03;27;07 - 00;03;46;05
Unknown
Yeah, yeah it does. I can imagine there's quite a few challenges just in terms of indexation of really both sites, right? Just being so large. We work we work with a lot with real estate sites and right, you have properties that sort of come and go on the eBay side and you have products that are or listings have sort of come and go
00;03;46;07 - 00;04;04;19
Unknown
jobs, certainly the same. Do you have sort of a blanket methodology in terms of how you approach those types of things? or are they unique to, I don't know, maybe how popular the product or the job is? Yeah. I mean, the core page type, the indeed use it uses for SEO, if you like, is a job search.
00;04;04;19 - 00;04;25;26
Unknown
So it's an aggregation of jobs rather than any individual job. And behind that, and we spent the last like two years really building quite a sophisticated machine learning model that determines whether or not a page should be in our URL, should be indexable. So we use all these kind of factors around, as I say, people coming from other places and searching the number of jobs visible, how much they interact with them, how much they seem to like them.
00;04;25;29 - 00;04;52;05
Unknown
All those go into a model and then itâs continually learning about what gets indexed and not. So that's the kind of the the beating heart of it. And then for individual jobs, because we participate in the Google for jobs platform, a lot of those jobs are indexable, but not all of them, because kind of one of indeed, I guess, core competencies is, I guess, judging the quality of a job says a whole universe of, you know, bad and scam and low quality jobs out there.
00;04;52;06 - 00;05;07;18
Unknown
So upstream of even before it gets to a job search on Indeed there's a whole process where often it's looked at by a human or various machine learning models look at it and assess its value, and only at that point, when it gets through all those hoops do we say, yeah, that's a good job that we'd want to put on the site.
00;05;07;18 - 00;05;24;05
Unknown
And if it's good enough to go on the site, it's usually good enough to go into Google's index as well. That's super interesting. So you're so you don't publish everything that's submitted? No. Well, I mean, this, I mean, if you think about the kind of things that jobs that people might put on online, there's a whole world of like scams
00;05;24;05 - 00;05;39;05
Unknown
and I don't want get into it too much, but you know, when you really lift the cover of the huge bucket of jobs, you can look into oriental systems and say, jobs that we don't publish. And you look at you like, My God, that's a lot of dodgy stuff in there. And so it's almost I mean, these have been around for like 20 years now and sort of a core competency.
00;05;39;05 - 00;05;56;13
Unknown
The organization is saying, I, we scrubbed the whole web for every job is out there. But whatâs actually a good job that people might want to actually apply for. And so, yeah, by the time that it gets to visibility on Indeed internal search or something that we might want to put into Google's index, it's gone through a whole bunch of checks before then to make sure that it's a good job.
00;05;56;19 - 00;06;13;21
Unknown
And as I say, on the kind of the aggregation of job side, then we it's it's a relatively small minority of the total number of unique intents that we know about, if you like, you know, combination of a job title or a contract type or a skill or any kind of number of keyword searches, if you like, plus a location.
00;06;13;21 - 00;06;37;10
Unknown
And then we know about a very large number of those. And then we put a relatively small minority of them into Google's crawl path and index and so on. Thatâs super interesting. I was, I don't remember where I saw it, I'm sure it was LinkedIn. One of the, one of the job hacks was, rather than providing a link for someone to actually submit a resume, the suggestion was rather than to submit a resume through, you know, a link.
00;06;37;12 - 00;06;54;09
Unknown
The suggestion was to go to Google, search the brand plus the job title, right. So you're, you're sort of generating brand searches. And then to go to the website directly. And so that one's I don't know if you call that spam per se, but I would imagine that would get flagged in your, in your list, machine learning checks.
00;06;54;09 - 00;07;22;06
Unknown
Yeah. I mean, I was there's that from the context, from the perspective of a job seeker trying to get their application through of an employer trying to get more applications. I think it was honestly more it was less about those two things and more about generating brand searches in Google. Oh, I see, I see, I see, yeah. I mean, yeah, there's a pretty there's a pretty well built out set of independent variables that we use to kind of determine indexability, that you can push really hard on one.
00;07;22;06 - 00;07;41;08
Unknown
But like the it's kind of, I guess overwhelmed by the others. It means. So it's very hard to spam, I guess. Basically. Yeah. Yeah, totally. I'd love to learn a little bit more about the the job sort of API relationship with Google. I guess maybe to start like, what does that impact then on getting new job listings indexed?
00;07;41;08 - 00;08;02;07
Unknown
Is it near immediate or was there any uplift in visibility for those new listings? Yeah. I mean, indeed, historically didn't index individual jobs. And the rationale was really around cool budget. Like it's actually indeed great to work for as an SEO because that the kind of the lineage of it is SEO. So I occasionally get to speak to very, very rare to get to speak to the CEO.
00;08;02;09 - 00;08;23;07
Unknown
And he'll ask me questions about like index bloat and stuff. And I'm like, well, hang on. You see, you have a major problem. Yeah. What do we think about this stuff? But so even the most senior management in the company, like from very early on, was concerned about sort of SEO hygiene historically, the tactic was just not to have individual jobs indexed in order to save cool and indexation budget and so on.
00;08;23;09 - 00;08;42;02
Unknown
And we've to make tons and tons of changes. I mentioned to the ML model we did it cleaned up a whole bunch of internal linking and stuff. So we've effectively I, I see it as we've bought us a bit of space, in the index and in the core budget to, to put new stuff in there. So with that, we've had most of all of the jobs indexed or is indexable.
00;08;42;02 - 00;08;59;08
Unknown
And then we really started using the indexing API specifically when moving into Google for jobs, who for jobs is like it's not like huge in terms of a portion of traffic that we get, but it's it's a good way of getting jobs there from to job seekers. Some people use it. And for those people that do use it, then yeah, we can help them out getting a job.
00;08;59;08 - 00;09;16;25
Unknown
But for their jobs there. So some of it I kind of see it almost the indexing API. And for that matter, index now being sort of semi open source alternative is kind of like a misnomer with it's not really because the name kind of suggests that you just kind of pop your order into Google's index, but really it's about getting into front of the queue I think.
00;09;16;25 - 00;09;32;10
Unknown
So I use this like analogy of a nightclub, which is now you're all there. I want to get to the nightclub, which is Google, Bing, or whoever's index and what something like the indexing API will do is take you from the back of the queue and put you at the front, but the bouncer still has to look at you and check that you're a good fit.
00;09;32;10 - 00;09;51;10
Unknown
You know, you be of the right age or whatever. You meet the criteria to get in and only then can you get in. So the crawl is almost immediate. And so both of those services are but the indexation is not certain. But it's kind of really to participate in Google for jobs as a job site. Using the indexing API is kind of a must really.
00;09;51;16 - 00;10;12;17
Unknown
Just the scale of it is such that crawl is not going to be. As you mentioned, jobs are quite transient and crawl is just to slow. So that's hence why they have this indexing endpoints. And so that's why that's why we use it. So it hasn't been like a massive business game changer for us. But again, if it's useful in terms of helping job seekers find jobs to put jobs in that placement, then we do it.
00;10;12;19 - 00;10;31;27
Unknown
So there is still a, an authority or sort of a quality threshold, even with the index API, that still needs to. Yeah. Okay. Yeah, yeah, yeah. I was curious if you're using the API, are you also using XML sitemaps? Like, do you see those things as redundant or is it still valuable to have those listed across both of those items?
00;10;31;27 - 00;10;47;08
Unknown
Yeah. We do use XML sitemaps for all the major page types, including individual jobs. But it probably is redundant, to be honest with you, when using the indexing API. It's funny. Like I say, Indeed is a very strong tradition of SEO. And when I joined, the sitemaps were pretty weak. And I was like, oh, OK,
00;10;47;29 - 00;11;06;20
Unknown
why have we got all these errors? Whatâs going on? And the philosophy was really that internal linking is what counts. And if youâre falling back onto sitemap.xml for discoverability, it means youâre not doing your job properly in internal linking. And so it was like a crutch that you weren't allowed to use. And so it was like, yeah, some examples of the weak . Just use your internal linking, get it right.
