How to tell whether ChatGPT knows your company — and what to do about it

In short: Ask ChatGPT about your company five times in a fresh chat — directly, by category, on facts, on sources. A single answer tells you almost nothing: according to a variance analysis published in July 2026, one individual AI answer has virtually no power to distinguish between brands. Only the hit rate across several runs and languages counts as a finding. And a strong Google ranking barely helps: only 12% of URLs cited by AI tools also appear in Google’s top 10.

Why does ChatGPT know some companies and not others?

Because AI models learn about your company not from your website, but from how other people write about you. In December 2025, Ahrefs studied 75,000 brands and matched brand signals against actual mentions in ChatGPT, Google AI Mode, and AI Overviews. The result is unambiguous: YouTube mentions correlate most strongly with AI visibility (around 0.74), closely followed by branded web mentions (0.66 for ChatGPT). The number of pages on your own site correlates at 0.19 — which is to say, barely at all.

Translated: more content on your own domain does not make you more visible in AI answers. Being mentioned does — in trade articles, directories, forums, videos, comparisons. For mid-sized companies that is good news. These signals cannot be bought, but they can be earned.

How do you test in 30 minutes whether ChatGPT knows your company?

With four question types, each asked five times, every time in a fresh chat with no history — and in every language you do business in. The history matters: if you follow up in the same chat window, the model answers based on what you just told it yourself.

1. Direct: “What is [company name]?” — Does the model recognize you as an entity at all?
2. Category: “Which providers of [your service] are there in [your region]?” — Do you get named without your name appearing in the question? This is the test that counts, because that is how buyers ask.
3. Facts: “Where is [company name] based, how large is it, what does it offer?” — Is any of it correct?
4. Sources: “Where does this information come from?” — Which pages get cited? That is where your visibility is actually decided.

Record the hit rate per question type across all runs, not your impression after the first answer. Repeat the round in ChatGPT, Perplexity, and Google AI Mode: an analysis of 680 million AI citations found that only about 11% of domains are cited by both ChatGPT and Perplexity. Checking one platform means checking one ninth of the picture.

Why is a single AI answer worthless as a test result?

Because the noise is larger than the signal. A variance-components analysis by Dmitrij Żatuchin (Estonian Entrepreneurship University of Applied Sciences, arXiv, July 2026) decomposed roughly 12,900 AI answers covering 20 brands across eight languages and three models. The sobering headline figure: a single query reaches a reliability of 0.010 — it effectively does not distinguish between brands. Around 35% of the observed spread comes from nothing more than asking the same question repeatedly.

More interesting than the noise is where it comes from. In the same analysis, the query language explains around 32% of the variance — more than model, brand, and prompt combined. A further 8.6% falls to the interaction between brand and language. A company can be highly visible in German and practically invisible in Polish without a single change to its website. For companies doing business across D-A-CH and Poland, that is not a detail; it is the metric.

The practical consequence from the same paper: adding languages buys roughly fifteen times more explanatory power per query than adding repetitions of the same question. Past about five repetitions, the sixth is no longer worth it — spend that budget on languages and platforms instead.

What do you do if ChatGPT does not know your company at all?

You earn mentions in places AI systems read — and you clean up your entity. Wikipedia is the dominant reference at roughly 48% of ChatGPT’s most-used top-10 source share; Reddit dominates Perplexity at around 47%. Neither can be steered by marketing, only by genuine relevance and honest participation.

What you can steer is the entity foundation. According to the 5W analysis, a consistent entry in Wikidata plus presence on at least four authoritative third-party platforms is associated with a 2.8x higher likelihood of being cited. In practice: identical company name, address, and service description across business registries, industry directories, LinkedIn, and specialist platforms — plus structured data (Organization schema) on your own site. It is unglamorous homework, which is exactly why so few companies do it.

The second most common blind spot: expert content that delivers a verified figure with a clear source. AI systems extract quotable statements, not marketing copy. How to prepare your own pages for that is covered in our article AI visibility: optimize your website for AI.

What do you do if ChatGPT says something false about your company?

You correct the source, not the model. LLMs do not form an independent judgment about your company; they reproduce the statements and the sentiment of the sources they draw on. If an old directory entry still lists your 2019 address, that is the address you get back.

That the models also simply invent things is documented: Ahrefs checked 16 million URLs cited by AI assistants. 2.38% of the addresses cited by ChatGPT led nowhere — against 0.84% for Google’s top-20 results. Perplexity (0.87%) and Gemini (0.86%) sat essentially at Google’s level; ChatGPT was the clear outlier. So in your self-test, always check whether the sources named actually exist.

