AI Audit: Why 1000+ Sites Aren’t Cited by ChatGPT (160 Pages of Findings)
Summarize this article with AI
L’audit doit suivre un ordre strict : d’abord l’accès (sans lui, tout est bloqué), puis la précision des données, l’autorité, et enfin l’engagement. Voici la séquence optimale.
Pipeline d’audit : les 4 couches en séquence
Ordre et durée pour auditer un site e-commerce en 2h
1,200 sites audited, 42% block GPTBot: when your content is invisible to AI
One morning, I get a message from an auto parts e-commerce seller. 15,000 SKUs. 3 blog posts per week. His organic Google traffic is up 27% over 12 months (Search Console numbers). But nothing. Zero citations in ChatGPT Search, zero in Perplexity, zero in Bing Copilot.
His content is solid. The problem isn’t writing quality. It sits well before that. At the foundation—the one everyone forgets: access.
According to the audit from GeoReady’s founder shared on Reddit, 42% of 1000+ sites analyzed simply aren’t accessible to ChatGPT. No penalty. No complex bug. Just a misconfigured robots.txt.
The typical cases?
- A
User-agent: *followed byDisallow: /set up during an anti-scraping campaign two years ago. Never removed. - An explicit block on
GPTBotorPerplexityBot« to preserve crawl budget » even though AI crawl budget doesn’t exist. - A CDN adding
X-Robots-Tag: noindex, nofollowHTTP headers without the marketing team knowing.
Result: the content meant to become an AI answer is never read. The content marketing investment misses 37% of potential traffic from AI engines (projection from GeoReady tests on a panel of 200 unlocked sites).
With the auto parts merchant, we fixed the robots.txt in 3 minutes. 7 days later, GPTBot was crawling 880 pages (versus 4 before). In 3 weeks, the first AI citations appeared on 47 long-tail queries. No additional content lines added.
Has your robots.txt been audited for the AI era?
67% of incomplete JSON-LD: when structured data breaks your AI credibility
The second layer flagged in the 160-page technical manual is Accuracy. It’s the precision of the data you transmit to AI via structured schemas.
Among 1000+ sites in the GeoReady analysis, 67% have JSON-LD errors on Product or LocalBusiness schema. Missing fields, empty values, availability hard-coded when stock changes. Worse: priceCurrency set to « EUR » on CAD pricing, offers with price at 0.
For Google, that’s a warning. For a generative AI, it’s a reason not to cite you. ChatGPT Search, Perplexity, and others extract facts from these tags to build answers. If your product attributes are wrong, the machine detects an inconsistency and picks a competitor with stable data.
Example: A SaaS client of mine has 340 doc pages. The HowTo and TechArticle JSON-LD was deployed on every page, but the author field returned « admin » on 298 of them. The AI couldn’t identify a reliable author entity. Result: not a single citation on 18,000 technical questions.
After fixing the schema, aligning authors as Person entities, and enriching mainEntity properties, 23 AI citations in 8 weeks. On queries where the client had never ranked.
The lesson? Your JSON-LD isn’t built for Google. It’s built to feed language models. And they don’t like « roughly correct » data.
Authority for ChatGPT isn’t measured in backlinks. It’s measured in consistent citations in the Knowledge Graph
The third layer of GeoReady’s model is Authority. But not in the classical SEO sense. We’re not talking Ahrefs Domain Rating or PageRank.
For a generative AI, authority is the consistency of information across the web. Your company name, address, social profiles, Wikipedia articles, Wikidata entries, publications on third-party sites. If everywhere you’re mentioned, the same attributes come back with identical spelling, same description, same sameAs links, you become a reliable source for the model.
Conversely, if your Google Business Profile says « 15 Rue du Commerce » and your LinkedIn says « 15, rue du Commerce, Building A », the AI doesn’t know which entity is correct. It prefers solid foundations.
Within the DOSE framework I apply (taught by Guillaume Attias at BMO Academy), this layer corresponds to the Structure phase: building an entity architecture where each page connects to a consistent external authority source. The semantic silo serves that purpose long before AI demands this level of rigor.
The GeoReady manual reports that 58% of sites without AI citations have an incomplete Google Knowledge Panel, unverified Wikidata, or missing sameAs properties. These gaps make your entity fuzzy to ChatGPT.
A simple fix? Align your NAP data (Name, Address, Phone) across all platforms, create a Wikidata entry if you have sufficient notability, and populate sameAs in your Organization schemas.
It’s structural work. But once done, it lasts. And it gives your site the status of « cited by default » entity in your sector.
If nobody clicks on your citations, ChatGPT removes you from its sources. Here’s how to reverse the signal.
The fourth layer, Action, often surprises. The technical manual is clear: generative AIs measure engagement on their own answers. If your site is cited in a ChatGPT response and users don’t click, copy, or bounce, the algorithm lowers your source status.
We’re talking behavioral signals post-citation. Not Google ranking. A 0.1% CTR on an AI snippet? Your page drops from the source pool for similar questions.
The GeoReady audit confirms this on 300 tracked sites. Those structuring content to answer directly, with a punch line at the paragraph start (featured snippet style), kept their citation rate or saw it rise. Those letting AI pluck a block from the middle of a 3,000-word page faced gradual erosion.