00;11;06;20 - 00;11;26;27
Unknown
We did do a little work to get sitemap.xml just reliable and complete and stuff. I actually find them most useful for reporting actually. So we subdivided, obviously, by geography and page type and subcategory of page, like a job category, for example. And super interesting to see what crawl and indexation Google is devoting to those sub-pages.
00;11;27;02 - 00;11;51;09
Unknown
Thatâs probably, if anything, thatâs probably the most useful thing for some XMLs for me. Yeah, when we start with a client, one of the first things we ask them to do is to separate out different Google Search Console accounts, different XML sitemaps based on site structure, specifically for that reason. It's A, on the reporting side, itâs much easier, but on the indexation and quickly identifying key areas of the site that there's challenges.
00;11;51;09 - 00;12;12;27
Unknown
Super, super helpful. I wanted to talk about job schema a little bit. I have not worked a ton with it. So I know that there are unique attributes to apply for like a remote job versus something who has to go to the office every day. And I know that sometimes not applying that correctly can result in maybe not showing for what you would want to.
00;12;12;28 - 00;12;28;13
Unknown
Can you can you talk a little bit about that? Yeah. I mean, there's a minimum, minimum set of fields that are required to appear in Google for job. So use those on Google for jobs, sorry, on individual job pages. And then we go a little bit beyond that. But I might get like cast out to the SEO community for saying this really.
00;12;28;13 - 00;12;48;23
Unknown
But I kind of almost see itâs an exchange of value with Google because I mean, this attribute is not on the page visible to users. It's not helping anyone directly get a job. And so Google said, hey, tell us absolutely everything you know about all your products. Like from my perspective, unless that's actually helping people get a job, I'm not going to go out of my way too much to kind of give them more context and information than necessary.
00;12;49;00 - 00;13;09;27
Unknown
So we did a whole bunch of testing in the early days of Google for jobs where we just added additional fields, added new schema to tinker with the values and stuff. Honestly, it really wasn't a I donât think we had any kind of real material wins in terms of new traffic coming in as a result of that. So my philosophy is really being, hey, let's just give them what they would a little bit more than that, but not much and make sure it's accurate.
00;13;09;27 - 00;13;24;05
Unknown
And then that's it really. So in terms of schema for jobs we do almost the bare minimum. And then of course we do. Well not of course I mean but we also use schema across various different parts of the site for FAQs and stuff like that as well. So we do we do dotted around the site.
00;13;24;05 - 00;13;43;05
Unknown
But I know there's like some people it sounds like you're pretty expert in this area, but some people who really go to town with like overlapping semantic relationships and stuff. But from my perspective, kind of just I just kind of, I do whatâs required and no more basically. Yeah, thereâs typically just a small handful of required fields.
00;13;43;07 - 00;14;05;04
Unknown
And then of course, like you said, you can go nuts with layering and all that sort of stuff. I wanted to talk a little bit about keyword research. Like is that even applicable for a site like yours where, you know, really it's the job title and you maybe are just bound by consumer interest in that particular title, or maybe that, you know, that company per se.
00;14;05;06 - 00;14;28;03
Unknown
Like, how do you guys think about, you know, keyword research? Yeah, that's a good point. And the answer is not, not really keyword research in the traditional sense that you go and look at a tool and see what people are searching for. Yeah. So the reason I say the really nice thing about indeed being this huge monster website, which is a destination site, is you get a whole bunch of people to show up to the site and then end to keyword search folks, and that is that exact keyword research that a market inside.
00;14;28;05 - 00;14;53;15
Unknown
And as I say, we have this, machine learning model that continually runs and continue tries to improve itself at predicting which of those intents, as I said, like marketing jobs in Austin, Texas, for example, is that going to get any kind of SEO traffic? Does not care about that on Google or not. And we use quite a few different, independent variables that say that are predictive of that dependent variable, which is the SEO sessions.
00;14;53;15 - 00;15;11;14
Unknown
And the nice thing about it is, is like as new sort of job titles or locations or anything comes up, it's effectively the users doing keyword research for us, you know, so people always jump to the, example of like, oh, prompt engineering to new jobs. You know about that? Yeah. And you're like, yeah, well, yeah, people do search for it.
00;15;11;14 - 00;15;27;25
Unknown
And so yes, we do have a rank for it. And itâs all good. But more or more specifically, it tends to be like, I don't know, I imagine a new restaurant opens in a small town or something. And then people drive by and see a new restaurant is opening, and I mean, jobs in use, more restaurants. And then effectively, if enough people search for that, weâll be like, this is something that people care about on Google.
00;15;27;25 - 00;15;50;20
Unknown
And so weâll have that linked and indexed at the web poster site. It's really nice. And then vice versa. You know, if people stop caring about a specific job title and company goes bankrupt or stopped hiring or whatever, we'll take that out of, out of the, Google SERPs. So itâs kind of a self, I think of it like a plant that grows towards the light. If the light is SEO traffic, then effectively, the site kind of grows itself and we just make sure it doesnât do anything terrible.
00;15;50;20 - 00;16;15;21
Unknown
I was really curious about, so the job title, How do you decide what jobs to match to if you're able to talk, I mean, not the actual formula, but like, maybe Iâll ask it a different way. is your machine learning model understanding that, you know, SEO director might have sort of a similar relationship to, I don't know, a senior manager.
00;16;15;21 - 00;16;41;09
Unknown
Right. Like SEO senior manager. And so if I do a search for SEO director, you're sort of combining the listings across those types of jobs. Is that, like a fair way to think about how you're trying to map jobs together? Because I think that one of the challenges that must be is that companies have, you know, it's like e-commerce, right, where you sort of name the product, some name that has nothing to do with the actual thing.
00;16;41;09 - 00;17;05;21
Unknown
And so the consumer doesn't really know what to find. And I imagine that must be somewhat true for companies as well. So how do you guys think about sort of making sure you're elevating those types of listings as well? Yeah. So there's two things. One is is there's what appears on indeed internal search. So if someone has I don't know, SEO manager, SEO director, senior SEO specialist or whatever.
00;17;05;26 - 00;17;25;10
Unknown
And and we think, hey, that was a probably the same thing based on the skills and the content of the job site will basically make sure that if someone searches for SEO director and then, we have a kind of a head of SEO job that will show that. And it's kind of exactly like any search engine like Google, you know, Yelp or Amazon or whatever.
00;17;25;10 - 00;17;47;11
Unknown
There's a whole, whole process to build synonym rings to understand, like what the, the, the, synonyms between the various terms are then in an ML component that again seeks commonalities between jobs. And so that all comes together to form the search results page. So you say like, you know, director of SEO is your job search. And then on that page as a head of SEO job, because we think that might be useful.
00;17;47;13 - 00;18;11;20
Unknown
So then you've got a page which is Director of SEO, say in New York or something, and then the next, the next step is what do we want that to go into Googleâs index enough? I donât know if people care about those jobs to have that in Google and the machine learning model I mentioned earlier. Is that really so we're saying, all right, well, you know, there's 200 people a month, coming to the site searching for director of SEO jobs in New York.
00;18;11;20 - 00;18;24;22
Unknown
And we've got a lot of jobs for it. And, you know, a lot of people end up applying. So it's not it's not a huge balance, right? All these things come together in like ten other things. And so, yeah, that's a good page. We want that to be in Google. And only then we show up internal linking, some XML
00;18;24;22 - 00;18;44;23
Unknown
we'll take off the no index tag and stuff. So brilliant. I donât understand, thatâs machine learning or this is manual intervention to try to crop. Is it machine learning or is it manual, youâre making the decisions by yourself as a team. The indexation is machine learning. Yeah. So so I mean not to get too much in and to give you the secret sauce, but basically there's a dozen variables.
