SEO & GEO Summit 2026: cited by AI, or actually recommended?

<p>On October 15 and 16, 2026, the SEO &amp; GEO Summit brought agencies, brands, and AI visibility tools together at Disneyland Paris. One 30-minute talk captured the hallway conversations: <em>How do AI assistants choose the brands they recommend?</em>, by Guillaume Guersan, built on AI Labs Audit data.</p> <p>The premise is one sentence everyone agrees with and almost nobody acts on. A mention inside ChatGPT or Gemini looks like an impression in Search Console. It is a signal. On its own, it makes zero euros.</p> <h2>A word about Guillaume Guersan</h2> <p>Before the numbers, it is worth saying who is speaking. <strong>Guillaume Guersan</strong> is a digital marketing and acquisition consultant, specialised in SEO, Google Ads and Meta Ads. More than ten years in the field, over 1,000 clients supported, over 11,000 people trained. He is not a theorist of AI visibility: he has lived through algorithm shifts, Google updates, and then assistants landing in everyone's pocket.</p> <p>His favourite question for a prospect who already worked on GEO is worth keeping as is: "When an AI cites your brand, does it actually recommend you?". In one line, he moves the debate from vanity metrics to the commercial question. At the SEO &amp; GEO Summit, he shared the outcome of that field-level view, backed by the study run with AI Labs Audit. This post pays tribute to him by restating the core of his talk, without flattening it.</p> <h2>Three levels people still mix up</h2> <p>The talk starts with a simple grid, and that grid is already a filter. Every brand wants to show up in AI answers. Few know <strong>where</strong> they show up.</p> <table style="width:100%; border-collapse:collapse; margin: 1.2em 0;"> <thead> <tr style="background:rgba(102,126,234,0.1);"> <th style="text-align:left; padding:8px;">Level</th> <th style="text-align:left; padding:8px;">What the model does</th> <th style="text-align:left; padding:8px;">Example</th> </tr> </thead> <tbody> <tr> <td style="padding:8px;"><strong>Cited</strong></td> <td style="padding:8px;">The brand appears in context, without being offered as a solution.</td> <td style="padding:8px;">"Delsey makes suitcases in several ranges."</td> </tr> <tr> <td style="padding:8px;"><strong>Considered</strong></td> <td style="padding:8px;">The model places it in a shortlist.</td> <td style="padding:8px;">"Samsonite, Delsey and American Tourister are among the cabin options."</td> </tr> <tr> <td style="padding:8px;"><strong>Recommended</strong></td> <td style="padding:8px;">The model advises it for a specific need.</td> <td style="padding:8px;">"I recommend the Samsonite Lite-Box if you often travel with a cabin bag."</td> </tr> </tbody> </table> <p>Moving from the first cell to the third is moving from awareness to revenue. A prospect who reads your name in a list has not chosen yet. A prospect who gets a product named for <em>their</em> case has something to buy.</p> <h2>The protocol behind the numbers</h2> <p>To leave storytelling aside, the study crossed <strong>9 French markets</strong>, <strong>20 questions per market</strong>, <strong>5 AI assistants</strong>, and two answer modes (with and without web search). Total: <strong>4,320 answers</strong>, 24,974 citations, 6,147 recommendations, and 19,317 source links exposed in web-enabled answers.</p> <p>The markets are the same ones we publish in the Observatory, with the AI answers available market by market:</p> <ul> <li>Electric bikes: <a href="/en/observatory/studies/velos-electriques">see the study and the answers</a></li> <li>Online accounting: <a href="/en/observatory/studies/comptabilite-en-ligne">see the study and the answers</a></li> <li>Dog health insurance: <a href="/en/observatory/studies/mutuelle-sante-chien">see the study and the answers</a></li> <li>Car insurance: <a href="/en/observatory/studies/assurance-automobile">see the study and the answers</a></li> <li>Mattresses: <a href="/en/observatory/studies/matelas">see the study and the answers</a></li> <li>CRM software: <a href="/en/observatory/studies/logiciels-crm">see the study and the answers</a></li> <li>SEO agencies: <a href="/en/observatory/studies/agences-referencement-seo">see the study and the answers</a></li> <li>Cabin luggage: <a href="/en/observatory/studies/valises-cabine">see the study and the answers</a></li> <li>Invoicing software: <a href="/en/observatory/studies/logiciels-facturation">see the study and the answers</a></li> </ul> <p>Models: Claude Sonnet 4.6, Gemini 2.5 Flash, Mistral Large, ChatGPT (GPT-5 mini), and Perplexity Sonar Pro, on the same analysis grid. The questions test three behaviours: discover (which brands exist), compare (A or B), choose (what do you advise for my case). One answer is an anecdote. A protocol shows what repeats.