The crawlers that write AI answers
never run your JavaScript
Your browser fills the empty container in. GPTBot does not. That single difference decides whether an assistant can quote you, and no report in a normal marketing stack will ever show it to you. We publish the measurements, the method and the claims we refuse to repeat.
No signup, no email wall. The result renders on the page.
Four numbers, each with
the date it was true
These are the load bearing facts under everything else on this site. The honest caveat belongs on the first one: the field keeps re-citing a single study from December 2024 rather than re-running it, so treat non rendering as a reliable operating assumption rather than as physics. When somebody re-measures it, this band changes and the old row stays visible.
Major AI crawlers that executed JavaScript
Vercel and MERJ server log study, 17 December 2024, cross validated on unrelated stacks. The agents fetched script files and ran none of them.
llms.txt adoption across roughly 300,000 domains
SE Ranking, May 2026. Adoption among the top 1,000 domains by traffic was zero, and no citation lift survived controls for authority, schema density and recency.
What Google states Search does with llms.txt
From Google documentation, not from a vendor. Ship the file anyway if you want, it costs a quarter of an hour and carries no risk. Just never report it as a result.
FAQ rich results in Google search
They stopped appearing on 7 May 2026 and the reporting was withdrawn the following month. Keep the question and answer section, drop the expectation attached to the markup.
Two gates you control,
and one you have to earn
Everything published here sits under one model, run in strict order. The crawler has to receive your words. The page has to contain something worth lifting on its own. Then something you do not own has to say the same thing about you. Failing a lower gate makes every higher gate irrelevant, which is why content led GEO engagements so often spend a quarter of a year on drafts that no assistant ever fetched.
Crawler access
Which agent strings to test, and why a site can fail at the edge rather than in the front end. Cloudflare has blocked AI crawlers by default on new domains since 1 July 2025, and nobody on the marketing team is told.
Extractability
Whether the initial HTTP response contains your body copy or an empty root element, and what a section has to carry to survive being quoted alone: a definition, a number with its denominator, a comparison, a procedure.
Schema and entity strategy
Google states plainly that structured data is not required for its generative features. Microsoft has said on the record that schema helps its models understand content. Both are true, and the reason to ship it is entity disambiguation rather than a ranking effect.
Measurement
The two free first party reports almost nobody has switched on, why referral traffic undercounts when native app traffic sends no referrer, and how to run a fixed query panel against the engines that report nothing at all.
Fix the fetch first,
then earn the quote
The order is the method, and it is the part most of this category gets backwards. A rebuild that starts with content is starting at gate two on a site that is failing gate one, and the whole content budget is spent before anyone checks whether the pages were ever readable.
Crawl audit
Every route type fetched with each documented agent string, two checks per route. Does it return 200, and does the returned HTML contain your actual copy rather than an empty container. The second check is the one that catches the real problem and the one nobody runs.
Extractability rebuild
Prerender, server side render or static generate, whichever suits the stack. Then restructure the content itself: answer first openings, a definition in the sentence after every heading, numbers carrying their denominators, and sections that make sense lifted out of the page.
Entity and schema layer
Organization and WebSite sitewide, one Person identifier reused wherever the author appears, article and breadcrumb types on posts. No how to markup, which died in 2023, and no rating markup for a testimonial nobody signed.
Measurement and corroboration
Switch on the Bing and Search Console AI reports in week one, because a baseline is worth more than a month of tactics. For the assistants with no first party reporting, a fixed panel of queries run the same way every month. Then go and earn a mention on something you do not own.
The site is the argument,
not the brochure
Almost every claim in this category is unfalsifiable by design. Ours are checkable with curl, on this domain, today. That only works because this is a standalone property rather than a folder on an agency site, which is the whole reason it was built this way.
Audit us before you believe us
Real HTML in the first response on every route, an explicit crawler allowlist, one consistent entity graph, and our own extractability audit published with the failing rows left in. We would rather be audited than believed.
We name the invented numbers without printing them
The citation multipliers circulating for comparison tables, question and answer markup and content recency come from a single vendor with no published method, no denominator and no replication. We say which claims those are. We never reproduce the digits, because a debunked figure lifted out of context is still that figure inside an answer.
Part of the Outbound Pros group
InboundPros is run by Jānis Plūme, who also runs the B2B outbound agency in the group. That is stated on every page here, because a site that recommends its own group without saying so is misrepresenting its independence, and because stating the structure is what lets a model resolve the properties to one organisation.
The questions that arrive
before the first call
Do ChatGPT and Perplexity crawlers execute JavaScript?
No, on current evidence. The Vercel and MERJ server log study published on 17 December 2024 found no JavaScript execution by any major AI crawler. The agents fetched script files and ran none of them. Nothing published since contradicts it. The honest caveat is that one study is now carrying an entire field, and most 2026 sources re-cite it rather than re-run it, so hold it as an operating assumption you would happily see overturned.
Is my React site invisible to AI assistants?
To the non rendering crawlers, yes, unless you prerender or server side render. Googlebot and Bingbot both render, so you can sit perfectly indexed in Bing and still contribute nothing to a ChatGPT answer, because the search bot that fetches the URL to extract a quotable sentence gets the empty container. Being indexed and being quotable are separate problems and passing the first tells you nothing about the second.
Does llms.txt actually do anything?
Not for search visibility. Google states that Search does not use it. The SE Ranking study across roughly 300,000 domains found 10.13% adoption, zero adoption among the top 1,000 domains by traffic, and no citation effect once site authority, schema density and content recency were controlled for. Ship it if you like, since it costs fifteen minutes and carries no risk, and some developer tooling does read it. Just do not let anyone put it in a report as a win.
Is GEO just SEO with a new name?
Substantially, with two real differences. Both reward being indexed, being specific and being worth reading, and Google says directly that its AI features carry no additional technical requirements. The first genuine difference is rendering: the AI crawlers do not execute JavaScript and Googlebot does, so a site can pass classic SEO and fail AI retrieval entirely. The second is that retrieval scores passages rather than pages, so a self contained section answering one narrow question can enter an answer the whole page would never have ranked for.
Which schema types are worth shipping?
Organization and WebSite sitewide, Person for the author with one identifier reused across every property, article and breadcrumb types on posts, and the software application type on tools. Google requires none of it for generative features and says so in its own documentation. Microsoft says schema helps its models understand content, which is the strongest available justification and it does not come from Google. The how to type is dead, FAQ markup produces no Google rich result since May 2026, and fabricated rating markup is a manual action waiting to happen.
Who should not work with us?
Anyone who wants a single AI visibility score that climbs every month, because the engines cite almost entirely different sources and averaging them produces a number that cannot move for a reason you can act on. Anyone wanting a guaranteed citation count, which nobody in this field can honestly offer. Ecommerce, local and consumer search, where we would be learning on your budget. General SEO, which we do not do. And anyone unwilling to publish anything original, because the strongest lever here is a number that exists nowhere else, and without it the ceiling on the engagement is low.
Fetch your own homepage
the way a crawler does
The checker prints the exact command that requests your page without JavaScript execution, then scores what comes back across crawler access, extractable HTML, answer formatting, schema coverage and entity consistency. Most people who book a call have run it first, and the call is better for it.
Ungated. Nothing stored, nothing emailed.