00;18;44;23 - 00;19;06;24
Unknown
I mentioned a few of them. So like number of people coming to a dean searching for something, how much traffic it's got, retrospectively, you know, what kind of traffic it gets from other paid media, etc., etc., etc. and so those all come together into a score. That score, is effectively I get beaten up by a data analyst for saying this, but basically itâs effectively a prediction of how much SEO traffic that individual is going to get.
00;19;06;24 - 00;19;24;17
Unknown
And then each market has a threshold. So it was out. Right. I expect, you know, if itâs going to get more than X traffic, then it goes into contention for getting into internal placements, the sum of X amount and so on. There's a bit more logic on top of that. There's an interesting trade off there.
00;19;24;17 - 00;19;44;03
Unknown
Then you've got like to say I, I'm just making these numbers up. So please don't quote them. But let's say youâve got a billion potential intents that you might want to put into Google that you know about.So marketing director in New York or whatever, accountant in London, whatever. And then you say, all maybe there's 10 % of those that you think, okay, yeah, that's probably going to get some SEO traffic.
00;19;44;03 - 00;20;02;28
Unknown
And so those go into the signup XMLs and all the rest of it. And then after that, we do some stuff with that SEO score around weighting internal link placements and stuff like that. So something thatâs very high potential, get traffic, gets more links and so on. But it's a bit, yeah, there's a tension between like high SEO score and relevance of the link placements.
00;20;03;03 - 00;20;19;17
Unknown
Gets a bit nuanced there. But yeah, that's fundamentally how you can think of it. It's just a big regression analysis at its core. Itâs a big regression analysis, gives you a number. There's a threshold for that number. If it's above it, then it goes onto the sun, basically. Do you follow that same approach when you have a job title you've never seen before?
00;20;19;20 - 00;20;36;24
Unknown
Yeah, yeah, so it goes on so if you go on and you search for you know, Chief yogurt weaver or something like that would be like, okay, that's interesting but if you if if no one else ever searches for that then obviously won't get through but if you know hundreds thousands of people start showing up and searching for that search then weâll be like ah,
00;20;36;24 - 00;20;50;27
Unknown
this must be a job people care about and assuming I'm kind of being facetious but like assuming there were jobs for that and people were applying for them and having a good experience or whatever, then at that point the model will be like, looks like legit to me, so it goes onto the site.
00;20;50;27 - 00;21;21;28
Unknown
I hope youâre not giving away a trade secret here, Chris. We just had a conversation with a LinkedIn ads expert a couple of weeks ago who's telling us how poor LinkedIn's title targeting is with the advertising product where 50 % of the titles people use in their LinkedIn profile just don't exist anywhere else. They're like, Iâm the chief garbage man. Meanwhile, you're the CEO of a huge waste removal company. So there's one limitation of LinkedIn's advertising is just you can't reach people by their title because only one person with that title.
00;21;21;29 - 00;21;40;06
Unknown
That's fascinating. Just as a complete aside, a long, long, time ago, because I built this tool, have you ever heard of the concept of LinkedIn X-ray search? So basically what recruiters do sometimes, if they don't want to pay for a LinkedIn account, probably shouldnât say this if that guy's listening, but they'll do a site command in Google, so sitelinkedin.com, Chief Garbage Officer,
00;21;40;09 - 00;21;54;02
Unknown
And that brings back, obviously, the whole bunch of profiles. So I built basically a little form that you can basically structure those searches for us. Not everyone wants to go and write a big, Boolean string into Google, so you just go onto this site and you basically flex for hint, and it gives you the string to do that search.
00;21;54;08 - 00;22;10;08
Unknown
Yeah. it's been running along. It gets a few thousand users a day, and it's been running along for years now. So I've got this huge database that's creaking along that has all these different job titles in there. I should redo something with it, actually. But yeah, I do see a lot of, in that context as well.
00;22;10;08 - 00;22;28;04
Unknown
I see a lot of interesting job titles. And also I have to say, again, I probably shouldn't say this, but a whole bunch of profanities. I don't know what the audience for this podcast is, so Iâm going to start quoting them. But a lot of we had to work really hard to get like profane job searches. A lot of people search for stuff which is.
00;22;28;04 - 00;22;45;18
Unknown
Yeah, not safe, not safer, not safer work, let's say. So, yeah. It's like adult industry, is that what mean? Like along those lines? That and also Can I curse in this podcast? Is that okay? Is it an adult audience? So people which like like I want a shit job a shit job, whatever.
00;22;45;18 - 00;23;05;24
Unknown
And then you end up saying, if enough people say that, then the model left its own devices. like, hey, shit jobs, want them. And so if you search for that, then you'll end up with shit jobs at top of Google, which just looks terrible. And so, yeah, again, we have yet another AI-based process, which basically goes through and goes, huh, that's kind of adult. We're going to knock that one out.
00;23;05;27 - 00;23;27;22
Unknown
So yeah, that's a little nuance of the industry. I pity the group searching for I'm looking for a shit job. You wouldn't believe it, right? Yeah. So to build these models, had to have human reviewers go through these things. And so we did a first pass and weaned out the ones we saw. Hang on, those look a bit suspect.
00;23;27;22 - 00;23;42;02
Unknown
And then these huge, great big lists of queries go over to the... Because obviously we operate in 60 different markets, so youâd have a lot of different languages to do, you pass that over to the Japanese language expert who'll go through and go, yeah, that's a problem. That's a problem. And I'd be like, I am so sorry. You look at some of these queries
00;23;42;02 - 00;23;59;17
Unknown
and some lovely person who spends their time normally translating microcopy on tooltips or something, youâre like, hey, I've just sent you a massive list of profanities. I'm so sorry. Just do your And so now it's more or less automated. We just kind of review it semi-regularly. But yeah, that's a nuance of the field.
00;23;59;19 - 00;24;21;07
Unknown
Maybe another nuance is, I mean, you technically have two audiences, right? You have the businesses that you have to pay attention to, to some degree, because they're the ones who are doing the listings. Certainly the job seekers are what sort of provides really the value of the site. I guess, how do you, so if you go to the homepage today, it's clearly geared toward, you know, a job seeker..
00;24;21;14 - 00;24;46;14
Unknown
How do you guys think about, you know, those relationships with the businesses, do you create sort of separate content for them or?. Yeah, I mean, it's interesting actually by market, thereâs root terms in, let's say, recruitment industry, which are pretty consistent. For example, in the US and most English speaking countries, and actually some non-English speaking countries, like surprisingly France, for example, it's just the word jobs.
00;24;46;14 - 00;25;09;04
Unknown
So people will search for skill, contract type, or company, or whatever, jobs. And that's really easy to target because just add that as a suffix to the query or whatever. And for employers It's a bit more nuanced, but typically thinks something like higher app thing. And so we just create content. On the employer side, I don't directly look after that, but it's mostly informational content.
00;25;09;06 - 00;25;28;29
Unknown
So if you search for like post a job, it's very heavily monetized, but we'll rank there, I think weâre number one. Post a job for a thing or just post a job generally. And then beyond that like the SEO element to visit, we'll just look at common employer questions and then we create long-form content that answers that.
00;25;29;09 - 00;25;45;05
Unknown
So we'll just do, if someone's looking for like, how do I set up a 401k for my employees? How do I do health insurance for my employees? So we'll basically work around that. But, honestly, on both cases, it's pretty straightforward. Then we do see people getting lost.
00;25;45;05 - 00;26;03;16
Unknown
So we'll see employers come job search and vice versa. And we've got like a guess a little process that says, well obviously weâve got links that say, hey, you're to post a job because it's over here. We have a little process where we really think someone's an employer and theyâre kind of wondering on job search last week, weâre like, hey, are you an employer? Do you want to post a job? Because it's over here. And we'll sort of intervene and try and push them to the right place.
