The short version
We ran Quince, the affordable-luxury clothing brand, through ChatGPT, Google Gemini, and Anthropic Claude, then read every question real buyers bring to those models. The questions did not come from one customer. They came from at least six very different people.
- You don't have one audience. You have several. The questions about a single clothing brand came from new parents, capsule-wardrobe minimalists, value hunters, sustainability skeptics, plus-size shoppers, and athletes, each asking from their own life.
- AI learns what you're for from where those people talk. Most of what AI knew about Quince came from third-party sites, and the loudest ones were niche mom blogs and sustainability guides, not a generic fashion forum.
- Typing your own prompts into an AEO tool is keyword tracking by another name, a guess at words your real buyers may never use. Find the people before you write the questions, or you test a conversation none of them are having.
Who actually asks AI about your brand?
Several different buyers ask about you, each from their own life. Reading one brand's real question set, you can sort the people behind the questions almost by hand.
When we ran our report on Quince, the questions AI fields about it sorted into clearly different lives:
- New parents. "Best jeans for postpartum bodies with wide hips and smaller waist no crotch bunching." "Best bras for exclusively pumping moms with sensitive nipples and chafing." "Best underwear styles that don't irritate a post C-section incision during healing."
- Capsule-wardrobe minimalists. "What brands make clothes that work as a capsule wardrobe, pieces that go together and don't go out of style?"
- Value hunters. "Which brands make really nice quality clothes without charging luxury prices? Looking for something that feels expensive but doesn't cost that way."
- Sustainability skeptics. "Are Quince's organic clothing claims actually certified, or is it just marketing hype?"
- Inclusive-fit shoppers. "What clothing brands actually carry a full range of sizes from XS to 4X or beyond?"
- Athletes. "What activewear brands actually perform well, moisture-wicking, durable, and comfortable for workouts?"
That is one brand, one report, 219 real conversations across the three models, and the people behind them split into six distinct audiences. A new mom and a minimalist are both Quince customers, and their questions barely overlap.
Why does the persona decide the question, not the keyword?
The persona decides the question because people ask AI from a situation, not a keyword. A keyword gives you a Google search ranking. A person gives you a concern with a budget, a body, and a deadline attached.
Nobody guessing keywords for a clothing brand writes "C-section incision." A marketer would type "postpartum clothing" and move on. The real buyer types the whole worried sentence, because she is solving a specific problem six weeks after giving birth. Change the persona and the question changes completely. The minimalist isn't worried about her waistline, she's worried about buying pieces she'll still wear in three years. Same brand, opposite questions, and you only find them by knowing who is asking.
Where does AI learn what your brand is for?
AI learns what you're for from the specific communities your buyers read, not from the category's default forum. Trace the sources behind the answers and you find the audience, not the industry.
When AI answered those Quince questions, 93% of the citations came from third-party sites rather than Quince's own pages. The most-cited blogs and editorial sites were not a big fashion hub. They were Apartstyle, a personal style blog, with 102 citations, then specialist sites like Closo, Dana Berez, The Mom Edit, and Wardrobe Oxygen, alongside sustainability guides like Ecocult and The Good Trade. Those are exactly the places a new mom or a sustainability-minded shopper goes for advice. AI reads them, so AI's picture of Quince is built from them. Win coverage in the niche your audience actually trusts and you teach AI to recommend you to that audience. Chase a generic category page and you are shouting in the wrong room.
What happens when you track keywords instead of customers?
Tracking keywords instead of customers hides you exactly where buyers are deciding. A keyword list only ever produces broad, generic phrasings, so you can look fine on those and still be invisible the moment a real person asks.
Start from real conversations instead of a keyword list and the competitive set widens too. We found 150 brands fighting for these answers, far more direct and indirect competitors than Quince would have guessed. Against that real field Quince came up in 7.3% of answers and ranked #7, behind Everlane at 21.1%, and across the 78 broad awareness questions a keyword approach would generate, AI named it in only 3. Those numbers are real, but they flatten the story. The interesting failure isn't the average. It's that Quince was missing from the postpartum questions, the capsule questions, the certified-organic questions, the exact high-intent moments the brand was built to win. A keyword check would have shown a respectable mid-pack score against a short list of competitors you already knew, and never flagged that the people most likely to buy couldn't find it.
How does Adacity find your real personas and their questions?
Adacity starts by finding who your buyers are, then matches them to where those people talk, so you never have to guess. You bring your website. We bring the audience and the questions.
We build from a corpus of real consumer conversations, then add your own site and the sources your category trusts, and we work out the distinct personas who buy from you. We match each one to the communities its real customers read, so a niche brand gets checked where its buyers talk, not just in the category's default forum. Then we phrase the questions the way each persona actually asks, run them across ChatGPT, Google Gemini, and Anthropic Claude with live web search on, and hand back the questions, the answers, and where each answer came from. A single Full Report runs 150 to 400 or more questions. The hard part was never asking the model. It was knowing who to ask for.
How is this different from picking your own prompts?
It's different because you never have to know your audience in advance, or guess their words. Most tools hand you an empty box and ask you to type in the prompts you want to track. That is keyword tracking, repackaged: it assumes you already know every persona, and the exact words each one uses, before you start.
Starting from real conversations finds the persona you didn't know you had. We dug into the question-versus-keyword side of this in what AI says about Quince; this piece is about the step before that, finding the people behind the questions in the first place.
See who's really asking AI about you
Your buyers aren't one audience typing one keyword. They're several different people, asking AI real questions to solve problems in their lives, in the places they already trust. Want to see who is asking about your business, what each one wants, and where AI gets its answers? Run the $10 check. You get your questions, your competitors, and what to fix, in your inbox within the hour.