Should ChatGPT Visibility Reports Separate Search, Non-Search and Shopping Answers?

What are the modes, and how do you tell them apart?

A single answer can mix modes. A search answer may include a product carousel, so record each element: whether search ran, whether citations appear, whether product cards appear, and whether an ad was shown.

Why does mixing them mislead?

Suppose your brand appears in 60% of search answers and 20% of non-search answers for your question set. If 40% of answers used search this month, your blended appearance rate is 36%. Next month, ChatGPT searches for 60% of the same questions, and the blended rate rises to 44%, without any change in how often you appear in either mode. The "improvement" is entirely a change in ChatGPT's behaviour. It can go the other way too, making a real gain look like a loss.

Similarweb estimated that citations appeared in about 6.8% of US ChatGPT prompts by May 2026, up from about 1.6% a year earlier. That was across all prompts, not just commercial ones, but it shows how fast the mix can change. When and why ChatGPT decides to search is covered in how ChatGPT decides when to use its search tools.

What does each mode respond to?

• Non-search answers reflect what the model learned. They change slowly, mostly when a new model is released. The levers are long-term: being described consistently and widely across the web, in the kinds of sources that end up in training data. Measure them, but do not expect monthly movement.
• Search answers respond to crawl access for OpenAI's search crawler, pages that answer the question clearly, rankings in the search sources ChatGPT draws on, and third-party coverage. Citations here can be traced to specific pages.
• Shopping results respond to product data. ChatGPT's carousel largely reflects Google Shopping: one study found 83% of carousel products came from Google Shopping results, so a complete Google Merchant Center feed matters, alongside a feed submitted through OpenAI's merchant programme, complete product pages, prices and availability that match, and presence on retailers and marketplaces. One 2026 analysis of 8,520 ChatGPT shopping queries found retail sites made up 89% to 99% of citations across buying stages, with brand sites barely registering. For many brands, retailer listings matter more than their own site here.
• Shopping research responds to detailed, accurate product information across sources, because the mode checks products against several constraints at once. Merchants can apply through OpenAI to be included.
• Ads respond to ad spend. OpenAI says ads do not influence answers, and they are labelled and kept separate, so report them as paid media.

How can marketers separate ChatGPT Shopping Research from normal ChatGPT answers?

Shopping research is a distinct mode, not just a different kind of question. OpenAI launched it in November 2025 for logged-in users on the Free, Go, Plus and Pro plans. The user describes what they need, and ChatGPT asks clarifying questions about budget and preferences. It then researches for a few minutes and returns a personalized buyer's guide. Users can steer it by marking products "Not interested" or "More like this". It runs on a version of GPT-5 mini trained specifically for shopping. OpenAI's internal evaluation found it met users' requirements on multi-constraint questions 52% of the time, against 37% for ChatGPT search.

To separate it in your tracking:

1. Run it as its own test. Where your ChatGPT app offers shopping research, start it deliberately for a set of buyer scenarios, rather than hoping a normal prompt triggers it. Record the scenario, the clarifying answers you gave and the final guide.
2. Script the clarifying answers. Personalization is the point of the mode, so fix your answers to its questions for each scenario, or results cannot be compared.
3. Record the guide's structure, not just mentions: which products are top picks, which are alternatives, what reasons are given and which sources are cited.
4. Keep it out of your normal prompt metrics. A guide that takes minutes to produce and follows a conversation is not comparable with a single-turn answer.
5. Run fewer, deeper tests. Each run takes time, so a handful of well-chosen scenarios per product category beats a large prompt list.

For normal answers, record whether a product carousel appeared and which products and merchants it showed, separately from the text of the answer.

How should the report be laid out?

Report each mode as its own section, with its own measures:

• Non-search: appearance and recommendation rate from model memory, by model.
• Search: appearance rate, recommendation rate, citation rate and which of your pages are cited.
• Shopping results: how often your products appear in carousels, at what price and from which merchants.
• Shopping research: top-pick and alternative placements per scenario, with the reasons given.
• Ads, if you run them: impressions and clicks, reported as paid.
• Mode mix: the share of your questions that triggered search or shopping, so readers can see when a change is in ChatGPT, not you.

If you need one headline number, show it next to the mode mix so readers can tell what moved. Differences between tracking through the API and the app are covered in why ChatGPT API answers differ from the web app when tracking citations.

What might an illustrative case look like?

Consider an illustrative kitchen appliance brand whose monthly ChatGPT report showed its visibility falling from 41% to 33%. Splitting answers by mode showed the brand's rate in search answers was steady at 55%, and in non-search answers steady at 25%. What had changed was the mix: fewer of its questions triggered search that month. Meanwhile, product carousels showed the brand's blenders mostly through one large retailer, at a price above the brand's own site, which the blended report had never shown. The team kept the split report, fixed the retailer's price data through its account manager, and started a separate quarterly shopping research test for three buyer scenarios.

What do people ask about ChatGPT visibility reporting?

Can I make ChatGPT search when testing?

You can ask it to search the web, or choose search in the app. For measurement, mostly test without forcing search, because forcing it hides how often ChatGPT chooses to search on its own, which is the mode mix your report needs.

Do shopping carousels count as recommendations?

Count them separately. Appearing in a carousel is visibility, but it is not the same as being recommended in the answer text.

Should ads be included in visibility reports?

Only as paid media, clearly separated. Mixing them into organic figures overstates what the answers themselves say.