00;26;04;11 - 00;26;24;25
Unknown
generally from an SEO perspective, quite straightforward actually. I was curious when you guys are creating content you use the example of, I donât know, set up a 401k for my employees? Do you guys think through the application of E-A-T, or do you rely on the strength of the domain to overcome some of that? More than that
00;26;24;29 - 00;26;48;11
Unknown
Really. Maybe Iâm sort of a bit old-fashioned. I guess when I think of traditional SEO E-A-T, itâs about things like author bylines and stuff like that â basically saying, âHey, Iâm an expert in this field.â And we do have, like, we have a career advice section thatâs targeted on how to write a rĂŠsumĂŠ for various positions, how to write a cover letter, how to resign gracefully from your job, workplace etiquette, stuff like that.
00;26;48;13 - 00;27;17;12
Unknown
And it's been really successful over the years. We basically said, all right, well, people who are looking for a job are typically finding us. But people who are maybe thinking about changing career, weâll try and help those guys too by giving them this content. And I think because of the strength of the domain, that long form content is tremendously well in Search. And for those articles, we did do that traditional EAT thing of having an author photo and byline and stuff.
00;27;17;12 - 00;27;32;15
Unknown
It's not my team that does that, it's an adjacent team, but I think it was more really about building credibility with the audience really. Because SEOs have kind of ruined the internet to a certain extent with just endless content. And so I think that team really wanted to be like, hey, we actually know what we're talking about here.
00;27;32;16 - 00;27;51;22
Unknown
I remember I commissioned some content from, this was way before the LLM revolution that we're living through now, but commissioned content on eBay, just for category pages. And it was like, all right, like, I fridges, digital fridges or something. And the content that came back is that when choosing a fridge, make sure the fridge fits in your kitchen. It must keep food cold.
00;27;51;22 - 00;28;10;26
Unknown
And I'm like, oh, god, this is so worthless. You know, like, it's just redundant information. And so I think the team that write this content for Indeed is very focused on making sure it's actually stuff that might be useful. And really, I think the byline stuff is really just about trying to say to users, hey, we're not just like spewing this out. We actually know what we're talking about. So thereâs that.
00;28;10;27 - 00;28;28;25
Unknown
But then, I mean, your point was exactly right. We really rely on the strength of the brand for that. So we've got an economist team they do, because we kind of know so many jobs, like a lot of the Federal Reserve and various central banks and stuff, rely on that for just looking at inflation and economic trends and stuff.
00;28;29;00 - 00;28;53;26
Unknown
And so if you go to data.indeed.com, there's a really cool sort of economist-focused data dashboard there that you can slice and dice. what's happening to job postings for tech jobs in California or something. And you can see over time, like salaries and trends and stuff like that. So we do stuff like that rather than explicit like E-E-A-T stuff. That's very cool. I didnât know that was open to the public.
00;28;53;26 - 00;29;13;16
Unknown
I mean, I always hear about economists talking about Indeed data. I just thought they had special permission from you guys, but you make it all public. Yeah. Make me. Yeah, pretty much. think we do, I don't know, it's kind of not my area again, but we have like, as I say, this team of economists, I think they do special stuff for like, if the Federal Reserve called, we'd probably give them some special stuff. But, we yeah.
00;29;14;13 - 00;29;30;06
Unknown
it's super depressing, man. I guess a lot of your audience like working tech, don't look at US tech jobs, it's about 2022, it's horrendous. So yeah, it's kind of, it's interesting, it's interesting. So more and more states across the US now are mandating salaries as partners job title, sorry, the job ad.
00;29;30;06 - 00;29;45;15
Unknown
And so we've got pretty good salary data as well. So we were actually part of the same company as Glassdoor. And so they have really good salary data and we kind of, you we make use of that too. you can kind of see salary trends and job posting. And then things like, I think it's like pretty ad hoc right now,
00;29;45;15 - 00;30;06;12
Unknown
but you can look at like, for example, if certain skills are being mentioned in a job title, so like how many jobs are mentioning AI and stuff. So it's quite cool. I should redo more to kind of build that into the core experience, I think. for now, yeah, we kind of do stuff like that to demonstrate expertise, I guess, and authority in that area, rather than telling people. I would imagine it's a nice little link hub as well.
00;30;06;12 - 00;30;27;14
Unknown
I mean, it's having that data available seems like a natural place for people to reference and point links to. Yeah, yeah, I'm not sure it's terribly well marketed. As I say, it's kind of like a, it's a team of economists that run it and they're not really part of the commercial team. So Iâm in the product team, kind of part of broader growth team. So ultimately I need to bring more job seekers into the site.
00;30;27;14 - 00;30;49;14
Unknown
But that team is really just focused on sort of like, you know, the academia of it rather than the growth of it. So it's not really, I think, exploited as a link opportunity. But yeah, you would imagine, I mean, think the St. Louis Fed, which is the Federal Reserve sort of, I donât know what term theyâre on, site, we tend to get cited by lot of journalists and stuff.
00;30;49;14 - 00;31;04;06
Unknown
I think that links out to Indeed data, certainly references Indeed data. So stuff like that, yeah, is super useful. You said you're part of the product team, not the commercial team? Yes, so part of the product team, that's right. So I've bounced around in my career a bit between marketing, like marketing growth teams and products,
00;31;04;06 - 00;31;25;03
Unknown
but yeah, I sit in the, and the whole SEO team sits in the product team. Exactly, yeah. How does how do you fit in within the product team I assume there's advertising, there's PR. Well, so my boss, for example, will be responsible for SEO, but also mobile apps and a lot of stuff, like maybe some data quality stuff,
00;31;25;03 - 00;31;41;20
Unknown
I can't remember now, but whatever. So yeah, there's various different things and inside that is SEO and our OKRs are around, in my case it would be JobSeeker, new JobSeekers coming in via organic So yeah, that's how we fit in. But then also we are also kind of part of the marketing organization.
00;31;41;20 - 00;32;02;04
Unknown
So marketing will then have like, know, paid search and display advertising, social, whatever. So our OKR kind of rolls up to there, if that makes sense. Yeah, I was, I wanted to dig into that. only because I think I saw some LinkedIn advertising on perplexity. Would that be right? If I saw a LinkedIn ad on perplexity?
00;32;02;06 - 00;32;26;23
Unknown
Yeah, Perplexity does run ads around specific queries. I don't think, like Perplexity just doesn't have traction with users right now. So we, I donât know what to just say, but we dabbled with it. Sponsored prompts, but it just really isn't very, like honestly, if the standalone LLM, so excluding Google, my both internal and extended system, the ChatGPT is really the only service that's got any kind of consumer traction really.
00;32;26;23 - 00;32;50;01
Unknown
There's ChatGPT and then there's everything else. Yeah. Did you have a chance? I guess. Sounds like it was just a short, maybe short advertising test with perplexity. I but but I wonder, is there an opportunity there for an SEO professional to try to do some, like, prompt discovery? Yeah. I mean, I, you know, I don't say too much, but yeah, we did kind of like they shared some anonymized prompts and stuff with us.
00;32;50;01 - 00;33;12;26
Unknown
And that was super interesting, actually, because we're all, I guess maybe weâre talking a second about LLM visibility monitoring and stuff. And the core problem is that unlike Search, where thereâs pretty well understood and well known, either the service providers themselves will give it out for ads or ISP-based tools like prompt or plugin-based tools like SimilarWeb or whatever will give you data on search terms.
00;33;13;01 - 00;33;34;20
Unknown
No one really knows truly outside these tools themselves what people are prompting them for, except we did get this data that kind of gave us a bit of an idea. Super interesting, super noisy, to be honest. I you know, youâve got this kind of classic shape of demand search, got a few head terms, some body, and then a lot of the classic long tail. Really, like, I think on these AI-based services, it's basically flat to the axes.
