How to Measure AI Visibility for Your Products
You cannot rank-track an AI answer. You sample it. A fixed prompt panel, run weekly across engines and scored for mentions and citations, is the method, and here is how to do it with tools or by hand.
- By
- VendaHawk
- Published
- Updated
- Reading time
- 7 min read
Measure AI visibility with a prompt panel: a fixed list of 30 to 150 shopper questions, run on a schedule across ChatGPT, Google AI Mode, Perplexity and Copilot, with each answer scored for whether your brand is mentioned, where it sits in the list, and which page is cited. Report share of voice against a named competitor set. Google already computes exactly this inside Merchant Center; for every other engine you sample it yourself or pay a tool to.
Why rank tracking does not work here
AI answers are generated, not ranked. The same prompt returns different brands on different days, and the pages an engine cites have little to do with Google's top ten: only about 12% of ChatGPT, Gemini and Copilot citations rank in Google's top ten, and ChatGPT cites pages from position 21 or lower about 90% of the time (Ahrefs overlap, Semrush study). Being cited is not the same as being recommended either: in one Ahrefs experiment, 43% of answers that cited a promotional page never mentioned the brand (Ahrefs).
So the unit of measurement is the prompt, and the metrics are rates across many samples.
Build the prompt panel
Write 30 to 150 prompts per brand, grouped by buying stage:
- Discovery: "best cordless stick vacuum for pet hair", "what should I look for in a toddler car seat"
- Evaluation: "X vs Y", "is X worth it", "X problems"
- Ready to buy: "where to buy X", "X under a set budget", "X with free returns"
Use four phrasing styles for the same intent (question, command, long conversational, short keyword) and make about 20% of the panel branded or competitor prompts. Localise by country and language. Source the questions from your Amazon Q&A, review complaints, and the top terms Merchant Center reports for AI surfaces.
Choose the engines
Sample ChatGPT, Google AI Mode and AI Overviews, Perplexity and Copilot for every brand. Add Amazon's Alexa for Shopping and Walmart's Sparky for marketplace sellers, since those assistants read the listing rather than the web and need their own panel. Run each prompt logged out, in a fresh session, so personalisation does not pollute the sample.
Score each answer
For every prompt and engine, record: brand mentioned (yes or no), position in any list, product or page cited (URL and domain), sentiment, and the competitors named. Then compute:
- Share of voice = your mentions divided by mentions of you plus the competitor set
- Citation share = answers citing one of your pages divided by all answers
- Top cited domains per prompt cluster, so you know which third-party pages to target
Expect variance. Semrush's 2026 AI Visibility Index measures around 15 sources per ChatGPT response, so sample each prompt at least five times before you trust a trend. Store the raw answers, not just the scores; you will want to reread them when a competitor appears.
Tools and what they cost
| Tool | Entry price (vendor pages, Oct 2026) | What you get |
|---|---|---|
| Google Merchant Center AI performance | Free | Share of voice and missing attributes for AI Mode and AI Overviews; English queries; US, CA, AU, IN, NZ (Google) |
| Google Search Console | Free | AI Overview and AI Mode clicks, inside the Web totals |
| Otterly.AI | $29/mo Lite (15 prompts, 4 engines); $189 Standard (100 prompts); $489 Premium (400) | Prompt tracking plus GEO audits (Otterly) |
| Ahrefs Brand Radar | Custom prompts from $50/mo as an add-on; AI Visibility Index from $199/mo (83 prompts a day); $699/mo for all models | AIO, AI Mode, Copilot, Gemini, Perplexity, ChatGPT, Claude (Ahrefs) |
| Semrush AI Visibility Toolkit | $99/mo per domain on annual billing; 25 prompts; 50 more for $60/mo | ChatGPT, Google AI, Gemini, Perplexity (Semrush) |
| Peec AI | About $95/mo Starter (50 prompts, 3 models), $245 Pro, $495 Advanced, per third-party listings | Daily tracking (Peec) |
| Profound | No public price; free trial of 50 prompts for 7 days; enterprise custom | Up to 9 engines including ChatGPT Shopping (Profound) |
Start with the two free sources. If you sell through Merchant Center, its AI performance report is the only first-party share-of-voice number any engine publishes, and its "popular attributes missing from your product data" list is a ready-made fix list.
