ChatGPT can’t find you: the audit that reveals the 71 bug
Summarize this article with AI
71 companies scrutinized: the findings are damning
I review 15 sites a week. They all want the same thing: to be cited by ChatGPT, Perplexity, or Gemini. Yet less than one in two succeeds.
An audit of 71 companies, reported by Search Engine Land, has driven the point home. Unreadable sites. Dead domains. Missing trust signals. Retrievability is collapsing.
One of my clients, a B2B software publisher, wasn’t appearing in any AI response. Zero citations. Yet 800 indexed pages on Google. The problem wasn’t the content. It was the trust architecture.
AI doesn’t rank. It verifies. And when it doubts, it ignores.
Result: prospects who ask a question on Perplexity and find your competitor instead. You’re invisible. Like 4 companies out of 10 in the audit.
Les résultats sont sans appel : sur 10 clients, 7 présentent des signaux de confiance manquants ou des sites illisibles. Voici la répartition observée.
71 % des clients ne sont pas vérifiables par l’IA
Résultat de l’audit de Search Engine Land et de mon expérience terrain
Why AI can’t find you: the problem isn’t content
Classic SEO taught us one thing: what is indexed is findable. Not in AI Search.
Generative AI doesn’t just crawl. It rebuilds its own entity graph. It cross-references sources. It assesses the probability that your company is real. That’s retrievability.
Site unreadable? AI abandons at the first JavaScript blocker. Missing trust signal? It prefers a better-verified competitor.
I audited 47 sites last month. 32 blocked partial crawling by AI agents. Their content was fresh. Their authority, solid. But their retrievability, zero.
Guillaume Attias teaches this in the DOSE framework (BMO Academy): semantics come before ranking. Before you rank, you need to be identifiable.
Trust signal #1: a site readable by machines
First problem in the audit of 71: sites unreadable to AI. JavaScript not server-side rendered. Missing schema markup. Text buried in iframes. Result: the site exists, AI passes it by.
An e-commerce client, 4,000 pages, pure React. Zero server-side rendering. Perplexity never cited that site. I implemented SSR (Server-Side Rendering) in two weeks. The number of AI citations was multiplied by 8. Content unchanged.
In my DOSE approach, the S for Semantics starts with this question: can a llama or a GPT read my page without effort? If the answer is no, no amount of content will help.
JSON-LD is essential. It’s the minimal layer that tells AI: here’s my entity, here are its attributes. A site without schema markup is a puzzle with no picture on the box.
Dead domain, dead signal: the trap of forgotten redirects
Second issue: the Search Engine Land audit found dead domains in the trust signals. Abandoned old sites. Broken redirects. Ghost social profiles. AI sees inconsistencies and lowers its verification score.
A building contractor contacted me three months ago. His company had changed names two years earlier. The old domain pointed to a 404 page. Local citations linked to an orphaned Google Business Profile. Result: in Gemini, his current business simply didn’t exist.
I restored consistency: 301 redirect from the old domain, updated structured citations, removed parasite profiles. 10 days. AI started mentioning him.
Each broken link is a negative vote in the trust graph.
Missing trust signals: entity pages, Wikidata, verified profiles
Last blocker: external verification signals. AI doesn’t take a site at its word. It wants proof. Wikidata. Google Knowledge Graph. Press mentions. Official social profiles with reciprocal links.
In my audit of 71 companies, most didn’t have a structured entity page. Some weren’t even in Wikidata. Others had LinkedIn profiles not linked to their domain.
I fixed that for a cybersecurity SME. Created a sourced Wikidata entry. Aligned the site, profiles, and registration ID. In 48 hours, trust was back. Perplexity started citing the company on technical queries.
The knowledge graph isn’t a luxury. It’s your sesame for AI Search.
48 hours to become visible again: my retrievability audit method
I don’t believe in 80-slide audits you read in two weeks. My diagnosis happens live, on screen. 45 minutes. We look at three things.
One: AI crawl. I submit your site to a GPT agent and watch what it reads. If main content is invisible, we see it right away.
Two: entity mapping. I verify your Knowledge Graph entry, your schema markup, your web citations. At each broken link, I note the missing signal.
Three: peripheral domains and profiles. I track dead redirects, old brands not aligned.
This trinity follows the DOSE discipline: Data, Organization, Semantic, Expérience. AI verification demands order. No band-aids, no vague promises.
One call. You see your gaps. You know exactly what to fix.
See your gaps live, not in a report
I take your site, run it through AI scrutiny in front of you. In 45 minutes, you’ll know exactly what’s blocking you. No pitch. Just pages, signals, and the plan.
Book a strategic call — 45 minFrequently Asked Questions
How do I know if my company is verifiable by AI?
Ask a specific question about your company in ChatGPT, Perplexity, or Gemini. If AI doesn’t cite you or gives outdated info, your retrievability is weak.
How long does it take to fix missing signals?
Technical fixes (readability, dead domains) take 24 to 72 hours. External signals like Wikidata take about a week. The effect on AI citations appears in less than 15 days.
Can a well-ranked WordPress site be invisible to AI?
Yes. If the theme loads content via JavaScript without server-side rendering, or if schema markup is missing, AI can ignore the site. Google ranking doesn’t guarantee AI will retrieve the content.
Do I need to create a Wikidata page to appear in AI Search?
Not required. But it helps. Wikidata feeds many knowledge graphs. A sourced entry with external identifiers (registration ID, official links) is easier to verify.
Does the retrievability audit include fixing the site?
The live audit identifiés blockers and missing signals. I give you a roadmap with specific actions for your tech team. I can help apply them, but that’s not included in the initial 45-minute audit.