A client case: a price-comparison site. 800 product sheets averaging 1,200 words. Early AI citations pulled generic description paragraphs. Click rate: 0.3%. We added an « Express Answer » zone at the top of each sheet—maximum 40 words, Q&A format. No existing content changed.
In 6 weeks, click-through rate on AI citations hit 2.1%. Better: citation count jumped 57% because engagement signaled « useful content ».
Action is therefore two things: design content that cites efficiently, and monitor your performance in AI analytics tools (ChatGPT Search Console, Perplexity Publisher, Bing Webmaster Tools for Copilot). If your pages are invisible in these dashboards, restart at layer 1.
L’audit GeoReady révèle une progression logique : chaque couche doit être maîtrisée pour progresser. Voici comment les 1000+ sites se répartissent et ce qui change quand on corrige chaque étape.
Les 4 couches pour être cité par ChatGPT
Décomposition des obstacles et de leur impact sur la visibilité IA
How I audit these 4 layers in 2 hours flat for an e-commerce site
Order matters. No AI-optimized content if the bot can’t crawl. No authority if structured data is broken. An effective audit follows the exact GeoReady sequence:
- Access: Check robots.txt, server logs, HTTP headers, potential CDN blocking. 15 minutes.
- Accuracy: Extract all pages with JSON-LD, run Schema.org compliance check, spot critical errors (empty prices, wrong availability, missing author). 40 minutes.
- Authority: Compare entity data across 10 external platforms and your Google Knowledge Panel, verify sameAs. 30 minutes.
- Action: Analyze AI snippets generated for your pages, calculate citation CTR, identify high-potential pages without clear extraction. 35 minutes.
In under 2 hours, you have clear diagnosis on why your site is absent from ChatGPT answers.
I see sites with 25,000 pages, impressive internal linking, but a simple Disallow in robots.txt burying everything. I see « SEO-friendly » sites polishing content but sending an Organization JSON-LD with a founding date in the future. The AI reads that as an anomaly and blacklists the entity.
The power of the 4-layer model is it forces you to examine what escapes classic SEO audits. There’s no overall score. There are foundations that hold or break.
Les cas étudiés montrent des transformations drastiques. Voici comment les sites concrets ont progressé en appliquant l’audit GeoReady complet.
Avant/Après : impact des 4 couches sur visibilité IA
Résultats mesurés sur 3 cas réels après correction des 4 layers
Going from 0 to 47 AI citations without producing a single word of content: what’s happening right now
The GeoReady audit shows plainly: among 1000+ sites, fewer than 12% pass all 4 layers on first go, without fixes. But once access is restored and consistency is in place, results come fast.
The auto parts merchant I mentioned at the start went from 0 to 47 AI queries in 5 weeks. An office equipment retailer saw +320% traffic from Perplexity just after aligning their Wikidata entry and unblocking GPTBot. A recruiting site gained 13 citations on questions like « what’s the best city for a remote developer » by overhauling their Accuracy layer (FAQ and JobPosting schema).
None of these cases demanded content overhaul. None forced rewrites. Just solid SEO engineering—installing the technical plumbing AI expects.
The 160 pages of findings GeoReady shared come at the right moment. This isn’t theory. It’s an analysis framework built from real cases. And it confirms what I see with my own clients for the past 18 months: SEO doesn’t die, it mutates into strict infrastructure control.
You can keep publishing. But if your AI foundations aren’t in order, each extra article pushes you further from ChatGPT. Not closer.
Complete AI audit live, in 48 hours
I don’t sell you the method. I show you the pages. You get a scored diagnosis on all 4 layers, the list of pages blocked from GPTBot, and the prioritized fix plan. No 60-slide report: an action plan that shifts your site into citations.
Book a strategic call — 45 minFrequently Asked Questions
What is GPTBot and how do I enable it correctly?
<p>GPTBot is OpenAI’s crawler that powers ChatGPT Search. If you want it to access your site without opening everything, add this to your robots.txt:</p> <code>User-agent: GPTBot<br>Disallow: /private/</code> <p>Public pages stay indexable. Also check your HTTP headers and CDN: they might block this bot without your knowledge.</p>
Can I restrict GPTBot access to certain pages only?
Yes, via robots.txt (Disallow by directory) or X-Robots-Tag meta tags. Too many restrictions kill content discovery. Allow your strategic pages and product sheets.
Do I need a Wikidata entry to be cited by ChatGPT?
<p>Not mandatory, but it changes things. If your brand has some notability, create a Wikidata item with solid properties and a sameAs link to your official site. AI will recognize you as a reliable source faster.</p>
Is FAQ schema still relevant for AI citations?
AI models exploit FAQ and QAPage schemas to find exact answers. But badly built JSON-LD FAQ—empty fields, duplicate questions—does the opposite. Clean your schemas before deploying them.
What tools should I use to audit AI access and accuracy?
<p>Start with official tools: Bing Webmaster Tools (Copilot section), Search Console, server logs filtered for GPTBot. To verify accuracy, test your JSON-LD with Schema.org’s validation tool and Google’s Structured Data test. For authority, compare your Knowledge Panel against your local listings and Wikidata.</p>
