GEO Readiness Study 2026: 2,563 Sites Ready for Google, Not for ChatGPT
We audited 2,563 sites: 63.5/100 on SEO, 45.1 on AI visibility, and on 97.8% of pages fewer than half the sections open with a citable passage.
· 9 min read
The average site in our sample scores 63.5 out of 100 on classic SEO and 45.1 out of 100 on visibility in AI engines. That's the sharpest finding from our August 2026 study, covering 2,563 websites and 6,455 pages audited between June and August 2026.
Put plainly: after twenty years of collective work on ranking in Google, most sites now walk up to ChatGPT or Perplexity carrying pages those engines don't know how to quote. Here are the numbers, the method behind them, and what they don't prove.
What does the study show, at a glance?
According to our August 2026 study of 2,563 sites and 6,455 pages, six measures are enough to capture the gap between classic search optimization and visibility in AI engines. All of them run on the same sample, and each one drops from its denominator the pages where the check doesn't apply.
| Measure | Result | Basis |
|---|---|---|
| Average SEO score | 63.5 / 100 | 2,563 sites |
| Average AI visibility score | 45.1 / 100 | 2,563 sites |
| Sites below 50 on AI visibility | 67.5% | 1,730 of 2,563 sites |
| Pages where fewer than half the sections open with a citable passage | 97.8% | 5,146 measured pages |
| Sites with no llms.txt file | 77.2% | 1,530 of 1,982 sites checked |
| Pages with no link to an authoritative source | 11.3% | 6,455 pages, no exclusions |
Only one of those six measures is good: links to authoritative sources, failing on just 11.3% of pages. These sites know how to cite others; they just don't know how to be cited themselves, which is the paradox this study measures.
How did we measure it, and what can't the sample tell you?
Every site in our August 2026 study went through 46 checks: 21 on classic SEO, 25 on visibility in AI engines. For each domain we kept only the most recent audit completed between June 1 and August 27, 2026. A site audited twenty times therefore doesn't carry twenty times the weight.
Honesty requires one clear caveat. These sites weren't drawn at random from the web: they're our customers plus companies reached through targeted outreach, overwhelmingly small and mid-sized French businesses. A panel of large US publishers would produce different percentages, likely better on structured data.
What this sample does describe well is the real state of the French business web in 2026, the one an agency or a marketing lead deals with every day.
One more note on method: most checks don't apply to every page. A page with no numbers in it can't be judged on whether its statistics are dated; a contact page has no business carrying article structured data. Those pages drop out of the calculation, and each time we give the number of pages actually measured. Exactly one check applies to all 6,455 pages with no exclusions at all: the one on links to external sources.
Why is there an 18-point gap between SEO and AI visibility?
Our August 2026 study measures an average of 63.5 out of 100 on classic SEO against 45.1 on AI visibility, an 18.4-point gap. And 67.5% of sites, 1,730 out of 2,563, fall below 50 on AI visibility. On the SEO side, the 63.5 average puts the bulk of the panel above that same line.
There's nothing mysterious about the gap. SEO rewards tasks an entire industry has learned to do, and that plugins now do unattended: fill in a title tag, compress an image, generate a sitemap, avoid broken links. A decent site clears those without anyone consciously trying.
Visibility in AI engines rewards something else: the shape of the text. An engine composing an answer doesn't rank pages, it extracts passages. It's looking for a block that stands up on its own, with its number, its date and its source inside it. In twenty years of search optimization, nobody ever asked writers for that. Hence the score.
On 97.8% of pages, fewer than half the sections open with a citable passage
A citable passage is a 40-to-60-word paragraph placed under a subheading, answering that subheading without requiring the reader to have read anything before it. Our August 2026 study asks that at least half a page's sections open with one. Against that threshold, 97.8% of the 5,146 pages measured fail.
The test itself is simple: no pointer back to an earlier paragraph, no implicit reliance on what came before. An answer, right there, in the first paragraph, with its number and its context inside it.
Against that threshold, 5,031 pages fail and 115 pass — 97.8% of the 5,146 pages measured. The remaining 1,309 pages drop out, as they can't be judged on this criterion.
Two other measures from the same study point the same way. Across the 5,099 pages where the check applies, 91.0% don't have two question-shaped subheadings followed by a self-contained answer. And across the 3,027 pages where the check applies, 89.2% carry no article structured data, the invisible markup that tells machines the title, author and publication date of a page.
They know how to cite sources; they don't know how to be one
On linking out to quality sources, only 11.3% of pages fail, according to our August 2026 study. It's the one check out of 46 applied to all 6,455 pages with no exclusions whatsoever, and nearly nine pages in ten pass it. These sites know how to lean on sources; they haven't learned how to become a source others reuse.
That skill matters all the more because engines draw heavily on brand-owned sites. Profound's analysis published in July 2026, covering 11.8 billion citations across eight engines, found that 57% of citations come from a brand site on average, ranging from 16% to 74% depending on the sector. Your own site is often the first source an AI has to hand.
The study published by Muck Rack in May 2026, covering 25 million links cited by ChatGPT, Claude and Gemini, shows for its part that journalism accounts for 27% of citations. And Ahrefs' December 2025 research across 75,000 brands established that brand mentions predict AI visibility better than traditional inbound links. We walked through all four studies in our analysis of brand citations versus press citations.
The difference is a reversal of perspective. Citing sources means writing for a reader who wants to verify. Being citable means writing for an engine that has to lift an answer in three lines without getting it wrong. The good news is that this skill isn't something you buy — it lives in the layout.
Undated numbers, the most widespread blind spot
Among the 895 pages that quote at least one statistic, 94.5% attach neither a date nor an identifiable source, according to our August 2026 study. That's 846 failures against 49 passes. The other 5,560 pages either quote no figures or fall outside this check.