00;33;34;20 - 00;33;52;21
Unknown
Youâve got a tiny number of head terms, for want of a better term, and then basically an infinite number of prompts. And when you look at what people are prompting them for, a lot of it is so contextual. it's like, you look at a mid-prompt conversation. Itâs like, so how do I optimize my resume for this?
00;33;52;21 - 00;34;15;24
Unknown
You know, like, well, when the bots proceeds proceeded to eventually pointless. Or maybe there's a file associated with which you donât obviously And yeah, there's a relatively small number of prompts that we could take from that and go, yeah, that's something we want to actually look at. So maybe I got a two part question. The first, just to clarify, it sounds like the perplexity advertising tool, not really a help to SEO pros right now.
00;34;15;26 - 00;34;34;27
Unknown
Fair? I want to say, yeah, but I'm gonna say no. Honestly just because of the scale of perplexity I just feel that it's not that useful really but yeah it was interesting looking at some groups. Certainly this was like six months ago or something so it's kind of like it was useful then to get an intuitive sense of what people are using them for
00;34;34;29 - 00;34;54;18
Unknown
but I don't necessarily think that it would be something I would rely on necessarily. And then follow up, I saw a recent post you had about the UDM equals 50. Can you talk a little bit about that? Maybe. Yeah, so obviously, well, you know, as I see it, there are two players in this game right now. Maybe that'll change,
00;34;54;18 - 00;35;20;05
Unknown
but right now you've got Google and you've got OpenAI. Google's obviously pushing AI mode increasingly hard and there's this UDM parameter which just, it basically tells you or determines what experience Google puts on its surface, so images, news, jobs, whatever, has its own unique value for that UDM parameter. And so yeah, mean, just get me through. It looks like Google is using UDM50 for AI mode.
00;35;20;05 - 00;35;38;16
Unknown
so because unlike chat GPT, I always get this the wrong way around. So forgive me. But I think I'm right saying that Google is a post request. That is that the user submission is visible in the URL. That means that a tool like SimilarWeb or Comscore or whatever can then look at, does this URL contain UDM50?
00;35;38;16 - 00;35;55;18
Unknown
Yes, it does. All right, great. What was the prompt? What was the search query that prompted, I guess, users to go over to toggle into that experience. so, yeah, again, it was just in service of trying to understand what people are using this for. That was the intent. I think that's that's the challenge right now.
00;35;55;18 - 00;36;19;07
Unknown
Right. With a specific keyword, you have a pretty good sense of what the intent is. Maybe even if it's a little blurb like Transactionally an informational with LLMs. Just thinking about my own use, it is very specific. â And the follow-ups, the follow-up prompts don't necessarily carry that specificity because now you're talking about something within the response that's also specific.
00;36;19;07 - 00;36;35;24
Unknown
Right, yeah. And I see a lot of people are jumping on LLM visibility tools because I mean, it's really valid criticisms which will say things like, oh, it's completely contextual to the user. just putting a generic prompt in doesnât make any sense. Or it's obviously non-deterministic, so it might be different every time.
00;36;35;24 - 00;36;59;10
Unknown
Anyway. And therefore like, you know, why even do this? My, my personal kind of feeling is, is. Yes, that's true. As you mentioned, there's like infinite, infinite prompts and they're all extremely contextual. My feeling is really that it's still worth measuring just because there is, like, if you're ranking for the default, I guess, like the default query, so non-contextualized to that user.
00;36;59;10 - 00;37;16;27
Unknown
and for something, yeah, which might be less specific than the thing they're asking, then at least directionally, you can see compared to other competing brands, like how you're doing. So it's not, like, perfect. But it's at least an indication. And think it's a relatively, and it's kind of the same in Google. I mean, Google contextualizes or customizes
00;37;16;27 - 00;37;44;14
Unknown
your answer based on your previous search history. All these tools are coming in, they're doing a request from a given location. And thereâs no guarantee that your users are coming in from that location as well. So there's some of the same kind of pushback supply to them. So I think that picking some representative prompts and then measuring them is still valuable for looking at what's happening at the top of the funnel, if you like, for AI traffic. And we use like, we call them synthetic prompts, which is a nice way of saying we made them up.
00;37;44;14 - 00;38;07;09
Unknown
So we basically use like, people also ask some of that data I mentioned earlier, bit of common sense, and we kind of come together and that's how we build out our prompt tracking population. I was I was just going to ask that because a lot of, a lot of the AI tools that have come out, who will track this sort of stuff, rely on the customer to define the initial set of prompts.
00;38;07;11 - 00;38;27;22
Unknown
And I've signed up for a couple just to play around, and every time I'm hoping that they're going to suggest the prompts for me. So you mentioned, like, do you have any tips for anyone when they're starting to think about, like, what are a core set of prompts that I should, you know, track like, or do you start with the end term like jobs and then work down from there?
00;38;27;22 - 00;38;48;05
Unknown
Or do you really look towards something that's longer tailed? And I don't know, maybe something you're seeing gets shown up in Google Search Console that's, you know, ten words long and you could theoretically assume that's an AI mode query. Yeah. I mean, the two tools we use is are profound and SEO clarity. And both of them have a I don't mean you could like a prompt research feature right now.
00;38;48;12 - 00;39;10;22
Unknown
Honestly, I think it's pretty in like a nascent field. But they are saying that I think I'm right to say I'm regurgitating some. So maps has told me so maybe completely off base, a pinch of salt. But I believe the the ChatGPT plugin, I think does make a Post request is visible in the URL so ISP can pick up, and I believe some Android devices rather horrifyingly will actively keylock their users.
00;39;10;22 - 00;39;32;08
Unknown
And so I think from those, those sources, there are some providers now that will, give you some indication it's a different it's a different methodology actually. They when you for example, professionals at the if you search in profound for a prompt or do that prompt research, it will give you the very roughly estimated volume for that substring.
00;39;32;11 - 00;39;50;10
Unknown
So if you say, like, resume, itâs not telling you the number of people who have just gone to ChatCPT and searched for resume. it's people all the prompts that it knows about or has estimated, contain the word resume. So it's very different. So there are some tools there. Yeah, the one that I found super useful actually is People Also Asked
00;39;50;12 - 00;40;20;21
Unknown
And they've got some very good data on people. Also asked values. ChatGPT is super normalized and I don't know how useful it is, but it does have a suggested prompt function. So we use isitalsoasked.com, and theyâve got some really good data on the People Also Asked values. ChatGPT, itâs super normalized, and I donât know how useful it is, but it does have a suggested prompt function. So if you go and start typing your prompt, it'll give you some example prompts. And then yeah, just a dose of common sense, to be honest with you, on what people might be using it for. Letâs say for us, it was slightly informed by that raw prompt data we got. it gave us a bit of a direction. But a lot of it is really those different data sources put it all together and say,
00;40;20;21 - 00;40;45;01
Unknown
yeah, we think reasonably this is a good set of prompts. I wanted to ask about log file analysis, correct me if I'm wrong, on a relatively recent podcast, I'm forgetting who the host was, but you mentioned something about tracking ChatGPT user agent â as sort of like a proxy for citation mentions.
00;40;45;03 - 00;41;06;14
Unknown
Yeah, sure. I assume you're using log to some sort of like log file analysis to sort of pluck that out. Yeah, Users and bots are all logged. So all the analytics I look at is log file analysis. so, yeah, so it's relatively straightforward. It's actually amazing now because we do have some JavaScript instrumentation, like we have GA on the site, but we typically use it.
00;41;06;22 - 00;41;35;06
Unknown
So everything is logged into the user log, into the server logs. And then obviously we extract that and take subsets of that for different indices. We then actually query for our analytics. And it's amazing now because in the world of LLMs, we have LLMs. You can go and just do a natural language query. So show me all the LLM user agents that have queried this page type in Japan in the last 30 days or something. Get the SQL query straight away. Or actually, proprietary query language you have internally. whatever. Letâs say SQL query. And then run it, and it's great.
