Can I Test 20 Real Buyer Questions Across ChatGPT, Perplexity, Gemini, Google AI Overviews and Claude for a CPG Brand?

Why do CPG brands need their own question set?

Shoppers rarely ask an assistant about a brand by name. They ask about a need: a protein bar without sugar alcohols, a dishwasher tablet that works in hard water, a sunscreen that does not sting sensitive eyes. The answer names a handful of products, and the products it names come from what retailers, reviewers and publishers say about them. A question set built from your brand name or your marketing categories would miss exactly the moments where a competitor wins.

CPG answers also lean heavily on pages you do not control: retailer listings with ingredients and ratings, review roundups, and product databases. Testing shows which of those the engines trust in your category, which tells you where to work.

How do you pick the 20 questions?

Spread them across the ways shoppers decide, and keep brand names out of all but a couple:

Find the wording in what shoppers already write: customer service emails, retailer reviews and Q&A, search queries in Search Console, and the questions sales teams hear from buyers. Then rewrite each one as a person would type or say it to an assistant, with the constraint that makes it specific. More on sourcing wording in how to choose the right prompts to track for AI visibility.

How do you run them across five engines?

By hand, the method is simple but slow:

1. Fix the conditions. Same location setting, signed out or in a clean profile, no earlier conversation in the thread.
2. Run each question three times per engine. Answers vary from run to run; three runs show whether a mention is steady or a one-off.
3. Record every answer. Paste the full text and the source links, not just a yes or no.
4. Repeat on a schedule. Monthly is enough for most brands; weekly in a launch or a contested season.

Twenty questions, five engines and three runs make 300 answers, which takes a person most of a working week to collect and log. That is why most teams either cut the set or use a tool. The full manual method is in how to manually track AI visibility by testing prompts across AI engines.

What should you record for each answer?

• Named or not. Whether your brand appears at all.
• Recommended or just listed. Whether the answer presents your product as an option to buy, or only mentions it.
• Who else. Every competing brand named, in order.
• Cited sources. Every link, and what kind of page it is: retailer listing, review roundup, brand site, publication, forum or product database.
• What it said about you. Ingredients, claims, prices and availability, checked against your own product data.

The cited sources usually explain the result. If every answer about gluten-free crackers cites the same two roundups and a retailer's category page, those three pages are where the next month's work belongs.

What can 20 questions tell you, and what can't they?

Twenty questions are a small, sharp instrument. They are enough to:

• find the needs and constraints where you are absent while competitors appear;
• see which engines treat you differently;
• identify the handful of pages the answers rely on;
• spot wrong ingredients, claims or prices being repeated.

They are not enough to estimate your share of all AI shopping answers in a category, to track dozens of product lines at once, or to detect small changes reliably. With 20 questions, a shift of a couple of mentions can be chance, so judge change across several readings and look at which questions moved, not only the total. For brands with many lines, run a separate 20 for each important line rather than one diluted set.

Can a tool do this for a CPG brand?

Several trackers will run a fixed question set across engines on a schedule; they differ in which engines they cover, how they collect answers and whether they show the cited pages. Retail-focused tools also track product cards inside ChatGPT's shopping results, which a question-based test does not cover.

Kovmere, our product, runs this exact design. A $19 Check asks up to 20 buyer questions for one market, up to 18 that do not name your brand plus up to 2 brand comparison questions kept apart from the main figures, three times each across OpenAI, Perplexity, Google Gemini, Anthropic Claude and Google AI Overviews. That is up to 300 answers; the report shows the answers themselves, counts every cited URL and labels the most-cited pages by type and by whether they mention you. The report separates mentioned, recommended and cited, and ends with up to three priority fixes, such as a product-page draft or a source worth approaching, which you review and publish yourself. No account is needed, and Pro at $49 a month re-asks the same frozen questions each week of the billing period.

Two limits matter for CPG brands. Kovmere samples the engines through their search-enabled APIs and Google's AI Overview results, so it does not show the consumer ChatGPT app or its shopping cards. And the questions are written for your market rather than taken from logs of what real shoppers typed: Kovmere has no access to real users' conversations, and the panel-based question data some trackers sell covers only a sample of opted-in users.

What do people ask about testing buyer questions for a CPG brand?

Should the questions include my brand name?

Mostly no. Keep one or two head-to-head questions with a main rival, and report them separately; the unbranded questions show whether shoppers who do not know you yet will hear about you.

Do I need to test every retailer's own assistant too?

Not at first. Start with the general engines shoppers use to research, then add retailer assistants such as Amazon's or Walmart's if most of your sales run through them, because they answer from their own catalogue.

How often should a CPG brand re-test?

Monthly for most categories, and weekly around launches, reformulations and seasonal peaks, using the same questions so the readings compare.

What if the answers get my ingredients wrong?

Correct the sources the answers cite, starting with retailer listings and product databases, and make the current ingredients easy to find in plain text on your own product pages.