GPTBot does not follow JavaScript-injected links, so any page reachable only through client-side navigation stays out of ChatGPT's index. In episode five of The GEO Show, Paris Childress walks through a 41-day crawl test in which GPTBot reached zero of those pages, then discovered 250 of them within 48 hours of the links being converted to plain HTML. The episode also covers ChatGPT Ads view-through conversions, GEO moving in-house at Cisco and Replit, and the EU AI Act transparency rules that took effect on August 2nd, 2026.
Googlebot settled the JavaScript question years ago. The crawlers behind ChatGPT, Claude and Perplexity have not, and a 41-day field test puts numbers on the gap. Across roughly 1,000 pages and 30,180 logged bot requests, Googlebot and GoogleOther traversed JavaScript-injected links. GPTBot, ClaudeBot, Perplexity and the other tested crawlers reached zero pages accessible only through those links. Once the links were converted to plain HTML, GPTBot discovered 250 pages, the first within 48 hours.
That test frames episode five of The GEO Show. Paris Childress works through eleven stories from the week in AI search, and most push the same way: assumptions carried over from SEO do not survive contact with generative engines.
Because the LLM crawlers are not rendering the JavaScript that builds those links. In the test, the pages behind JavaScript-injected navigation existed and Googlebot reached them, and GPTBot logged zero visits. They were not deprioritised, or crawled and discarded. They were never found.
Paris treats this as a reset rather than a discovery. The SEO industry established years ago that Googlebot could crawl, render and understand JavaScript, so a JavaScript-heavy site was not at a particular disadvantage. That conclusion does not transfer to GEO.
Just know that if you're JavaScript only, you're gonna have a tough time with GEO.
His fix is deliberately unglamorous. If JavaScript is what the human experience needs, keep it, but ship a non-JavaScript HTML page as well, ideally with JSON structured data in it, so LLM bots can crawl and read it. He notes the resemblance to the old practice of double serving.
The hiring evidence says yes, and it is coming from large enterprises rather than startups. Cisco Careers has posted an SEO/GEO Content Strategist role whose remit includes establishing the company's Generative Engine Optimization and Search Engine Optimization content practice. Replit is recruiting a lead tasked with growing AI referrals, winning citations, running experiments and tying AI visibility to sign-ups and to revenue.
The acronym war is visible in the job titles themselves, with roles stacking abbreviations behind slashes because nobody has settled the naming. More consequential for agencies: these roles are being insourced, the way in-house SEO roles ended up competing with agency engagements. The same pressure shows up on the product side, where Surva.ai now documents a single platform spanning ChatGPT, Perplexity, Claude, Gemini and Google AI Overviews, with visibility tracking, competitor analysis, content generation and referral measurement. The integrated monitor, diagnose and draft workflow is commoditising too.
Only loosely, and the episode argues the proxy is weaker than the industry assumes. No keyword research tool will tell you how many times a specific prompt has been posed to ChatGPT. That data does not exist, so practitioners, Hop AI included, have used keyword search volume for the same topic as a stand-in.
Duane Forrester of Search Engine Journal argues that AI systems collapse many phrasings of the same need into relatively small consideration sets, which makes "is this category still contestable" more valuable than tracking dozens of semantic keyword variants. Paris notes the commercial interest Forrester discloses through Citation IQ, treats the argument as a hypothesis, then agrees with the mechanism: a keyword with 5,000 searches per month cannot be rewritten as a question and assigned 5,000 prompts per month.
The consequence is a different unit of work. GEO may resemble category ownership more than long-tail keyword capture: you map the topics and problem statements around a category instead of harvesting keyword variants.
Interactive ones, by a wide margin. Lawrence Hitchens of StudioHawk analysed 1.2 million AI referral sessions across more than 600 businesses. Interactive tools, templates and calculators earned 7.5 times their share of AI traffic relative to how many such pages existed overall. How-to content, definitions, comparison pages and guides also over-indexed. Standard blog posts under-indexed.
