EN
FR
ES
DE
IT
KO
JA
<p><em>We regularly publish aggregated, anonymised figures from our platform. No consultancy estimates: real, dated measurements with their exact scope. Because a statistic without a scope is worth nothing.</em></p>
<h2>The scope</h2>
<p>Between September 2025 and June 2026, the platform ran <strong>104 complete audits for 65 brands</strong>, analysing <strong>6,820 answers</strong> produced by <strong>44 distinct AI models</strong> (ChatGPT, Claude, Gemini, Perplexity, Mistral, DeepSeek and others), in native mode (model memory) and web mode (online search). The full methodology, confidence intervals included, is documented on our <a href="/en/ags-methodology">methodology page</a> — including what we do not measure.</p>
<h2>Finding 1 — Unprompted visibility: 16%</h2>
<p>On questions a customer would ask <strong>without naming the brand</strong> (“which solution for…”, “who would you recommend for…”), the audited brand appears in only <strong>16% of answers</strong> (106 mentions out of 657 answers, a sub-sample of 13 brands with entity extraction). In other words: in more than 8 answers out of 10, a competitor — or nobody — holds the ground.</p>
<h2>Finding 2 — AIs invent web addresses for 66% of brands</h2>
<p>This is the figure that surprises our clients most: <strong>699 non-existent URLs</strong> cited by AIs were detected, affecting <strong>43 brands out of 65</strong> (66%). AIs fabricate plausible links to pages that never existed. And it is not theoretical: <strong>8 of these invented addresses received real visits</strong> — humans clicking and landing on an error page.</p>
<h2>Finding 3 — Memory and web: two brand images, 7 points apart</h2>
<p>Across the 63 audits measuring both modes, the average visibility score is <strong>63.4/100 in native mode</strong> (the model's memory, no browsing) versus <strong>70.5/100 in web mode</strong> (online search enabled). Model memory under-represents brands: what an AI “knows” about you dates from its training, and only web mode reflects your recent content. Any strategy ignoring one of the two modes measures half the problem.</p>
<h2>Finding 4 — When a brand appears, it ranks well</h2>
<p>Across the 4,757 answers where the audited brand is mentioned, its average position is <strong>2.0</strong>. The practical conclusion is counter-intuitive: the battle is not about rank within the answer, it is about <strong>getting into the answer</strong>. Once cited, you are rarely buried.</p>
<h2>The limits of these figures</h2>
<p>This data comes from brands that chose to audit themselves — a sample that is not representative of all companies. Sub-samples are stated for each finding. Figures evolve with every model update; we will republish this barometer with fresh data.</p>
<p>Want to know where a specific brand stands? That is exactly what an audit measures — and post-action re-testing proves progress, numbers in hand.</p>
When a customer asks AIs without naming a brand, someone else gets recommended. What about the brands you manage?
Every question asked to ChatGPT without the brand in the answer is a competitor recommended in its place: measured across 6,820 real AI answers.