00;41;35;06 - 00;42;00;18
Unknown
And you can really get that fine-grained, nuanced control then. It's quite difficult with a GUI-based tool like GA. And yeah, exactly. So with that, ChatGPT you think not always, but often when ChatGPT cites a URL, it will send off a request at runtime to that URL. I assume itâs just to check the page is still live and the content is still what it was when it first indexed it. And then
00;42;07;18 - 00;42;22;05
Unknown
So it's not every time, but it seems to be a lot of times. So if you zoom out to a page level, like subset of pages, you can definitely see this is the number of times people are signed, category pages are being cited, individual job pages for me, and so on and so forth, home page.
00;42;22;08 - 00;42;39;12
Unknown
And so that gives you an an interesting insight into what's showing up. And then obviously you can see the clicks that come from that. And so therefore you can derive a rough CTR, and you can see conversion. And obviously you have that data I mentioned Profound. As you can see, whatever data I've talked which shows you the visibility.
00;42;39;17 - 00;43;04;12
Unknown
So it's very very nascent compared to SEO but you can start to see a kind of funnel between visibility to conversion a little bit more. Yeah. So you've you I feel like you're a little bit ahead of the game as far as big enterprise companies go in starting to already try to understand and optimize for LLMs. At this point, you probably have 1,000 prompts that are important to you and you're optimizing for them.
00;43;04;12 - 00;43;20;06
Unknown
Is that fair? To a degree, yeah, to a degree, it tends to be. Not so much. We're going after specific. Again, it's programmatic. So we try to do things at very large programmatic scale. So it's not like I'm like, oh, you know, we're you know, we're like the second most visible part time job is we should do something there.
00;43;20;10 - 00;43;42;17
Unknown
We try to do things that have a template level that will lift all boats. Does that make sense? So you you come up with your initial set of prompts that are important to you and you say, I mean, maybe it's profound that you used to say, how am I doing against these important prompts? And you say, oh my God, there's 900 of them that I'm like, I'm, I have some traction against like 90% of the prompts that are important to me.
00;43;42;22 - 00;44;14;27
Unknown
Then you've got to kind of reverse engineer, well, how did I, before I was doing any optimization for LLMs, like what did we do right? And what are the kinds of things that you're seeing in your reverse engineering of just by best practice from some other discipline you've had, you've had this early success. Honestly, the absolute truth of it is we started doing a lot of this in January and I had no preconceptions about how well indeed we'd be doing in these surfaces.
00;44;15;00 - 00;44;29;11
Unknown
And so we set it all up, as I we mentioned it, chose the prompts, etcetera, and I was like peeking at it and I was like, my God, we did really well. And I meant to do this, this was always my plan. And a lot of it's been born out of the pre-existing SEO success actually.
00;44;29;16 - 00;44;56;26
Unknown
So when ChatGPT, for example, was relying very Bingâs index, Indeed is extremely well in Bing, even better than Google. And so when ChatGPT was going off and doing its RAG thing to synthesize its answers, Indeed was just showing up a lot there. So it's really, been honestly, it's not a satisfying answer, but it's really been traditional SEO that's really, I guess, got us to the most visible site in that job space from day one.
00;44;57;03 - 00;45;13;29
Unknown
Now, in terms of things we're doing, honestly, again, nothing revolutionary. I think the whole world and this dog is now just doing FAQs across the site because they do really seem to be, like, just having that very plain answer. Like, here's the question, hereâs the answer. They seem to do really, really well when it comes to influencing responses. do really, really well, when it comes to influencing responses.
00;45;13;29 - 00;45;31;13
Unknown
So we're doing a lot of testing right now. We do something called cause and impact studies. So we'll basically take a bunch of pages and we'll say, all right, like, this is the metric that we care about. Maybe it's like, traffic from LLMs, maybe it's citations or mentions or just regular traffic or whatever.
00;45;31;19 - 00;45;47;22
Unknown
We'll have some metric we we care about. We'll grouped together a bunch of pages that typically perform similarly, and then we'll have like, like a synthetic control. So we say, okay, this is what happens. So you'd reasonably expect if we do this here this would continue to happen. But if we make this intervention here what happens. Oh it goes up.
00;45;47;22 - 00;46;04;24
Unknown
And therefore we'll be able to look at the difference pre-post and put a number on that. So that's, that's that's the methodology we use. And then we just try various interventions, a lot of FAQs to be honest with you, to try and influence that. LLM traffic is probably our least noisy metric actually, so we'll try and influence that metric, yeah.
00;46;04;24 - 00;46;21;12
Unknown
Yeah. I asked that question thinking I was taking you up to give a compliment to your PR. Yeah, yeah. Well, no, I think it is. I honestly, my my, my the dirty secret of SEO. I don't tell my boss is that I personally think this is just my observation. I've a lot of time working in it.,
00;46;21;14 - 00;46;35;26
Unknown
is that to do well in SEO, really, it's like, need to have a site that's indexable and all the rest into the right content on the head. Fine, thatâs the table stakes. Everyone can do that, really. But then you need to have a good product that people actually like. And that's probably like 40 % of the challenge.
00;46;35;29 - 00;46;55;00
Unknown
You're the 40 % I think is brand. So like the PR team, the brand teams going out there to encourage people to search for you. And then really the stuff that I do, tinker around with links and indexation and meditation and stuff like that all day, that's really the last 20%. Like if you have an awful site that no one knows about, then it matter how good you are at SEO, you'll just get nowhere. And vice versa as well
00;46;55;00 - 00;47;15;09
Unknown
actually, some awful sites. Craigslist, Craigslist still ranks from the jobs, it hasn't changed since 1995. And so, you know, you've people like you and they find it useful then youâre going to do well. So I guess, yeah, I don't say it because I want to sound important internally, but truly like the brand team is probably the real SEO team to be honest. You mentioned
00;47;15;12 - 00;47;40;04
Unknown
ChatGPT using Bing results. There's been a couple of case studies on LinkedIn and elsewhere that are saying it's only using Google results now. I've also heard that it still grounds the response in Bing results, but if it's something that Bing doesn't have a good response for, then it goes to Google. Do you have a strong sense of where that is today, like in terms of where they're actually getting their results from?
00;47;40;04 - 00;47;54;21
Unknown
Yeah, think, think the, so one of those people on LinkedIn was Gus Pellogio, who's kind of like a, almost like a R and D PM. So he does a lot of the kind of forward thinking stuff and yeah, he reached much the same conclusion that theyâre coming from Google. I believe, yeah, I believe theyâre coming from Google. Yeah,
00;47;54;29 - 00;48;18;06
Unknown
on the traditional kind of like RAG grounding, I'm pretty sure it's coming from Google. I heard that it was coming via some intermediary swiping services. They kind of sneakily engaged and that's why we have this whole thing over the weekend of Google disabling pagination parameters. You know, but anyway, I'm not quite sure about that. But I think I might be saying, I don't want to spread fake news here, but I think the agent experience is still coming from Bing. So
00;48;18;06 - 00;48;37;27
Unknown
if you give it a generics prompt, I donât know, find me a job, you toggle on the agent and then you say, I want to find a job in my locality or whatever, then I think I'm right in saying, I've kind of put them in two windows and watched them, the results that come back on that agent are Bing. So the virtual machine that it uses is using Bing, I think.
00;48;37;29 - 00;49;00;10
Unknown
I don't want spread fake news, but that's been my assessment based on some non-scientific testing. I mean, it would make sense that they wouldn't rewrite the whole way that they've approached it from the ground up, although I guess totally possible. So it would make sense that they're for different queries or prompts. I guess in this case, they they may be using different methodologies to try to get to an answer.