Google building its own tools into AI Mode and AI Overviews answers looked like a reason to stop making them. The referral data says otherwise, and vibe coding has made them cheap to build. Planning is changing too: Google's AI Mode data shows the average query is three times longer than a traditional search query, and planning-related queries have grown 80 percent faster than AI Mode queries overall. Briefs should model the whole sequence, with sections that independently answer the buyer's second, third and fourth questions:
No. Ivris Tech reports two pages that accumulated 542,615 Google impressions at an average position of 3.6 and produced seven clicks between March 19th and August 11th. The authors cannot prove those impressions came from AI systems, synthetic agents, rank trackers or Google's own testing, but none of it can be ruled out. The zero-click problem was always framed as falling click-through rate. This adds a second layer: the impression count itself is inflated by machine activity, so rising impressions may describe bots rather than buyers. Gemini behaves with similar noise. Steady Demand found that across 14,472 citations, repeated identical queries shared citation sources only about 40 percent of the time, and that Gemini and ChatGPT cited the same domains just 8 percent of the time.
The through-line is that GEO is not SEO with new acronyms attached. The crawlers behave differently, the demand data is missing, and the deliverables that used to sell are becoming free.
Hi, everybody. Welcome back to The GEO Show, episode five. We've got some great stories to run through today, so let's dive right into it. I'm your host, Paris Childress from GEOforge, and I do hope that at some point you will have a look at GEOforge for your AI visibility needs.
So story number one is about ChatGPT Ads. Very recently, ChatGPT Ads has added view-through conversion reporting with a one-day view-through conversion window, which means that when someone converts after seeing but not clicking an ad within one day, OpenAI is going to report that as a view-through conversion.
So people that have done advertising, paid ads, especially on platforms like Meta or other paid social platforms, you're probably familiar with view-through conversions, and I think this is very appropriate for OpenAI and ChatGPT Ads. So OpenAI is clearly now building the measuring stack that's needed for serious performance advertising.
That's probably gonna even muddy the waters further when it comes to attribution because you will certainly have a case where you have, let's say, one conversion that is claimed as a last-click conversion from your Google Ads paid search account, as a view-through conversion from, let's say, from Meta or maybe even from LinkedIn ads, and also as a view-through conversion from OpenAI's ChatGPT Ads.
So as always, we always recommend that you ask that lead when they come through the door in that first form that they fill out, “How did you hear about us?” And ask them to select among a few options, one of those being AI chatbot. And it's not perfect, but that will give you a directional sense for your ROI and your investment in GEO or AI search.
Let's move on to the next story. GEO is becoming a dedicated in-house growth function. We have now seen Cisco Careers has a job ad that's posted with the title SEO/GEO Content Strategist, whose remit includes establishing its Generative Engine Optimization, GEO, and Search Engine Optimization, SEO, content practice.
Also, we have seen and observed that Replit is recruiting an AEO/GEO/SEO lead tasked with growing AI referrals, winning citations, running experiments, and tying AI visibility to sign-ups and to revenue. So we have clear evidence now that large enterprise companies like Cisco are hiring for GEO roles with GEO in the job description.
And it is interesting now how many slashes there are because we're still trying to decide among the acronyms here. But Cisco's job ad has SEO/GEO Content Strategist, which you could argue is actually three roles in one, and then Replit's role is AEO/GEO/SEO. So they really can't decide here, but just casting a wide net.
So I think what we're gonna see is that more and more of these roles are gonna be insourced. So for those of you who are agencies that are listening, just be aware it's similar to in-house SEO roles that sometimes compete with agency engagements.
All right. The next story is: a field test gives one of the clearest warnings yet about JavaScript-heavy sites. A forty-one-day test across roughly a thousand pages logged thirty thousand one hundred and eighty bot requests. Googlebot and GoogleOther traversed JavaScript-injected links. GPTBot, ClaudeBot, Perplexity, and other tested crawlers reached zero pages accessible only through those links. After the links were converted to HTML, GPTBot discovered two hundred and fifty pages, the first within forty-eight hours.
So what is this telling us? I think that this is telling us that JavaScript is not a friendly language for crawlers, both for SEO and for GEO. Now, in the SEO world, it was more or less established years ago that Googlebot could crawl and render and understand JavaScript so that JavaScript websites and JavaScript elements on pages were not penalized in any way, or they were not at a particular disadvantage.
But now I think we are resetting this in the age of GEO, and the LLM crawlers are certainly struggling to understand and to ingest JavaScript. So the advice here is if you need to use JavaScript for the human experience, fine, but make sure that you have a non-JavaScript and HTML page, ideally even with some JSON structured data as well in that page so that LLM bots can crawl it and read it as well.
