An AI engine can say what your company does, rarely what sets it apart. The cause is less invisibility than readability: the proof of your difference lives in the press or in JavaScript that engines do not read. In Fast Growth Advisors audits, AI engine readability is the weakest sub-criterion, at 0.91 out of 5. Here are the three stages where your positioning vanishes, and the thirty-second test to check it.
A French Series-B deeptech raised 30 million euros. Its process cuts production costs by 45% and carbon footprint by 60%. Those figures are in the trade press and in the funding announcements, but none of them exists on its own site in a form an engine can reuse.
On its homepage, a visitor reads: “innovative advanced recycling solutions.”
Ask an AI assistant about it. It will describe what the company does without being able to say what sets it apart: it will name the category instead, and cite as a reference the competitor who put its proof in HTML.
This case comes from our audits, anonymized, and there is nothing exceptional about it.
Why does the scrape-to-referral debate miss the point?
Because it counts what AI systems take, when the useful question for a B2B startup is what they keep from your page. The scrape-to-referral ratio is the number of pages an AI system extracts for every visitor it sends back to the site.
Since August 2026, Microsoft Clarity has shown this ratio in its dashboards, with a public example of around 6,000 to 1. Cloudflare had published nearly 70,900 to 1 for Anthropic’s crawler over one week of June 2025, then around 4,580 to 1 a year later.
Three real figures, none comparable.
Yet the debate will crystallize around them: AI takes a lot and returns little, so what does it owe us? A fair question. But for a B2B startup leader, it misses the only thing that matters: when an engine reads your homepage, what does it keep?
At Fast Growth Advisors, we measured it. The deeptech pattern shows up in 61% of the 83 detailed audits in the Fast Growth Advisors (FGA) Observatory, and our measurements confirm it: the figures that prove the value live in the press and the funding announcements, almost never on the website.
How does AI engine readability get lost in three stages?
It gets lost at the fetch, the server response or text extraction, and information can survive the first two stages only to die at the third.
Many teams think in a single step: “the bot sees the page.” That is wrong.
At the fetch, the agent requests the URL, and can fail on a firewall, a missing page or a consent banner. At the response, the server returns bytes without executing a single line of JavaScript: the Vercel and MERJ (web analytics firm) study published in December 2024 found no major AI crawler that renders JavaScript.
Extraction is the stage where a tool turns the HTML into usable text. It throws away part of the document along the way.
Each stage has its own cause of invisibility and its own fix. An audit that does not separate them produces a false diagnosis.
Almost everyone forgets the third.
Yet that is where the tests run by Fast Growth Advisors produced the least intuitive results, because extraction follows none of the rules that twenty years of classic search optimization have drilled into marketing teams.
What do the extractors that prepare large-model corpora throw away?
They throw away content built by JavaScript and keep content hidden by CSS, the reverse of classic SEO reflexes.
To check it, Fast Growth Advisors built a test page with twenty unique markers, one per edge case, embedded in a body of realistic length, then ran it through the three extractors used in those pipelines: jusText, Resiliparse and Trafilatura.
Three results are worth pausing on.
A pure client-rendered page, meaning a React application that builds all its content in the browser, serves 128 bytes of HTML. Usable text after extraction: zero bytes. No partial degradation, it is all or nothing.
A block hidden with CSS but rendered by the server is read in full. The reason is mechanical: with no JavaScript execution, no CSS is applied, so the engine has no idea the block is hidden. It is the exact inverse of classic SEO, where Google renders the page and can devalue hidden content.
Twenty years of reflexes to unlearn.
And the test everyone recommends is wrong. A curl followed by a grep on the raw response finds your sentence even when it is injected by JavaScript, because it sits as a literal string inside the script, precisely the part every extractor discards. Result: the test clears the very sites it was meant to catch.
Why does the proof disappear first?
Because on a typical startup, JavaScript carries precisely the proof: logos in a carousel, reviews in a third-party widget, monthly-annual pricing toggles, key figures animated on scroll.
Your message itself gets through fairly well.
To show it, Fast Growth Advisors set up a fictional B2B startup homepage with common defects: result figures, client logos, testimonial and pricing grid injected by JavaScript. We then ran the same three extractors on it.
| Reading | Bytes of text | Share | Tracked elements kept |
|---|---|---|---|
| Seen by a human, JavaScript executed | 1,450 | 100% | 8 / 8 |
| trafilatura | 706 | 49% | 1 / 8 |
| jusText | 796 | 55% | 2 / 8 |
| Resiliparse | 949 | 65% | 2 / 8 |
Volume reassures, detail damns.
Trafilatura keeps 49% of the text but only one of the eight tracked elements. In all three extractors, the promise gets through, and so does the old positioning left in display:none during a redesign. None of the six real proofs (the three result figures, the client logos, the testimonial and the prices) survives anywhere.
The engine can then say what the company does. It cannot say what sets it apart.
Add the 61% pattern noted above, and the mechanism closes: the proof that existed was already somewhere other than the site, and the proof that remained there is carried by JavaScript the engines do not read. Invisible twice over, then. You will find the screenshot, the three extractions reproduced in full and the complete protocol in the white paper The recoverable message, freely available, with its PDF version.
