AI visibility: the 68 000 answers method to replicate on your e-commerce

A 100-point score, three measurable levers

I look at 15 e-commerce sites per week. Their classic SEO holds up. Their visibility in AI answers remains low. The Neil Patel team analyzed 68 000 answers generated by ChatGPT, Gemini, Claude and Perplexity. They wanted to understand why some brands appear in these answers and others stay invisible. Three criteria decide. Frequency of appearance. Position in the answer. Number of catégories covered. These three criteria form a 100-point score. I now apply it to every e-commerce catalog I audit. Simpler than tracking 40 keywords. Take a purchase question: « What is the best stick vacuum under €250? » The AI returns a list of 3 to 5 models. Your brand appears in 2 out of 5 models? Your frequency is 40% on that question. Frequency alone is not enough. Cited last, your brand carries less weight. And if it only appears in the « vacuums » category, it is not recognized as a reference for filtration or battery life. The Neil Patel study confirms it: brands that dominate win on a scope of questions, not on one keyword. Classic SEO optimizes pages for queries. AI visibility optimizes a catalog for answers. A small appliance client, for example. Their site was in position 3 on Google for « stick vacuum ». In ChatGPT answers, it was absent. The content held. It was the structure of the answers that blocked it. The AI visibility score on 100 points forces you to think differently. You target less the first blue link. You target the source the AI cites. The 68 000 analyzed answers give a measurable method. Concrete. The rest breaks down each of the three levers. And how I deploy them on an e-commerce catalog.

Pour mesurer la fréquence, un point de départ simple : 100 questions de votre niche génèrent 400 réponses IA. Comptez ensuite combien de fois votre marque y est citée.

Frequency of appearance: be the name the AI repeats

Frequency of appearance is the number of times your brand is cited in AI answers across a set of questions. 100 questions in your niche generate 400 AI answers? Count how many times you come up. Neil Patel showed that frequency is the first factor of the score. On a catalog of 1 240 references, I identified 87 recurring purchase questions. What questions? Not keywords. Complete questions, worded as spoken: "Which double stroller for twins in the city?", "Which stand mixer for kneading bread dough?", "Which baby puree for a 6-month-old with no lumps?". Each question calls for an extractable answer. A block of text or data the AI can cite without reformulating everything. A classic product sheet is not enough. It tells the product's story. The AI, on the other hand, looks for an answer to a specific question. With that client, I created an answer block on each concerned product sheet. The block starts with the question. Then the answer in two sentences. Then the quantified specification. Example: "For dense bread dough, the X7 mixer with 1 600 W motor and spiral hook is suitable. Kneading time: 9 minutes for 800 g of flour." Result observed on 50 tested questions: the brand went from 12 to 38 appearances in the answers of the 4 AI engines in 90 days. An order of magnitude. Not a promise. The mechanism is simple. When the AI finds a sentence that answers exactly, it uses it. If it comes across a brand manifesto, it ignores it. Frequency rises when you multiply answer blocks on the questions your customers actually ask. At Neil Patel, high-frequency brands produce the most direct answers, not the most content. I often see sites with 800 product sheets and zero question-answer blocks. They win by adding a 60-word block on the 30 to 50 questions that matter. Frequency progresses question by question, without a full overhaul.

Position: being cited in the first 40 words

The second lever of the AI visibility score is position. In an AI answer, being cited first or second does not carry the same weight. Neil Patel measured it on 68 000 answers: the earlier the brand is mentioned, the higher the score. Normal. A language model builds its answer like an ordered list. The first names are the ones it judges most relevant for the question. On an e-commerce site, position is played on the first sentence of the product sheet. Not in the title tag. In the first visible paragraph. Look. A classic sheet starts with: "Our family business has been designing innovative stand mixers for 25 years." An AI-optimized sheet starts with: "The X7 stand mixer is suitable for kneading 800 g of flour in 9 minutes thanks to its 1 600 W motor and spiral hook." The first sentence answers no question. The second answers "Which mixer for kneading bread dough?". The AI cites the second. Not the first. I work with a baby care e-commerce merchant. On the question "Which car seat for a 2-year-old measuring 90 cm?", their site was in position 4 on Google. In Gemini's answer, it was absent. The first paragraph of their sheet described the brand. Not the answer. After restructuring the first paragraph, Gemini started citing it in second position. The Google page had not moved. The AI visibility score climbed by 11 points on that scope. Position in the AI answer also depends on factual density. AI models like precise numbers, dimensions, compatibilities. A paragraph that contains "90 cm, 15-36 kg, R129 standard" is easier to cite than a vague sentence. No need to rewrite the whole catalog. Just the 20 to 30 questions with high purchase intent. Start with the first paragraph of each sheet. Make it answer a question. Not a brand. The position gain is quick. AI models recalculate their answers at each update of their base. You see changes in a few weeks, not in six months.