00;49;00;13 - 00;49;23;15
Unknown
Yeah. Well the, the, the biggest thing with agents is that they get blocked all the time. And if you'd like experimented with operator or the agent mode or whatever, it's like Cloudflare is just like Agent Poison, and or Google reCAPTCHA. And so I would assume when they're doing this, they want to go with like a friendly service that's allowed them on, rather than just going, OK, I'm going to search for this for you. I was blocked. I'm going to search for this for you. I was blocked.
00;49;23;17 - 00;49;38;26
Unknown
Oh, well, good luck. That's not a great experience for their users. So I would assume even if they for whatever reason prefer to use Google data for their RAG. Maybe their agents, which are very prone to getting blocked, maybe they want to go with something that. That's something they can rely on a bit more, you know, conspiracy theory, but maybe.
00;49;39;02 - 00;49;58;08
Unknown
Yeah, it makes sense. I mean, I guess since we're talking about LLMs, Nick Leroy was one of our first guests this season. He runs SEOjobs.com and I think he just launched ppcjobs.com. I just pinged him real quickly to see if he would have any questions of interest for you. And he was curious about the impact of non-click searches,
00;49;58;08 - 00;50;18;14
Unknown
right? Just generally traffic sort of going away. Are you guys thinking about, or I guess how are you thinking about supplementing that traffic? Are you paying attention to Reddit, for example, or some of these other platforms that are getting elevated in search and LLMs for that matter? So are you thinking about more of a distributed strategy there?
00;50;18;14 - 00;50;37;14
Unknown
Yeah, I mean the big chunk of traffic that everyone's feeding right now is really on informational searches and just for the nature, we don't make money by necessarily selling ad impressions, so it kind of hurts us less if you like. we do have a ton of content out there around because they had to resumes, had to cover letters, and also a bunch of salary information and stuff.
00;50;37;14 - 00;50;55;19
Unknown
So if you want to know what the salary is for a given job title, location, or company or something, you have all that information out there. Obviously, AIOs is doing the right job answering that in the SERP, and so youâd see a reduction in CTR. No surprises or controversy there. My position on that one, our position on that one, is that really that was never a commercial, directly commercial play.
00;50;55;20 - 00;51;13;11
Unknown
And people would would do those kind of searches. They would get the information they need, they would move away. And we were content to provide that information because we want to help people get jobs. But also, it's kind of a brand play, effectively. Hopefully, they come get a good experience. They fall in a warm fuzzy. And when they do want to come to a job, find a job, which is where we make money, then they would come to us rather than a competitor.
00;51;13;11 - 00;51;33;08
Unknown
So that was always the play. Our position on this is that if we are appearing in these AI surfaces, whether it's AIOs, AI Models, ChatGPT, whatever, then, yes, it's kind of a less rich experience for someone coming to the site and seeing the logo, maybe getting some peripheral messaging or whatever. But on the flip side, like AIOâs impressions are huge.
00;51;33;08 - 00;51;53;29
Unknown
Like a lot of people see them much more than one click to any individual website, even a big website. And also people trust them. People look at them and it's kind of like a recommendation from a trusted friend, almost. So yeah, on one hand, you're not getting that rich experience where they come to the site. But on the other hand, we're getting a brand benefit from something, the something I guess the people trust saying, hey, this is a really good website.
00;51;53;29 - 00;52;13;20
Unknown
This is where I got this information. This is something you should believe in. So it's kind of brand by a different means, really. Transactional search is obviously I mean, again, not giving away too many amazing secrets like they haven't been impacted in the same way by, AI overview CTR, CTR capitalization. At some point, I do kind of expect agents to start, to, ease into that a little bit.
00;52;13;20 - 00;52;31;28
Unknown
Transactional searches for e-commerce and job search and travel and every kind of transactional industry out there. But when that happens, I actually think, again, we'll see how it plays out. Who knows? But again, you look at how these agents work. And if you if you tell it to go to Indeed.com, you'll get digital content. It will do what you want you to do.
00;52;32;01 - 00;52;47;07
Unknown
If you try to go and just find find me a job in this case or in a find, vintage item for sale or whatever it is, then it'll go to search. You go to search and you literally you can see it by clicking on the results, just like a user would. So I honestly feel like it's it's just SEO with extra steps actually.
00;52;47;07 - 00;53;06;01
Unknown
So I personally don't think there's a massive existential crisis there. I think it's just going to be maybe a nuance on what we do right now. I heard you say SEOs are more important than ever. Yeah. No, really. Totally, totally. I genuinely do think that. Okay, so here's my little pitch. Most of might be cope because I have a huge mortgage, is it.
00;53;06;01 - 00;53;26;03
Unknown
So people, know, there's fundamental principles. There are more goods and services that people could possibly know about in their own head at an even point. People are gonna wanna eat at restaurants and get jobs and buy things and take insurance out, whatever. And there's lots of brands competing for them. Like these services, these AI-powered services, I believe, is fundamentally more useful than traditional search.
00;53;26;07 - 00;53;53;09
Unknown
And so more people are going to turn to them in future than maybe in the past would have just like relied on brand reputation, saw an offline ad, you know, a referral from a friend or whatever. I think more people are going to just go straight to their LLM of choice and get that recommendation. And therefore SEOs, as we do manipulate algorithmic responses, like we're to be more valuable than ever in terms of like pointing people to our employers, real world and someone else.
00;53;53;09 - 00;54;11;01
Unknown
So I think, I think effectively it's going to be it's going to be a really bumpy few years. And like traffic forecasts and stuff are going to have to come down quite a lot. But the actual value that we generate in terms of like getting business in I think is probably going to be more than ever important. I hope. As I say, big mortgage.
00;54;11;02 - 00;54;28;25
Unknown
I'm kind of hoping. And there is this train of thought or many people saying that the traffic that does come from LLMs, even though it may be a smaller chunk, tends to be much higher converting. So I think there's probably truth in that. Yeah. And also like this. That's absolutely true. It's my experience too.
00;54;28;25 - 00;54;51;07
Unknown
And also there's a lot of, we started just asking, itâs like very, very basic, but like as an employer comes in and post a job, we just say, Hey, what do you hear about this? And we seeing people, maybe not clicking directly, but again, recommendation from a friend and people are coming in and they're posting jobs. Yeah. All right. Getting to the rapid fire round, if you're ready. Give me.
00;54;51;09 - 00;55;09;25
Unknown
All right. You had a lot of good tips around Google Search Console sort of breaking down XML sitemaps, things like that. If there was any metric that doesn't exist today in Google Search Console, what would you recommend that they add? The one that caused most pain for me was the enhancements section of the site.
00;55;10;02 - 00;55;26;20
Unknown
I believe that as gospel and then I think it was from Google. think I actually reached out to them and said, what's going on with this? It turns out that at least for big sites, think, maybe all sites, those numbers, so like on the job details, job listings, reports or whatever, they're all just samples.
00;55;26;20 - 00;55;41;24
Unknown
They're samples that most directionally tells you what's going on. Like, that's not very helpful because I actually need a real number. if they could make that real numbers, that would help my life a lot, especially for sites of your size. I can only imagine it's significantly sampled. Yeah. Oh yeah. Yes, it is.
00;55;41;24 - 00;55;55;01
Unknown
So I was absolutely scratching my head. Like I had a conversion, like a pyramid. So these are all the jobs you put in, cool. And then all of a sudden, like the number of jobs visible in Google for jobs, is a crazy little bit in the pyramid. And then more jobs getting traffic. And was like, what's going on?
00;55;55;04 - 00;56;14;21
Unknown
And so there was that. And then also what I would love to see if folks, the guys named Daniel, Daniel Wassenberg or something, the PM for things, if you listen to this podcast, like taking the pack, the cool and indexation data into the Google Search Console bulk export, that would be massive. So I rely on that a lot as a sense check for internal data.
00;56;14;24 - 00;56;33;09
Unknown
But if we could take that and put that into our internal data lake as well. Massive game changer. So yeah, I love that. I would love for them to split out AI mode searches versus traditional search. I mean, that would be... It makes impressions in CTR. Something almost kind of worth this now, right? Like because you yeah you know it it yeah that to that to all right.
