How to Find the Questions Shoppers Ask AI About Your Products

To find the questions shoppers ask AI assistants about your products, start from the shopper’s job, not your keywords: write down the use case, the comparison, the budget and the spec questions a buyer would type, phrase them the way people talk to a chatbot, and test each one in the countries you sell to. A list of 10 to 20 such questions per product line tells you more than any keyword tool, because it is exactly what an assistant is asked.

This guide shows how to build that list, how to check it, and what to do with the answers.

Why are AI questions different from search keywords?

A search keyword is short: “merino base layer”. A question to an assistant is a sentence with a situation in it: “what’s a warm base layer for skiing that doesn’t itch?”. The assistant answers that sentence by picking a few products and explaining why, so the words that decide whether you are picked are the situation words: skiing, doesn’t itch, warm.

Two things follow from that:

• You cannot see these questions in a keyword report. Assistants do not publish what they are asked.
• The questions you guess have to be tested, because an assistant may read your listing very differently from the way you wrote it.

Which kinds of questions should be on your list?

Most shopper questions about a product fall into five shapes. Cover each one for every product line you care about.

Shape	What the shopper wants	Example (running shoes)
Use case	A product for a situation	“best running shoes for flat feet on pavement”
Comparison	A choice between options	“trail or road running shoes for mixed terrain”
Budget	The best option under a price	“good running shoes under 100 dollars”
Spec	A product with one property	“lightweight running shoes with a wide toe box”
Audience	A product for a kind of person	“running shoes for a beginner training for a 10k”

Leave out questions with your brand name in them. You will almost always win those, and they tell you nothing about whether an assistant would pick you over someone else.

Where do good questions come from?

You already hold most of the raw material:

1. Your support inbox and chat. Questions customers asked before buying are the questions they now ask an assistant. Copy their wording, not your summary of it.
2. Your product reviews. Look for the situation in each review: “bought these for my first marathon” is a use case question waiting to be written.
3. Your returns. “Too narrow”, “not warm enough” or “smaller than expected” each point to a spec question your listing should answer.
4. Your category fields. Every attribute Shopify defines for your category (material, fit, size, pattern) is a filter a shopper can ask for.
5. Your listing, read cold. Read it as someone who has never heard of you. Whatever they would still need to ask is a question.

Write each one as a full sentence, in the shopper’s language, the way they would say it out loud.

Why test questions per country?

The same question gets different answers in different markets. Shopify’s AI catalog answers per buyer country and shipping destination, and shoppers phrase things differently from one market to the next: a Dutch shopper asks about a “thermoshirt”, not a “base layer”. If you sell in several countries, write the question in each market’s main language and check it there.

How do you test whether an assistant recommends you?

You can do it by hand: open ChatGPT and Gemini, ask the question, and note whether your store is named, whether it is recommended or just mentioned, and who is picked instead. Do it from the country you care about, and repeat it later, because answers change between runs.

Doing that for 20 questions in three countries every week is a lot of copying. This is the job we built the AI rankings page in Octolift for. Each question is run as a live search of Shopify’s AI catalog for that country, where your rank in the top 10 is repeatable, and asked of ChatGPT and Gemini at the same time. Generate questions writes a first draft of the list from a product’s title, type and category, in the country’s main language, and every question you ask stays in your list so you can check it again. See AI rankings for how it works.

What do the answers tell you?

Sort each question into one of three results:

• Recommended. Keep it on the list and watch it. Note what the assistant said about you; that is what your listing is communicating.
• Mentioned, but someone else recommended. The assistant knows you but found a better match. Compare your listing with the product it picked: which of the shopper’s words does theirs answer and yours does not?
• Not mentioned. Either the assistant cannot find the product or it does not connect it with the question. Start with whether the product is in Shopify’s catalog at all, then with the specifics in your listing.

Across the whole list, look at who keeps winning. A brand that is recommended in 6 of your 10 questions is the listing to study.

How do you turn a lost question into a better listing?

Take the words in the shopper’s question that your listing does not answer, and answer them with facts you can back up: material, dimensions, fit, who it is for, what it is not for. Put the structured ones in your category fields, where an assistant can filter on them, and the rest in your description. Then check the question again after Shopify has picked up the change.

Do not add claims your product cannot support to win a question. An assistant that quotes a wrong spec costs you a return, and AI rewrites are an easy place for invented facts to slip in; our checklist for AI written product descriptions covers how to catch them.

Summary

Build your question list from real shopper situations in five shapes (use case, comparison, budget, spec, audience), write it in each market’s language, and test it in ChatGPT, Gemini and Shopify’s catalog per country. The questions you lose show you exactly which facts your listing is missing. You can check your first five questions for free with Octolift on the Shopify App Store.