AIs don't agree about your brand: 1 question in 4

<p><em>We talk about "AI" as one thing. But ask ChatGPT, Gemini, Claude and Perplexity the same question: do they say the same thing about your brand? We measured. The answer is no — and the gap is wide.</em></p> <h2>The scope</h2> <p>We took <strong>617 questions</strong>, each asked to <strong>several models in parallel</strong> (4.3 models on average: ChatGPT, Claude, Gemini, Perplexity and others), in real audits. For each answer, we measure whether the audited brand <strong>actually appears</strong> (entity detection on the produced text). Then, question by question, we check whether the models <strong>agree</strong>.</p> <h2>The finding</h2> <div style='background:#faf5ff;border:1px solid #ede9fe;border-radius:14px;padding:18px 20px;margin:16px 0;'><div style='font-size:2.4rem;font-weight:800;color:#6d28d9;font-variant-numeric:tabular-nums;line-height:1;'>23%</div><div style='font-size:0.95rem;color:#4b5563;margin-top:6px;'>of questions where models disagree on whether the brand appears</div></div> <p>In <strong>nearly one question in four</strong>, at least one model cites the brand and at least one other does not — on the exact same question. In other words: your visibility is not a stable property, it is a <strong>variable that depends on which model is asked</strong>.</p> <h2>Why AIs diverge</h2> <p>Three reasons combine. First, each model has a <strong>different training memory</strong>, frozen at a different date: what one "knows" about you, another ignores. Second, some models <strong>browse the web</strong> and others answer from memory — two distinct brand images (we measured a 7-point gap in our <a href="/blog/en/ai-visibility-barometer-brands-2026">barometer</a>). Third, they do not <strong>draw on the same sources</strong>: LinkedIn, Wikipedia, reviews, registries weigh differently from one model to the next (see our <a href="/blog/en/sources-ais-cite-most-about-brands-study">study on sources</a>).</p> <h2>What it implies</h2> <p>A visibility audit on <strong>a single model</strong> measures one reality out of four. Claiming to be "visible on ChatGPT" says nothing about Gemini, Claude or Perplexity — and your customers use one or the other at random. That is precisely why we query <strong>50+ models</strong> and reason in distribution, with confidence intervals, rather than on a single number. The method is public on our <a href="/en/ags-methodology">methodology page</a>, and the <a href="/en/observatory">Observatory</a> tracks these figures over time.</p> <h2>The limits of these figures</h2> <p>This data comes from brands that chose to audit themselves — not a representative sample. "Disagreement" is about presence (the brand appears or not), not the detail of wording. Rates evolve with every model update.</p> <p>Want to know which models you appear on — and which ones you're invisible on? That is exactly what an audit measures, model by model.</p>