One more thing many underestimate: AI systems have a long memory for documented crises. An old conflict preserved in press coverage still surfaces in AI answers years later, in the present tense. The only remedy is newer, well-sourced substance — not an attempt at deletion.

Why does a strong Google ranking help so little here?

Because AI systems draw from a different pool. An Ahrefs analysis of 15,000 queries found that only 12% of the URLs cited by AI tools also appear in Google’s top 10. The other 88% come from sources that never show up on page one — forums, directories, trade portals, deeper subpages of authoritative domains.

The reverse holds too: classic search remains the channel that actually brings people to your website. How the two relate is laid out in our article GEO vs. SEO: get visible in AI answers. That is where weooo starts — building AI visibility as an addition to existing visibility, not a replacement for it. Measurable through a defined prompt set, GDPR-compliant, without customer data moving into third-party systems. No buzzword bingo, just a number you can measure again three months later.

What does AI visibility realistically deliver — and what doesn’t it?

Far fewer clicks than the debate suggests. By Ahrefs’ calculations, ChatGPT reaches about 12% of Google’s search volume for queries people used to google — but sends 190 times fewer visitors to websites. Its click-through rate is 96% below Google’s. Anyone selling AI visibility as a traffic channel is selling an exaggeration.

The value sits elsewhere: in the mention itself. ChatGPT names brands roughly 3.2 times more often than it links to them. Your company can appear on a decision-maker’s shortlist without a single click ever showing up in your analytics — or fail to appear without you ever noticing. That is precisely why the self-test is worth doing and an AI traffic dashboard largely is not.

And the honest limit of measurement: citation patterns are volatile. Observing over 230,000 prompts, Semrush saw ChatGPT’s Reddit share fall from around 60% to roughly 10% within a few weeks in September 2025. Building a strategy on a single source or platform is building on sand. Quarterly measurement is enough — daily measurement produces noise.

Conclusion

Whether ChatGPT knows your company can be settled in half an hour — but only if you ask repeatedly, in several languages, across several platforms, and evaluate the hit rate rather than the first answer. If the test comes back negative, the cause is rarely on your website and almost always outside it: missing mentions and a messy entity. The realistic payoff is not a wave of traffic but the chance to appear in the shortlist where it is increasingly decided who gets asked for a quote at all.

Want to know how your company is described in ChatGPT, Perplexity, and Google AI Mode — in German, English, and Polish? Talk to us and we will turn the measurement into a baseline you can rely on.

Sources

• Ahrefs, “Top Brand Visibility Factors in ChatGPT, AI Mode, and AI Overviews” (12 Dec 2025) — 75,000 brands, Spearman correlation; YouTube mentions ~0.737, branded web mentions 0.664 (ChatGPT), number of site pages ~0.194. Authors’ note: correlation is not causation.
• Ahrefs, “How Often Do AI Assistants Hallucinate Links?” (2 Sep 2025, updated 31 May 2026) — 16 million cited URLs; ChatGPT 2.38% dead links, Perplexity 0.87%, Gemini 0.86%, Google top-20 baseline 0.84%.
• Ahrefs, “ChatGPT Has 12% of Google’s Search Volume but Google Sends 190x More Traffic to Websites” (11 Feb 2026) — Ahrefs Web Analytics across 76,000 websites; click-through rate 96% below Google.
• 5W Research, “The State of AI Citations 2026” (May 2026) — synthesis including Profound (680 million citations), Semrush (230,000 prompts), Ahrefs (15,000 queries); only 11% domain overlap between ChatGPT and Perplexity, 12% URL overlap with Google’s top 10, Wikipedia ~47.9% of ChatGPT’s top-10 source share, Reddit ~46.7% for Perplexity, ChatGPT mentions brands ~3.2x more often than it links to them, entity presence on ≥ 4 third-party platforms associated with ~2.8x higher citation likelihood.
• Semrush, “The Most-Cited Domains in AI: A 3-Month Study” (Nov 2025) — 230,000 prompts over 13 weeks; ChatGPT’s Reddit share collapsed from ~60% to ~10% in September 2025.
• D. Żatuchin, “Where Does the Noise Come From? A Variance-Components Decomposition of Non-Determinism in LLM Brand Answers”, arXiv:2607.13304 (14 Jul 2026) — 12,933 usable answers, 20 brands × 8 languages × 3 models; single-query reliability 0.010, resampling ~34.8% of variance, query language ~32%, brand × language 8.6%. Limitations: point estimates without confidence intervals, a single collection window, temperature 0.3; the author has an interest in an AI visibility measurement provider.