Users talk in full sentences, add constraints, and treat the system like a staffer.
The old search model trained people to be brief. Type "Running shoes" and click around until something looks right. AI search flips that behaviour. Users talk in full sentences, add constraints, and treat the system like a staffer.
The tone of the prompts in the material is blunt about what people want. Persons give the AI their requirement priorities and have it do "all the necessary" to make recommendations. That is not browsing, that is delegation.
This is where "search" starts to look less like information retrieval and more like outsourced judgment. People are asking what to buy, when to buy, and how to defend the choice. They want a shortlist plus the reasoning, in the same breath.
How people ask for purchasing advice
A group of researchers built an database of questions users ask AIs by sampling posts and comments from five specific AI-focused subreddits (including communities like ChatGPTCoding, ChatGPTPromptGenius, and PromptEngineering)1, the prompts cluster around product decision support. Users are not asking for a definition of a category. They are asking to match a product to a life situation, a budget, a constraint, and a set of priorities, then justify the result.
In the product evaluation and recommendation bucket, the user hands over personal constraints. A person asks for sneakers similar to what they like, and with the purpose that they will be used. "Light running shoes for a marathon below 150$" The AI is being asked to act like a specialist sales associate who knows the inventory. Another example like "the perfect gift machine" prompt goes even further. It asks the AI to question the buyer about budget and interests, then calculate an "ideal" gift. That is a purchase decision framed as a guided interview.
In product selection and comparison, the prompts become more mechanical. People want the AI to cross-reference retail sites, pull prices, and compare multiple articles before they commit. One prompt asks for an AI that cross-references retail sites with eBay. Another asks the AI to extract raw facts from the internet such as dimensions, net weight, whether something is electronic, and whether batteries are included. That is an explicit request for structured product data. It reads like a shopping spreadsheet, except the user wants it built on demand.
Then there is quality assessment, where the AI is asked to ingest reviews and produce a verdict. One person describes asking the AI to research the net and analyze reviews to provide a "no shit take" on the product. That phrasing matters because it signals what the user thinks the AI is good for: cutting through marketing language and summarizing sentiment.
Finally, the prompts move from advice to action. The agency and automation requests are not subtle. They ask it to scrape new products and deals from e-commerce shops. One person sets a rule to check prices before quoting or estimating. Another wants to paste a shopping list into the AI, name a supermarket, and get a total cost. Another wants to upload pictures of a weekly flyer and have the AI plan meals using sale items. These are workflows. They require the system to pull live, exact values such as price, availability, and warranty terms, then do arithmetic and planning.
The common thread is that the AI is being treated as the interface to the market. Links are incidental. The output the user wants is a decision and a rationale, or a decision and a transaction.
This forses changes to stay vissbile
If users are asking the AI for shortlists and justifications, content that reads like a brochure is dead weight. It cannot support an answer that has to name tradeoffs and defend a recommendation.
The first requirement is intent coverage that is explicit. If people ask "best running shoes for sprint" or "lightest marathon shoes under 300$," the page has to contain sections that directly answer those prompts in plain language, with numbers attached. If the AI needs to weigh price versus warranty, your warranty terms cannot be buried in a PDF or hidden behind a contact form. If the AI needs to compare comfort versus performance, your content has to explain what "comfort" means in your category in a way that can be reused as a reason, not a vibe.
The second requirement is scannable comparison material. AI systems need to justify a pick. That means tables, labeled specs, and clear statements that separate claims from measurements. When the prompt asks for cross-referencing retail sites or extracting dimensions and net weight, a paragraph that says "compact and lightweight" is useless. A table that says "Shoe size: 42" and "Weight: 220g" is easy to extract and trust. When the prompt asks where warranties and maintenance plans "made sense," your coverage cannot stop at "3-year warranty." It has to define what is covered, what is excluded, and what the maintenance plan changes, in terms that can be compared.
The third requirement is readiness for agent behavior. If a user asks the AI to check prices before quoting, or to scrape deals, the AI needs clean, unambiguous data on your site. The material describes this as making the brand "API-able." The practical meaning is straightforward: price, availability, shipping, returns, and warranty need to be exposed in a way that a machine can pull without guessing. If those values are inconsistent across pages, or locked behind interactive elements that do not render cleanly, the AI will route around you and use someone else. The easiest and best way to enable this is to use Schema Markup
There is also changes now that users often perform "review or analysis this product" prompts. Most LLM's are trained to prefer earned media for this, then your site copy is not the main battleground for trust. The battleground is what third-party reviews say, how consistent they are, and whether the AI can summarize them into a clear verdict. If the AI is already being asked for a "no shit take," it will find the harsh parts as easily as the praise.
The switch from SEO to AEO, as described here, is not just a new checklist. It is a behavior change at the consumer level: people are moving from searching to assigning. The model becomes the buyer’s proxy, and your content becomes a data source and a stack of usable reasons.
