If you sell products, you have probably typed a shopping question into ChatGPT or Copilot and seen a row of product cards come back with pictures, prices and merchants. The next thing most people wonder is whether their own product shows up in those cards, and what price and merchant sit next to it. This guide walks you through a simple routine for AI shopping visibility: pick the products, write a fixed set of prompts, record what appears, open the product details, and repeat on a schedule. By the end you will have a worksheet you can keep running with a small team.
There is no official benchmark for how many prompts you should run, how often, or what a good appearance rate looks like. The numbers in this guide are practical starting points, so treat them as a place to begin and adjust them to the time you actually have.
You will need access to ChatGPT and Copilot, a spreadsheet, and a list of the products you care about.
Step 1: Pick the products and decide what you are measuring
Start with a short list of products or product families, not your whole catalog. For each one, write down the brand, the product name, the model or SKU if you have one, and the variant that matters (size, color, material). Then write down the question you want answered. It might be "does our exact model appear when someone describes this use case?" or "does our brand show up at all for this category?" or "which merchants are listed when our product does appear?"
Those are three different questions, and it helps to keep them apart. If you blend them into one score, you won't be able to tell what moved when the numbers change.
You are done with this step when every product on the list has a stable identity, so anyone on your team can tell your exact product from a similar one.
A common mistake here is counting a close match as your product because the title sounds right. A different model, a different size or a similar name from the same brand is not the same item. Write down the exact title you see, and mark the identity as uncertain so someone can check it later.
Step 2: Build a fixed prompt set
Write prompts the way a shopper would, and put in the details that matter to the buyer. A good starting mix covers a few kinds of request:
- A category and use case, such as "Find a lightweight carry-on for frequent weekend trips."
- Specific attributes, such as "waterproof," "wide fit," or a named material.
- A budget or price ceiling, if price is part of how people decide.
- A request that names your brand or model.
- A category request that does not name any brand, which shows whether your product comes up without being asked for.
These are examples of how to write prompts, and they say nothing about what either platform will return. For a first pass, ten to fifteen prompts across your priority products is plenty. Give each prompt an ID and save the exact text, the same way you would for a repeatable prompt set in any AI visibility routine.
Keep the branded and unbranded prompts in separate groups. A branded prompt tells you whether the platform can find a product it was told about, while an unbranded one tells you whether the product surfaces when the shopper only describes a need. They answer different questions, so a single blended number would hide both.
You are done when each prompt has a clear purpose, a rule for what counts as a match, and wording you will not change. If you do need to change a prompt later, save it as a new version instead of editing the old one quietly. Otherwise a change in the results could be a change in the question, and you would never know.
Step 3: Standardize the run conditions
Shopping results depend on who is asking and from where, so write down the conditions for every run. For each platform, record the date and time, the time zone, the country or market, the language, the currency, the device, whether you used the app or the web, the account type if you know it, and whether the shopping interface appeared at all.
For ChatGPT shopping tracking, check Settings > Personalization > Memory when you set up your routine. OpenAI says ChatGPT considers the query and its context when it picks products, including Memory and Custom Instructions. Starting a new chat does not remove that account-level context. So note whether the account is in its normal personalization state. If your team decides to change or turn off those settings for a cleaner audit, write down the change and keep it the same for every later run.
For Copilot shopping visibility, decide whether you will ask a product question in a conversation or open the Shopping homepage. Microsoft documents the route as the left side pane, then More, then Shopping. The homepage is personalized, and Microsoft says shopping and chatting with Copilot can make recommendations more relevant over time. A prompt typed into a conversation and the Shopping homepage are different surfaces (there is a separate guide on how to track Copilot visibility in general), so pick one route and stay on it, and don't compare them as if they were the same thing.
Availability also varies. Microsoft says Copilot features differ by market, language, device, platform and subscription, and the support article applies to personal Microsoft accounts. If a feature is not available to you, log that plainly, and do not quietly switch to another surface.
You are done when someone else on your team could repeat the same run with the same platform, route, account setup, market and prompt.
Step 4: Run each prompt and record the visible products
For ChatGPT shopping tracking and Copilot shopping visibility alike, run every prompt on its own in each tool, starting a new conversation each time. First note whether a shopping panel or a set of product results appeared at all. If it did, copy down the products in the order they are shown, using the exact title on the card and the brand or model if it is visible. Count how many results you can see, and save a dated screenshot or other evidence, following whatever your company's rules are for storing that kind of material.
It helps to keep a few things separate while you do this:
- Shopping product cards or a shopping panel, as opposed to ordinary text recommendations.
- Items you can see right away, as opposed to items that only show up after you expand or open a product.
- Regular results, as opposed to anything clearly labelled sponsored.