00;56;33;09 - 00;56;52;07
Unknown
What's your high score in maze speed scrabble? Haha, that's great! So I guess for context, youâre listeners. Pandemic locked in, my wife came up with this kind of game where we would just like come up with names in Scrabble really quickly and then we just did that for fun because we were bored out of our mind. And then, yeah, one day I was trying to learn React actually,
00;56;52;07 - 00;57;16;14
Unknown
I thought I should probably learn to do like coding React because it's important, itâs a big technology. So I sat down and I kind of codified this game in React as hand-coded it and then. Yeah, like literally, I donât know, a few months ago, I stumped up for the Claude premium subscription and I linked up this GitHub repo with, I put this game in with Claude code
00;57;16;16 - 00;57;31;06
Unknown
and I said, why is this any good? And he was like, I mean, the LLMs are unrelentingly positive, so it be really bad. He was like, no, no, it's actually not very good at all. Do want me to fix everything? And I was like, yeah, okay. So it literally went like this. I just turned on all accepted all changes and it completely rebuilt it from scratch
00;57;31;06 - 00;57;52;18
Unknown
like infinitely better. Look better, better, perform better, better you and everything. So I should never know what my high score is but yeah I was humbled, humbled by Claude there. Thatâs amazing. I mean, vibe coding is a little addicting, but also like incredibly impressive. Yeah. Yeah, yeah. How quick? Yeah, yeah. Anyway, I'm gonna hold the thing about.
00;57;52;18 - 00;58;13;23
Unknown
But yes, it's, especially for me as well, Game Changer has been like, just SQL queries. can like blitz out 200 lines SQL query now with a few English language sentences. So yeah, it's been a game changer. What's, this isn't a rapid fire round question, but I guess it could be. What's your favorite tool to to code with or until the code with?
00;58;13;25 - 00;58;43;04
Unknown
Well, it depends on what I'm doing. If I'm doing just, like, a personal hobby site or whatever, I would typically use VS Code, old-fashioned I guess, but VS Code. I was using the GitHub... I forgot what itâs called. I the GitHub premium tool. Itâs basically fancy what I suggest. I got rid of that and now I'm using Claude Code looking at my local repo. And then I'm using the terminal inside VS code to get to vibe code effectively and then just tinkering with it.
00;58;43;06 - 00;59;00;07
Unknown
The hard thing is, it's so easy just to accept all the changes it wants and then sometimes it gets itself in an absolute mess. So I think that the pro tip would be for any vibe code which is for me based on bit of experience, just check it into Git. Every time it does something, check it into Git so that if it gets too far down that line, you realize something is not terribly wrong.
00;59;00;09 - 00;59;17;00
Unknown
You can revert to the commit before it went horribly wrong. It's funny enough, a good friend of mine is Iâm coding an app now as everyone. he's like, it just went, you know, he basically vibe coded himself into complete chaos. And I was like, you gotta use Git, man. You gotta use source control, because otherwise, yeah, gonna get in a mess.
00;59;17;09 - 00;59;39;20
Unknown
So yeah, so Claude is my go-to, and then regular check-ins with source control to save my sanity. Source control is huge. that's a good tip. All right, two more. Any tips for writing a great dev ticket? so I'm kind of a PRD monster. Like, any question you ask me comes up with a three-page document with sub-bullets and everything like that. I tend to use...
00;59;39;20 - 00;59;57;09
Unknown
so the framework I used, one thing I remember from business school was SOS stack. So the situation, the objectives, the strategies and tactics, the actions that weâre actually going to do in the controller, how youâre going to measure it. And so I just put everything in that format. And that's just good context for engineers. Obviously, if I'm writing a small bug fix ticket, that's not necessary,
00;59;57;09 - 01;00;13;17
Unknown
But typically I'm writing like epics that will then get delivered by individual product managers. And so I tend to structure everything in that Solstack format, which is just a way of forcing me to think clearly about things. Now, if I put in that control, that measurement thing at the end, kind of...
01;00;13;17 - 01;00;27;11
Unknown
it really makes me think more closely about the objectives I'm setting for that project. it's all very well saying, oh, weâll make traffic go up. But then we actually go and try and write the query. And hey, we're going to measure with this query. You realize that the data you want is there and quite work out or whatever.
01;00;27;15 - 01;00;45;28
Unknown
yeah, it's a bit helpful. Nice. All right. Last one. Any books you would recommend could be self-help, could be fiction. Let me think, when was the last time I read a book? Yeah, I'm reading a couple right now, Iâve a couple of open books. the author is called Chip Nguyen. Oh, yeah.
01;00;45;28 - 01;01;14;05
Unknown
Designing Machine Learning Systems. Yeah, Designing Machine Listening Systems and AI Engineering, Building Applications with Foundation Models. I find it quite useful. There's a really good sort non-SEO podcast I listen to, which is called Machine Learning Street Talk, which is a, it's kind of like an AI researcher podcast. And a lot of people who are like on the frontier of the sort of AI research go onto it. And it's probably 50 % over my head, to be honest with you.
01;01;14;05 - 01;01;32;22
Unknown
It's quite like abstract and a little bit philosophical actually, but it's super helpful for understanding how these models work. And then those books I just mentioned by Chiep Hewton-Nguyen, she's really writing about from an engineer's perspective, how you build them. And then what do you think, I'm starting to think about like, well, how is Google actually choosing these results for its RAG
01;01;34;12 - 01;01;51;25
Unknown
like synthesizing its answer? Like looking at it from an engineering perspective, whether it's like from researchers or from like more practical like how it is actually built. It's actually very helpful, I find. So yeah, I would probably go with that. I'd probably go with that. Yeah. Early on, I took some classes on just general basics of machine learning
01;01;51;25 - 01;02;13;23
Unknown
and as LLMs have exploded, that sort of foundation was incredibly helpful just to sort of understand what's happening there. All right, before we let you go, we started a little bit of a prediction tradition. I'd love to get your thoughts on, you know, if you go to Google.com in 12 months, what you expect the job search experience to be if someone's looking for, I don't know, SEO director in Connecticut.
01;02;13;25 - 01;02;30;00
Unknown
How do you expect that to change in the next, I don't know, 12 or 48 months? Well, will people do that even? I don't know. Will they kind of assign an agent to go and look for jobs for them, whether that's like Google's agent or some standalone brand, like Indeedâs agent to do that. It's an open question. In terms of the traditional search experience,
01;02;30;00 - 01;02;46;22
Unknown
Honestly, my conspiracy theory slash doom protection is thereâs going to be a ton more ads, I think, in future because we're seeing a lot of... I've got some friends at Google and I kind of like talk to them about how it's going and stuff and they... Yeah, I mean... they need to keep their stock price going up into the right.
01;02;46;22 - 01;03;04;29
Unknown
and so they need to keep squeezing out more money from searches. And if searches are migrated to ChatGPT, or people are just getting their answers from AI mode, but not necessarily clicking ads as much, traditional search results are going to get a lot more ads, I think. yeah, I kind of went for the day when the first page of search is basically all ads, to be honest with you. Hopefully not,
01;03;04;29 - 01;03;23;16
Unknown
but we'll see how it goes. They got to replace that revenue somewhere. Right. Yeah. Yeah. All right. Well thanks so much Chris. This has been super insightful. Really appreciate you spending time with us before we let you go. Let let everyone know where they can find you online. I'm a very intermittent LinkedIner, so I'm sure you can find my LinkedIn somewhere, but very rarely on it
01;03;23;18 - 01;03;38;23
Unknown
And then I have a blog at q4query.com, which I also very rarely update, but yeah, itâs got whatever I do right up there. All right. Amazing. Chris, thanks again. And if you enjoyed the show, please remember to subscribe, rate and review. We'll see you next time. Bye bye.