What score does readability by AI engines get in our audits?
0.91 out of 5, under 20% of the potential: the weakest of the fifteen sub-criteria the FGA Observatory scores across 94 complete 15-criteria audits.
For context, none of the fifteen sub-criteria exceeds 60% of the potential, and the best-scored, differentiation and message clarity, top out at 52%. GEO readiness, meaning a site’s ability to be read and reused by generative engines, is therefore the weak point of an already weak set.
Should this worry you as early as 2026?
According to Demandbase, visits sent by ChatGPT to B2B sites rose from roughly 645,000 a month in June 2025 to 2.6 million in June 2026, a 303% increase. IDC (International Data Corporation) projects that 70% of B2B buyers in the United States will rely on generative AI to discover and evaluate their vendors by 2028.
Each will judge the urgency. The measurable point is already here: the channel is growing, and three quarters of the sites Fast Growth Advisors audits are not readable on it.
How can you test your site’s readability in thirty seconds?
Disable JavaScript in your browser, then reload your homepage.
What stays on screen is, as a first approximation, what an answer engine can read of you. If your message is there but your logos, your figures and your prices have vanished, you know what is left to do, and the fix is a targeted intervention, not a rebuild.
That leaves the extraction stage, which this test cannot see.
To go further, you have to measure what Fast Growth Advisors calls the recoverable message: the share of your positioning that survives a read without JavaScript, extractor included. Our audit protocol measures it, edge cases included, with three extractors and one check: do the promise, the differentiators and the proof come through in the extracted text?
The window is still open. It will not stay open: your competitors are publishing HTML.
This article explains why a page can be readable for a human and unreadable for an engine. We have since measured what three extractors actually keep on B2B vendor sites: in the worst case, 23% of the served text. ChatGPT visibility, what the engine sees of your site.
Sources
- Microsoft Clarity. Understand Which AI Crawlers Drive Real Visits with AI Scrape-to-Referral Insights, Microsoft Clarity blog, 13 August 2026. “AI Scrape-to-Referral Ratio” card, published example: 6,000 to 1. clarity.microsoft.com
- Cloudflare. The crawl before the fall… of referrals, Cloudflare blog, 1 July 2025: nearly 71,000 page requests by Anthropic for every page referral (70,900 to 1, week of 19 to 26 June 2025). Value of around 4,580 to 1 in June 2026 read on Cloudflare Radar, AI Insights page. blog.cloudflare.com
- Vercel and MERJ (web analytics firm). The rise of the AI crawler, 17 December 2024. Server-log study: none of the major AI crawlers render JavaScript; JavaScript files are fetched without being executed (11.50% of requests for ChatGPT, 23.84% for Claude). vercel.com
- FGA Message-Market Fit Observatory, V2 edition. Readability-by-AI-engines sub-criterion at 0.91 out of 5 across 94 complete 15-criteria audits; none of the fifteen sub-criteria above 60% of potential; “proof absent from the site” pattern in 61% of 83 detailed audits. Internal Fast Growth Advisors measurements. Observatory
- Demandbase. Press release of 12 August 2026: ChatGPT-referred visits to B2B sites rose from roughly 645,000 a month in June 2025 to 2.6 million in June 2026, a 303% increase. demandbase.com
- IDC. IDC blog, 10 March 2025: “By 2028, 70% of B2B buyers in the U.S. will rely on GenAI to discover, evaluate, and select vendors” (IDC, 2024). idc.com
- Fast Growth Advisors. In-house extraction measurement of 29 July 2026, reproducible (
cas_limites_realiste.html,test_extracteurs.py), across jusText, Resiliparse and Trafilatura, detailed in the white paper The recoverable message.
FAQ
What does readability by AI engines mean?
It is the share of your positioning an answer engine can actually read.
For Fast Growth Advisors, it is measured after the three stages that separate the URL from the text an engine uses: the fetch, the server response and the extraction. A site can render perfectly in a browser and still be unreadable to an engine that executes no JavaScript. Readability is judged on what the tool keeps, not on what the screen shows.
Why can ChatGPT or Claude ignore content that displays fine in my browser?
Because their crawlers read the raw HTML without executing JavaScript.
A Vercel and MERJ study published in December 2024 found no major AI crawler able to render JavaScript. Content built in the browser, such as a client-rendered React application, therefore arrives empty for those crawlers. Your browser does that rendering work, which explains the gap between what you see and what an engine keeps.
Can you judge your own site’s AI engine readability yourself?
Partly, with a browser and JavaScript switched off.
Reloading your homepage without JavaScript gives a rough, free idea of what an engine receives. One blind spot remains: extraction, which can still discard text that is present in the HTML. Measuring it fully means assessing the recoverable message: the share of positioning that survives a read without JavaScript, extractor included.
Do I need to rebuild my site to be readable by AI?
Rarely: in most cases, only the proof needs fixing.
On the sites Fast Growth Advisors audits, the message already gets through. What disappears are logos in a JavaScript carousel, figures animated on scroll, pricing toggles and reviews in third-party widgets. Bringing them back into server-rendered HTML is a targeted intervention. A rebuild is only justified when the entire site is client-rendered.