A bare "73% of companies," with no year and no author, is close to unusable for an engine that has to attribute what it says. The same study shows freshness suffers from the same problem: across the 2,578 pages where a last-modified date is measurable, 86.7% haven't been touched in more than 90 days.
Engines favor information that's recent, dated, and placed high on the page. That's the mechanism behind the slow decay of citations on pages nobody touches anymore.
77% of sites have no llms.txt file
Across the 1,982 sites where the check completed in August 2026, 1,530 have no llms.txt file at all — 77.2%. This plain text file, placed at the root of a domain, describes a site's important pages in a few lines for the benefit of language models.
Say it plainly: no engine officially recognizes it today, and it will never compensate for poorly structured content. We file it under low-cost, zero-risk moves: an hour of work, once, then an update when the site structure changes.
That three sites in four don't have one says less about the file than about the general level of readiness. It's the simplest move on the entire list, and it's almost never made.
What should you take away?
Three findings stand out from our August 2026 study of 2,563 sites. The first concerns the gap between the two scores, the second the shape of opening paragraphs, the third the dating of figures. They are ordered below by return, from the most profitable fix to the least.
- Your SEO score tells you nothing about your AI visibility. The average gap measured in August 2026 is 18.4 points, and two thirds of sites fall below 50 out of 100 on the AI side. A site ranking well in Google can be completely absent from ChatGPT's answers. These are two projects, not one.
- The first paragraph of every section decides everything. 97.8% of measured pages fail on that exact point. Pulling the answer, the number and the context into the first 40 to 60 words of each section is the cheapest and highest-return fix on the list.
- Date and attribute your figures. 94.5% of pages quoting statistics don't. Adding the year and the author behind each number takes minutes per page and turns unverifiable content into a reusable source.
"These sites aren't bad. They're excellent at a game whose rules just changed. What's left is closer to reformatting than to a rebuild," says Nicolas Meridjen, founder of LightSpot and author of this study.
Where does your own site stand?
The averages in our August 2026 study sketch a landscape, not your site in particular. A well-built one-page site can pass all 46 checks we measured while a 500-page site fails every one of them. Only an audit of your own URLs settles the question.
LightSpot's free audit runs the same 46 checks used in this study against your URL, on 3 pages, in a few minutes, with no account to create. You get both scores, SEO and AI visibility, plus the top 3 fixes written out page by page.
For the underlying method, our complete GEO guide walks through each of the checks measured here, and our piece on proving your GEO actually works covers how to tie a fix to a citation you've earned.
FAQ
- What sample is this study based on?
- 2,563 websites and 6,455 pages, audited between June 1 and August 27, 2026. For each domain we keep only the most recent completed audit, so a site audited ten times doesn't count ten times. These sites come from two sources: our customers, and targeted outreach to French companies. That makes it a sample of small and mid-sized businesses, overwhelmingly French — not a representative sample of the global web. The direction of the findings holds at this scale; the exact percentages would differ on a panel of large US publishers.
- Why is the AI visibility score so much lower than the SEO score?
- Because the two scores reward different work. SEO rewards things the industry has spent fifteen years industrializing — title tags, page speed, sitemaps, internal links — and that plugins now handle on their own. AI visibility rewards the shape of the text itself: an answer that stands on its own out of context, a figure with a date and a source attached, a page that shows when it was last updated. Nobody has ever asked writers to do that, so almost nobody does. The 18.4-point gap between the two averages measures exactly that lag.
- What counts as a citable passage, in practice?
- A 40-to-60-word paragraph sitting right under a subheading, answering that subheading's question without requiring anything above it to be read first. It carries its own definition, its own number, its own context. An AI engine assembling an answer looks for exactly that kind of block, because it can lift it as-is and cite the source. Our test asks that at least half a page's sections open with one. Against that threshold, 97.8% of measurable pages fail: on those pages, fewer than one section in two starts with a citable passage. It's the single biggest gap, and also one of the fastest to close.
- Is an llms.txt file actually worth adding?
- No engine officially recognizes it today, and it will never make up for badly structured content. But it's a plain text file at the root of your domain describing your key pages in a few lines, it takes about an hour to write, and it carries no risk. According to our August 2026 study, 77.2% of the sites we checked don't have one. We treat it as a supporting signal — do it once, keep it current — not as a lever on its own.
- How do I find out where my own site stands on these measures?
- Run a LightSpot audit on your site's URL. It applies the same 46 checks used for this study: 21 on classic SEO and 25 on visibility in AI engines. You get a score out of 100 for each family, a letter grade from A to F, and the top 3 fixes to make first, page by page. The 3-page audit is free and needs no account.
Nicolas Meridjen
Fondateur de LightSpot.ai — outil d'audit de visibilité IA (46 critères SEO + GEO)
Je construis LightSpot.ai et j'analyse comment les moteurs de recherche IA (ChatGPT, Perplexity, Google AI Overviews) choisissent les sources qu'ils citent. J'écris sur le GEO et le SEO à partir de données d'audit réelles.
On the same topic
AI Citations: Brand or Press? What 4 Contradicting Studies Actually Reveal (2026)
Two studies clash on AI citations: 86% brand-owned vs 84% earned media. Here's why the numbers disagree, and what it means for your GEO strategy.
ChatGPT Ads in Europe: What Changes for SaaS Marketers
ChatGPT Ads is expanding across Europe. Here is what its targeting, measurement and separation from organic answers mean for SaaS marketers.
Why AI Citations Decay — And What to Do Before You Lose Them
A study of 3.5 million citations found half of them vanish within 4.5 weeks. What 'citation decay' means, and how to build a refresh routine that works.