Catégories: a wide mesh does not dilute, it strengthens

Third lever of the AI visibility score: the number of catégories in which your brand appears. Neil Patel measured it on 68 000 answers. A brand that shows up in 3 catégories gets a better score than a brand that crushes a single category. For an e-commerce, that changes everything. Many sites only optimize their product sheets. Result: they show up in the "product" category. Never in "usage", "maintenance", "comparison" or "safety". A running shoe site, for example. Customer questions are not just about "stable running shoes". They talk about "how to avoid knee tendinitis", "which sole for running on asphalt", "how to choose your size for a marathon". With only product sheets, the site appears in none of these question catégories. Its counter stays at 1. The way out: create relay content. Guides, comparison articles, technical FAQs. Not for classic SEO. So that the AI draws from several catégories of your catalog. I build semantic cocoons around the main catégories. Each cocoon brings together 15 to 20 adjacent questions, linked to the category's pillar page. The AI reads this architecture as a coherent network. It cites you in more catégories, without diluting authority. The DOSE framework, taught by Guillaume Attias at BMO Academy, helps me structure these links between query, intent and content. Each question has a place. Each link strengthens the category. With a baby care client, I expanded the scope from 2 to 7 question catégories: "strollers", "car seats", "baby food", "sleep", "hygiene", "safety", "outings". Without creating hundreds of pages. Just 45 relay content pieces in 90 days. The number of catégories, measured on the 4 AI engines, doubled. The result is mechanical: the more question catégories you cover, the more you increase the probability of appearing in answers. And the more your 100-point score rises. Careful though: each relay content answers a specific question. A generalist 2 000-word guide on "how to choose a stroller" is not enough. The AI wants specific answers, question by question. Category coverage is long-term. It builds a global presence in the minds of AI models. You are no longer a site that sells shoes. You are a source that answers the runner's questions.

Voici le pipeline que j'applique pour passer de 0 à 100 points en 90 jours. Quatre phases, du repérage des questions jusqu'à la mesure du score.

E-commerce deployment: 4 phases over 90 days

Going from 0 to 100 points on the AI visibility score does not take a year. It takes 90 days, provided you deploy the three levers on a restricted scope. I apply 4 phases on e-commerce catalogs. Phase 1: auditing AI questions. I collect the questions your customers actually ask. In Google suggestions, forums, competitor FAQs, customer reviews. I group the questions by intent: purchase, comparison, usage. In général, I come out with 120 to 180 actionable questions for a category of 500 references. Phase 2: prioritization. Not all questions are equal. I keep the 20 to 30 that combine search volume, commercial intent and category coverage. That is the starting core. Phase 3: production of answer blocks. On the concerned product sheets, I add a "Question / Answer" block at the top of the description. On category pages, I add a "The 5 questions before buying" box. On the blog, I create relay content of 800 to 1 200 words targeting a precise question. Each block respects three rules: answer in 60 to 80 words, precise numbers, spoken wording. Phase 4: measurement. Every 2 weeks, I ask the 30 questions to the 4 AI engines. I note the frequency of appearance, the position, the number of catégories. I get a 100-point score. I adjust.

With a client of 2 300 baby care references, the AI visibility score went from 34 to 61 points in 90 days. It is an order of magnitude measured on the 4 engines, not a guarantee. Reproducible, however, if you apply the phases in order. You can deploy step by step. Starting with 15 questions per month is enough. The score rises gradually. Frequency, position, catégories reinforce each other. Everything rests on measurement. The 100-point score tells you exactly where you are winning. I advise you to start with your most profitable category. The rest of the catalog will come later. AI visibility is built scope by scope.

The shift: AI doesn't rank, it chooses

The Neil Patel study on 68 000 AI answers shows that the profession is changing. Classic SEO ranks pages by authority. Generative AI, on the other hand, chooses sources to build an answer. When a page is ranked on Google, you are a blue link among ten. The click depends on your meta title. When an AI generates an answer, it cites your information in a fluid text. No click. A mention. If you are not mentioned, you do not exist on that channel. According to the study, the brands that win are not those with the most backlinks. They are those that provide the most extractable answers. A block "Weight: 2.4 kg, battery life: 45 minutes, surface: 68 m²" is cited more than a storytelling paragraph. I verify it every week on the 15 sites I observe. Product sheets with specification tables at the top appear more often in ChatGPT answers. Those with a long presentation text remain little reused. The method is counter-intuitive. AI does not reward content volume. It rewards factual density. A 15-word sentence with 3 numbers is worth more than a 120-word paragraph without data. For an e-commerce, that means every product sheet must start with an answer, not a story. Every category page must list the questions before presenting the products. Every article must be an answer block, not an essay. You do not need to reinvent your site. You need to restructure your first paragraphs and add question-answer blocks. The 100-point score is a steering tool. It forces you to look at AI visibility as a measurable channel, not as a vague trend. The 68 000 answers analyzed by Neil Patel give a reliable base. They show that AI visibility is worked on upstream, on information structure. Once in place, it runs without you. My conviction: within 12 months, e-commerce merchants who structure their AI answers will capture a share of conversational traffic without paying for clicks. Those who wait will have more ground to catch up. One question remains: how many questions can your catalog handle today? What is the first purchase question your customers ask ChatGPT? And does your product sheet answer it?

Stéphane Jambu

SEO & AI Engineer

I build growth systems / AI / Neuroscience | 650+ clients · 80 LinkedIn testimonials · 30 years of expertise · 15 years of systems running without me.

Follow on LinkedIn