So it almost kind of harkens to the days of double serving or cloaking in a way. But just know that if you're JavaScript only, you're gonna have a tough time with GEO.
On to the next story. We have a new entrant that we're watching in the GEO software space, and they are collapsing monitoring and execution into one single product. So the company is called Surva, S-U-R-V-A.ai. And in their newly updated documentation, they describe a single platform that spans ChatGPT, Perplexity, Claude, Gemini, and Google AI Overviews with visibility tracking, competitor analysis, answer gaps, organic search data, content generation, and referral measurement.
So this is a full workflow of monitor, diagnose, and draft content. So this workflow, I know in a previous episode, we talked about how AI search visibility tools that are reporting only, read-only, are quickly becoming commoditized with even new entrants that are offering it entirely for free.
But now we're also moving towards the commoditization of the fully integrated workflow, the more vertical workflow, where it's not only about monitoring your share of voice and your AI visibility and diagnosing it, but it's also about doing something about it, taking action with targeted content production.
Next story. Fresh analysis reinforces that AI search should be organized around tasks, not keywords. Google's own AI Mode data shows that the average query is three times longer than a traditional search query. More than one in six US searches use voice or images, and planning-related AI Mode queries have grown eighty percent faster than AI Mode queries overall. Brainstorming queries have grown thirty percent faster.
So a new analysis published today from Google and posted on Google's blog argues that content architecture needs to reflect those multi-step behaviors. So this is interesting now because we're no longer trying to get a page ranked for a single query, but we're anticipating a longer journey with an initial prompt followed by follow-up prompts or follow-up questions in that conversation.
And what we wanna do is try to build out assets for that entire journey as opposed to a single page that can serve as a great landing page for a particular search query or a keyword. So content briefs should really model these sequences that go through this whole pattern or this process of exploring, comparing, validating, and then implementation.
And they ought to have, if it's going to be a full content brief, that should contain sections that are capable of independently answering the buyer's second and third and fourth questions and so on. So that is a major shift in the way that we are planning content production. And for those who are still doing content briefs, I think that's something very important to keep in mind, the full journey, including follow-up questions and follow-up prompts.
On to the next story. We have some strategic analysis from Search Engine Journal that says the real GEO opportunity set may be smaller than the keyword universe. Duane Forrester from Search Engine Journal argues that AI systems collapse many phrasings of the same need into relatively small consideration sets, making “is this category still contestable” potentially more valuable than tracking dozens of semantic keyword variants. He explicitly discloses a commercial interest through Citation IQ, so this should be treated as a hypothesis rather than settled evidence.
That is interesting because we have also taken a stab at trying to estimate prompt volume around a particular topic or a category. And bottom line is there are no tools that do this. There are no equivalent keyword research tools that can tell you that this particular prompt or this question has been posed to ChatGPT X number of times. That doesn't exist.
So what we need to do, and what we have been doing in our consulting work, is we are using keyword volume for those same topics as a proxy, keyword search volume as a proxy for prompt volume.
But this research is kind of turning that assumption and that mode of operation on its head, and I think rightly so, because a lot of keyword queries, it does make sense that they're collapsing into longer phrases so that if you have a keyword that has, I don't know, five thousand searches per month, you can't just convert that into a question and say that this is now a prompt that has five thousand queries or prompts per month.
So GEO strategy really may ultimately resemble more category ownership than long-tail keyword capture. And that is a very, very different mindset. Category ownership, meaning trying to figure out all the topics and problem statements around a particular category vertical rather than doing that long-tail keyword research.
All right, moving on to the next topic. At the same time, interactive tools are massively over-indexing in actual AI referral traffic. StudioHawk's Lawrence Hitchens analyzed one point two million AI referral sessions across six hundred plus businesses and found interactive tools, templates, and calculators earned seven point five times their share of AI traffic relative to how many such pages existed overall. How-to content, definitions, comparison pages, and guides also over-indexed. Standard blog posts, not surprisingly, under-indexed.
So the tools are not dead. I know that in a previous episode, we talked about that AI Mode and AI Overviews from Google are now building their own interactive tools right into the answer. But still, these tools are over-indexing on their share of AI referral traffic, and so they are still a very good idea to invest in building. And now building these interactive tools is easier than it's ever been with the dawn of vibe coding and tools like Lovable and Replit and of course Claude.