- A prompt that returned no shopping panel, as opposed to a panel that appeared without your product in it.
That last pair matters more than it looks. "No panel" and "panel without us" lead to different next steps, so they should never share a cell.
If your product only appears in a sentence of the written answer and not in a card, don't count it as a shopping-panel appearance. It may still be worth knowing about, but it belongs to tracking ordinary recommendation-style answers, which is a separate job. The same by-hand habit works there too, as in checking your ChatGPT visibility manually.
On ads, OpenAI says its product results are chosen independently and are not ads, and that ads are separate from product results. That does not mean you will never see an ad in ChatGPT, so if one shows up, record it on its own line. Copilot's support page says sponsored links are clearly identified, so do the same there.
One practical warning on how to collect this. OpenAI's Terms of Use prohibit automatically or programmatically extracting data or output from its services, and they prohibit getting around rate limits or protective measures. So this routine is done by hand, with a person reading the screen and typing into the sheet. It is slower, but it is the safer way to do it.
You are done when every prompt and platform pairing has either a dated result or a clear "no panel" or "no result" status, with the order of the products kept.
Step 5: Open each product and capture merchants, attributes and prices
The card is only the first layer. For each product that matches yours, open the detail view and record what you find there as a separate set of notes. Write down every merchant or offer shown, the order they appear in, the price and currency, and any attributes on the page. Depending on the platform and the merchant data, that might include size, model or variant, a review summary, a return window, availability, price history, shipping terms, a label, and how the buy button works. When something isn't there, write "not shown" or "not available" and move on, because filling gaps with a guess will cost you later.
In ChatGPT, the price on the first card usually reflects the first merchant listed, and OpenAI says it may not be the lowest. The detail view can show more merchants with different prices, and the merchant ranking can reflect availability, price, quality, and whether the merchant is the maker or main seller. You may also see labels such as "Best price" or "Instant checkout." Copy the label exactly as it appears. A lowest-price offer can be missing the "Best price" label if another offer carries something like "Instant checkout," so don't work out the meaning yourself. Also keep in mind that a label like "Budget-friendly" on a product image is generated by ChatGPT. It is not a guarantee and doesn't necessarily point to the lowest price.
Titles and descriptions can be simplified by ChatGPT using information from third-party providers, and review summaries come from public reviews that OpenAI says it does not verify. That is why you keep the exact card text in one column and your own product match in another.
In Copilot, record the retailers and any price history or return details that appear. Microsoft says what shows up depends on what the merchant provides, and that retailers set their own prices and availability and can change them without notice. If your audit needs a real purchase price, check it on the retailer's page and put that number in a separate field. It is not the same as the price shown in the AI result.
Shoppers often finish the purchase on the retailer's site anyway, so the verification gap in product content is worth reading about. Before you compare any prices, make sure you are comparing the same thing. That means the same exact model and variant, the same market, and the same treatment of shipping and fees. If those don't line up, mark the price as non-comparable instead of averaging numbers that don't belong together.
You are done when the worksheet keeps both what the card showed and what the detail view showed, and every price has a source and a time next to it.
Step 6: Log everything in one structured sheet
A spreadsheet is enough for a manual baseline. Use one row for each combination of prompt, platform, run, product and merchant. Add a run-level row for prompts that produced no panel or no matching product, so those don't disappear from your counts. A layout like this works as a starting point:
| Field group | Suggested fields |
|---|---|
| Run identity | Run ID, date, time and time zone, who ran it, platform, route, market, currency, device, account type |
| Prompt | Prompt ID and version, exact text, category, use case, branded or unbranded |
| Result | Panel shown, matching product shown, order shown, exact title, your canonical product match, sponsored label, screenshot reference |
| Merchant and offer | Merchant name, merchant order, price and currency, price label, availability, shipping or fees if shown, retailer check price |
| Product details | Attributes, size or variant, review summary, return window, price history, buy route, other labels |
| Quality notes | How sure you are of the identity, missing fields, odd results, interface changes, any deviation from the plan |
From that sheet you can work out a few simple measures. These are team-defined measures, not official platform metrics, so write down how you calculated each one.
- Product appearance rate: the runs where your defined product match appeared, divided by the eligible runs for that prompt and platform. Say how you treated the no-panel runs in the denominator.
- Brand share of visible product slots: the number of listings that match your brand, divided by all product listings recorded in the same sample. Decide whether a repeat appearance counts once per run or once per slot, and stick with it.
- Merchant presence rate: the product detail views that showed a particular merchant, divided by all the detail views you opened for that product. Don't work it out from the first card alone, since the merchant list may only appear after you open it.
- Observed price range: the lowest and highest price among comparable, timestamped offers for the same exact variant in the same market. It describes your sample, and it is not the full market or a promised checkout price.
