- Assistants combine what the model already knows with pages retrieved live from search indexes and their own crawlers.
- They name products they can justify against the shopper's constraints, using facts from product pages and confirmation from independent sources.
- Every engine draws on a different mix of sources, so visibility differs by engine and must be measured separately.
Ask ChatGPT for “a lightweight rain jacket for city commuting under $200” and you get a short list of named products with reasons. Ask Perplexity the same question and the list may be different. Neither answer is an advertisement, and neither is random. Both are the result of a retrieval and reasoning process that brands can understand — and influence.
Step 1: The assistant decides whether to search
For shopping questions, assistants usually search the web rather than rely only on what the model learned in training, because prices, stock and new products change. ChatGPT search, Perplexity, Gemini and Claude can all retrieve live pages. What they retrieve comes from search indexes and from their own crawlers — OpenAI's OAI-SearchBot, PerplexityBot, Google's index, and others.
Implication: if those crawlers cannot reach or read your pages, the assistant has nothing current from you to work with, and falls back on what other sites say about you.
Step 2: It turns the question into constraints
A shopping prompt is a bundle of requirements: product type (rain jacket), use (city commuting), attribute (lightweight), budget (under $200), sometimes identity or values (sustainable, petite, sensitive skin). The assistant looks for products that satisfy all of them and for evidence that they do.
Implication: a product page that states weight, waterproof rating, packability and price gives the assistant evidence for each constraint. A page that says “your new go-to layer” gives it none.
Step 3: It retrieves and weighs sources
The assistant reads a handful of pages for each answer. Studies of AI citations consistently show a few patterns:
- Lists and comparisons are cited heavily. In one 2026 analysis, “Best X” listicles accounted for 43.8% of the page types ChatGPT cited, followed by comparison guides and product documentation.
- Retailers dominate in some categories. Similarweb found that 54.7% of sources cited in ChatGPT beauty conversations were retail and ecommerce domains.
- Communities and video matter. Reddit, YouTube and Wikipedia are among the most cited domains across AI engines overall.
- Engines disagree. An analysis of hundreds of millions of citations found only about 11% overlap between the domains ChatGPT and Perplexity cite.
Implication: your own site is necessary but rarely sufficient. Assistants look for agreement between your product facts and independent sources.
Step 4: It looks for consensus and trust
When several independent sources name the same product for the same need — an editorial roundup, a retailer with strong reviews, a community thread — the assistant can recommend it with confidence. When sources conflict, or when a brand's claims cannot be found anywhere else, the assistant hedges or leaves the product out. This is especially visible in beauty, where assistants are careful with efficacy and safety claims.
Step 5: It writes the answer — and the reasons
Finally the model composes a short list, usually three to five products, each with a one-line reason drawn from the sources: “breathable 2.5-layer shell, packs into its own pocket, around $160”. The reasons are often paraphrased directly from product pages and reviews. Products without quotable facts rarely make the list, because the model has nothing to say about them.
What this means for fashion and beauty brands
- Be retrievable. Let AI search crawlers reach your pages and serve the content without JavaScript.
- Be quotable. State the facts that match shopper constraints on every product page.
- Be confirmed. Earn mentions in the roundups, retailers and communities each engine actually cites for your questions.
- Be measured per engine. Track the same questions in ChatGPT, Gemini, Claude and Perplexity; they will not move together.
Get your products named when shoppers ask AI
Sayrank tracks how ChatGPT, Claude, Gemini and Perplexity answer your shoppers' questions, fixes what stops AI crawlers on your site, and rewrites your product pages so assistants can quote them. Built for fashion and beauty brands.
Frequently asked questions
Can brands pay to be recommended in ChatGPT answers?
The organic recommendations in assistant answers are not sold. Some platforms have introduced or are testing advertising formats, which are labelled separately from the answer itself. Being named in the answer depends on what the assistant finds and trusts.
Why does ChatGPT recommend different brands than Perplexity?
Each assistant retrieves from different sources and weighs them differently. One analysis of hundreds of millions of AI citations found only about 11% overlap between the domains ChatGPT and Perplexity cite. Optimizing for one engine does not guarantee visibility in another.
Do AI assistants use reviews?
Yes, indirectly and directly. Review content on retailer pages, review sites and communities is a frequent source, and the overall sentiment it expresses shapes how an assistant describes and ranks a product.
Why does the answer change every time I ask?
Assistants generate answers probabilistically and may retrieve slightly different sources each time. That is why visibility should be measured over many runs and weeks, not from a single answer.