We tested 50 businesses on ChatGPT. 94% were excluded.

Between February and March 2026, the AVO research practice ran a controlled test across ten UK industries. We selected fifty businesses that, by traditional measures, were performing well: ranking on page one, receiving organic leads, with active content programmes.

We then asked ChatGPT, Gemini, Perplexity and Claude a set of buyer-shaped questions. Not branded queries. The questions a real buyer would ask: "best ergonomic office chair supplier in Manchester", "London accountancy firm for SaaS startups", "private cardiology consultant in Birmingham".

The headline

Of the fifty businesses, three were named at least once across all four AI systems. Forty-seven were not named at all.

That is a 94% exclusion rate from a sample that, by every conventional metric, is healthy.

The pattern of exclusion

The excluded businesses share four characteristics:

• Their entity is ambiguous. Service language is broad and aspirational rather than precise.
• Their signals disagree across platforms. Different addresses, different categories, different descriptions on the website, Google profile, LinkedIn and directories.
• Their authority is self-asserted rather than third-party verified. Testimonials on their own site, but not corroborated externally.
• Their content addresses keywords, not the actual buyer questions that AI systems answer.

The pattern of inclusion

The three businesses named by AI shared the inverse profile. Tight, specific positioning. Consistent metadata across every platform. Citations from independent third parties (industry press, professional bodies, customer-published references). Content that answers a real buyer question end-to-end rather than targeting a phrase.

Why this matters more than ranking loss

Conventional traffic loss is visible. You see it in your analytics, you can attribute it, you can react. AI exclusion is silent. The buyer never reaches your site, your analytics show nothing, the leads simply do not arrive.

The 94% are not in distress today. Most still receive enough demand from the residual SEO surface to feel comfortable. The diversion is gradual and undiagnosed. By the time it is felt, the AI's preferred set of recommendations is well established and far harder to dislodge.

What we recommend

Run the test on yourself. Ask the four major AI systems the questions a buyer would ask in your category. Note who is named. If you are not named, you are not losing rank. You are missing from the answer entirely. That is a different problem and it requires a different discipline.