</p> <h2>Four findings that change the work</h2> <h3>1. Popularity is not conversion</h3> <p>On dog health insurance, SantéVet dominates citations: 302 mentions against 69 for Lassie. On explicit recommendations, the gap shrinks (59 against 29). And if you look at how often a citation becomes advice, the ranking flips: <strong>42% of Lassie citations become a recommendation, against 19.5% for SantéVet</strong>.</p> <p>SantéVet wins surface. Lassie wins commercial efficiency. In the whole study, only <strong>24.6% of citations</strong> are explicit recommendations. Stop celebrating mentions without reading the next line.</p> <h3>2. The question steers the answer</h3> <p>The share of answers that recommend at least one studied brand depends mostly on how the question is framed. "For my case" reaches 55.7%. "A or B" sits at 4.6%. Vanity prompts measure almost nothing. The prompts your buyers actually type, those measure the deal.</p> <h3>3. With or without web, the market reshuffles</h3> <p>Out of 216 observed brands, <strong>214 show a non-zero gap</strong> in citation frequency between native and web mode. 154 brands lose visibility in web mode, 60 gain some.</p> <p>The sharpest case: invoicing software. Without web, Zervant leads (cited in 127 answers out of 180). With web, Tiime takes over (106 out of 180). Among the 180 web answers, 27 show a link to tiime.fr. Only 2 show a link to zervant.com.</p> <p>Model memory and the live web tell different stories. We already documented that gap in our <a href="/blog/en/native-vs-web-score-measure-both-realities-of-your-ai-presence">native vs web comparison</a> and in the <a href="/blog/en/ai-visibility-barometer-brands-2026">AI visibility barometer</a>.</p> <h3>4. There is no single AI market</h3> <p>Same 20 car-insurance questions, same conditions, three assistants, three leaders: Direct Assurance on Claude, Allianz on Mistral, MAIF on ChatGPT. The ranking measures citations, not product quality. It mainly measures that an "AI ranking" does not exist outside of model, language, date, and answer mode. The full SEO agency study is <a href="/en/observatory/studies/agences-referencement-seo">available in the Observatory</a>.</p> <h2>When the model is wrong about you</h2> <p>Visibility does not guarantee accuracy or freshness. In the invoicing corpus, two outdated tips keep coming back: QuickBooks France (local offer stopped at the end of 2023) and Debitoor (no longer a standalone app). Assistants can recommend a dead offer, invent a price, or mix up two product lines.</p> <p>Watch what they claim, not only whether they mention you. Same reflex as with <a href="/blog/en/hallucinated-urls-when-ai-invents-pages-on-your-website">invented URLs</a>.</p> <h2>Six decisions to move from cited to recommended</h2> <ol> <li><strong>Stop celebrating citations.</strong> Measure recommendations separately, with the prompt and the reason.</li> <li><strong>Watch the right prompts.</strong> Your buyers' prompts, not the ego ones. A stable panel, replayed over time.</li> <li><strong>Build two kinds of pages.</strong> "Your service for a precise persona" and "Alternative to competitor X", with limits, proof, and choice criteria.</li> <li><strong>Find the sources that matter.</strong> The ones already exposed by AI answers on your queries. Our study on <a href="/blog/en/sources-ais-cite-most-about-brands-study">the sources AI cites most</a> gives the order of magnitude.</li> <li><strong>Show customer proof.</strong> Context, problem, solution, measured results, verified reviews.</li> <li><strong>Measure everything, often.</strong> Citations, recommendations, false claims. Same prompts, native and web, every month.</li> </ol> <h2>Go deeper in the Observatory</h2> <p>The numbers presented at the Summit did not stay in a conference room. Same markets, same question families, AI answers in the open: all of it is published in the <a href="/en/observatory">AI Labs Audit Observatory</a>. Open a study, see who is cited, who is recommended, and what assistants say about a given brand.</p> <p>The useful question is no longer "do we show up?". It is: at which level, on which questions, against which competitors, and with which proofs?</p> <p><em>More studies and talks in the <a href="/blog/en">AI Labs Audit blog</a>, and our method on the <a href="/en/ags-methodology">methodology page</a>.</em></p>