- Position: the order the product was displayed in. It is an observation, not an official ranking score, and if you average it, remember that the result sets may be different sizes.
Please don't invent a weighted ranking formula, and don't treat a missing field as zero. "Not shown," "no shopping panel," "no matching product" and "not checked" mean four different things.
Pro tip: put a status column next to every measure, with those four labels as the only allowed values. It takes a few seconds per row and it keeps the blanks from turning into false zeros when someone builds a chart later.
You are done when a stranger could open the sheet and tell, for any single number, which dated runs it came from.
Step 7: Repeat on a schedule and look into the changes
For AI shopping visibility, a routine you can keep up beats an ambitious one you drop after a month. A reasonable starting point is a weekly run of your priority prompts and a monthly run of the full set, with a few independent runs of the most important prompts when you have the time. These are suggestions and not official standards or proven sample sizes. If you change a prompt, the platform, or your method, record it as a new measurement version and start a new baseline.
When you compare two runs, compare like with like: the same prompt version, the same surface, the same market, the same account conditions, the same product-match rule and the same way of observing. Then look for what moved. That could be whether the product appeared, its position, which merchants were listed, the attributes, the displayed price, a sponsored label, or whether the panel showed up at all. If a price or availability looks strange, check the retailer and keep the original AI result on file as it was.
A common mistake is reading one result as a stable ranking. ChatGPT says it chooses products based on what it reads as relevant to the shopper's intent, and that not every available product will necessarily be shown. Copilot says its options can reflect likely engagement and a shopper's own behavior. So a change can come from the prompt, the account, the market, the device, the time, or the product data, and not only from anything you did. Day-to-day noise in AI answers is normal, and it is one more reason to repeat runs. Write up what you find as "observed in these runs," and don't describe a single observation as a trend across the platform.
You are done when you can explain your comparison window, name the conditions that changed, and trace any reported change back to individual dated runs.
Step 8: Add scheduled tracking for the wider picture
A shopping-panel audit tells you about product cards. It doesn't tell you how your brand shows up in the other answers shoppers see, such as a comparison question typed in plain language or a "what should I look for" question. For that wider view, If you want to see how the wider tools compare, a roundup of AI visibility tools is a good place to start. DeepSmith's AI Visibility lets you define the buyer questions you care about, checks AI engines for them on a schedule, keeps the full answers, and reports mentions and citations separately. You can see mention rate and citation rate per prompt, and look at which competitor pages are getting cited.
To be clear about the limits, DeepSmith does not capture shopping panels, product cards, SKUs, merchant offers, prices or shopping screenshots, so it does not replace the worksheet above. Keep doing the manual routine for the product listings, and use scheduled tracking for the question-and-answer side. If you already track a set of buyer prompts there, you can reuse the wording of those prompts as the start of your shopping prompt set.
If you also care about Google's own AI answers, tools that track Google AI Overviews cover that side. Also keep in mind that a product card in a shopping panel is not the same thing as being cited as a source in an ordinary AI answer. Bing's AI Performance documentation describes citations as visible references and says its data is aggregated and sampled, so it works for trends and not for exact counts. Those numbers are not a measure of shopping-card impressions in ChatGPT or Copilot.
You are done when you can say which questions your manual worksheet covers and which ones your scheduled tracking covers, with no overlap you are counting twice.
What to do next
Pick three products, write ten prompts, and run them once in each platform this week, filling in the sheet as you go. After that first pass you will know how long a run really takes and which fields you keep leaving empty, and you can trim the worksheet to what your team will actually maintain. Once the audit shows where you are missing, a guide on how to get your products cited in ChatGPT shopping answers is the natural follow-up. When you want the broader AI answers tracked on a schedule too, you can start a free DeepSmith trial and set up your buyer prompts there.
Related guides
- How to Track Many Products Across AI Assistants and Report Up: Organize prompt sets per product, track each across the major assistants, and roll results into one leadership view that shows which products gain or lose presence without hiding.
- How to Track Your Brand in Google AI Mode: Track your brand in AI Mode: build a prompt list from real conversations, run repeat samples, record mentions and cited URLs, and log results so changes over time.
- How to Run a 14-Day Proof of Concept With an AI Visibility Tool: Run a structured two-week trial with a fixed prompt set, success criteria and a day-by-day plan so you can compare tools on equal terms.
- AI Visibility Tool Red Flags: Claims to Question Before You Buy: Spot vendor claims that deserve a follow-up question, from vague engine coverage to unexplained scores and guaranteed outcomes.
- AI Visibility Tracking vs Traditional Rank Tracking: What Changes: Learn which assumptions from classic rank tracking break in AI answers, such as fixed positions and stable results, and which metrics replace them.