All right, moving on to the next story. Gemini citation behavior looks radically less stable than traditional rankings. This is not a big surprise. Steady Demand analyzed fourteen thousand four hundred and seventy-two citations from fourteen hundred and eighty-seven local queries across fifty US metro areas.
Nearly sixty percent of Gemini citations pointed to businesses' own websites, but repeated identical queries shared citation sources only about forty percent of the time. Google returned the same top business only seven percent of the time, versus ninety percent for Google's local pack. Gemini and ChatGPT cited the same domains only eight percent of the time.
So this is about local search, and this is how local search translates to AI search visibility. And what we're seeing here is that there is very little overlap between getting ranked in the classic Google local pack, which is the maps results, and having your business cited in Google AI Overviews.
So just because you are ranked in the local pack, and maybe you've been there for a long time and you have a solid position there, that is no guarantee that you will be mentioned and recommended accordingly in AI Overviews, either as mentions or as citations. So as always, we should think about AI ranking more as probability distribution rather than something that is deterministic.
Next story. First-party Search Console data can look spectacular while producing essentially zero human traffic. I don't know if some of you have seen this. We certainly have. Ivris Tech reports two pages that have accumulated five hundred and forty-two thousand six hundred and fifteen Google impressions at an average position of three point six, but only seven clicks between March 19th and August 11th.
Queries of seven words or more generated four hundred and fourteen thousand three hundred and fifty impressions, but only eight clicks. Crucially, these authors explicitly say that Search Console cannot prove that these impressions have come from AI systems or synthetic agents or rank trackers or Google testing or any other machine activity, but those cannot be ruled out.
So I think what we're seeing here in this zero-click world is not only that the click-through rate from your search impressions is going down dramatically, but that the volume of impressions is now getting really muddled and inflated potentially by bot traffic and agent traffic. So for those who have reported on rising search impressions as a primary KPI for SEO, I don't think that's a trustworthy metric anymore just because so much of those impressions are coming from bots and not from humans.
All right. Story number 10. A new GEO product launch shows technical AI readiness audits are commoditizing quickly. A Romanian startup called Human AI Labs launched Blindspot, which evaluates websites against 70 deterministic criteria spanning AI crawler access, structured data, semantic clarity, citable content, authority, and trust signals. The company said its beta audits averaged 48 out of 100. Those scores and criteria have not been independently shown to cause higher AI visibility.
So the audit business is similar with SEO. I think now SEO audits have really moved to AI, and it's very difficult really to package an SEO audit as a service. I believe now that what this is telling us is that the same thing is true for a GEO audit.
And many SEO agencies that we have seen who are making the transition from SEO services into GEO services are taking that same audit concept from SEO into GEO, and I believe that it's a mistake actually because I think that a GEO audit can be done very, very thoroughly, almost instantly with AI. And we even have a tool that does a pretty good job of this.
So I just think it's gonna be very hard to charge for that and have that as kind of an initial project that precedes a long-term retainer with GEO clients. I just think that this is now part of the proposal stage where you give the prospect a full GEO audit and then try to start off right away with a long-term execution plan and a retainer.
All right. Finally, the last story of the day today is that EU AI content transparency rules are now operational and no longer theoretical. It's confirmed that Article 50 of the EU AI Act has applied since August 2nd of twenty twenty-six. Providers of generative systems face machine-readable marking requirements.
Deployers must disclose certain deepfakes and AI-generated text on matters of public interest when it lacks human review or editorial control. Penalties can reach fifteen million euro or three percent of worldwide annual turnover, depending on the violation.
So three percent of worldwide annual turnover for a company like OpenAI or Anthropic is massive, massive revenues. I think that the latest number that I heard was that Anthropic's revenue, their ARR or their annual recurring revenue run rate, is now approaching about seventy billion, and they'll probably end the year at over a hundred billion dollar run rate.
So three percent of that is a huge number, and that is why you're seeing that Anthropic has now come out and taken this issue seriously, and they have announced their watermarks. And I think that it has shaken up the conversation around GEO quite a bit.
But I think that as long as content that is generated through AI or with the assistance of AI focuses on information gain and proprietary knowledge coming from the brand source, then I think that there will not be a penalty or a suppression whatsoever, and in fact, that content will still thrive, regardless of the degree to which it has been assisted by AI in its production.
All right. That'll do it for today's episode. Thank you all for listening, and we'll see you on the next one. Bye.