The DIY route
If you would rather own the data, the stack is small: the OpenAI Responses API with web search for ChatGPT, the Perplexity Sonar API, the Gemini API with Google Search grounding, and a headless browser for AI Mode and Copilot. Store the raw answer and its citations, classify mentions with a small model, and keep a human spot check on a sample. One caveat: API answers differ from what the consumer app shows, so note which you sampled and do not mix them in a trend line.
Close the loop with referral data. Add chatgpt.com, perplexity.ai, copilot.microsoft.com and gemini.google.com as a channel group in GA4, and on Shopify use the thank-you page survey. Shopify's checkout UI extensions render on the Thank-you page and Order-status page on Basic plans and above, and Additional Scripts were retired on 26 August 2026, so the extension route is the supported one (Shopify). Ask "How did you first hear about us?" within two hours of purchase and put "An AI assistant (ChatGPT, Gemini, Perplexity, Copilot, Amazon Rufus)" on the first screen with a free-text follow-up. If you buy rather than build, Fairing is free up to 200 orders a month and Zigpoll's Lite tier is free for 50 responses a month, per their Shopify App Store listings.
A weekly routine you can keep
- Monday: run the full panel on every engine, five samples per prompt. Store answers and citations.
- Tuesday: score mentions, positions and cited domains. Update share of voice per cluster and per engine.
- Wednesday: pull Merchant Center AI performance and the missing-attributes list; pull GA4 AI referrals and the survey counts.
- Thursday: list the pages that were cited instead of you. Classify them as editorial list, YouTube, Reddit, review platform, retailer or brand. Those are next month's outreach targets.
- Friday: pick three fixes: an attribute to add, a FAQ to write from a new prompt, a page older than 12 months to refresh. Log them with the prompt that prompted them.
Monthly, add new prompts from Q&A and reviews, retire prompts no shopper asks, and re-check the engine mix. Citation sources move: Reddit went from a large share of ChatGPT citations in 2025 to under 1% of ChatGPT Search citations by August 2026 while staying strong on Google AI Mode and Perplexity (Otterly), so a panel that only watches one engine will mislead you.
Frequently asked questions
How many prompts do I really need?
Enough to cover your buying stages and hero products without being noisy: 30 to 50 for a single-category brand, up to 150 for a multi-category catalog. Fewer, sampled five times each, beats hundreds sampled once. Add prompts from real shopper questions as you find them.
Is "share of voice" comparable across engines?
Only within an engine. Each engine cites a different mix of sources and returns a different number of brands per answer, so report share of voice per engine and track the trend inside each one. A combined number hides which engine moved.
Can I just check ChatGPT myself once a month?
You can, but one logged-in check is a single noisy sample. Use a fresh session, repeat the prompt several times, and write down the result every time. A spreadsheet with dates, prompts and outcomes is a legitimate starting panel.
What should I do when a competitor appears and I do not?
Read the answer and note which page it cites. Most product recommendations draw on third-party lists, videos and reviews rather than brand sites, so the fix is usually to earn a place on that page, plus closing any attribute or FAQ gap on your own product page.
Does the free Merchant Center report replace a paid tool?
For Google AI Mode and AI Overviews in supported countries, it is the best number available because Google computes it from its own data. It says nothing about ChatGPT, Perplexity or Copilot, so you still need sampling or a tool for those.
Sources
- https://ahrefs.com/blog/ai-search-overlap/
- https://www.semrush.com/blog/ai-search-seo-traffic-study/
- https://ahrefs.com/blog/self-promotional-content-ai-seo-experiment/
- https://www.semrush.com/news/463141-semrush-releases-expanded-2026-ai-visibility-index-analyzing-126-million-ai-search-prompts/
- https://support.google.com/merchants/answer/17200695
- https://otterly.ai/alternatives/pricing/
- https://ahrefs.com/brand-radar
- https://www.semrush.com/pricing/ai/
- https://peec.ai/pricing
- https://www.tryprofound.com/pricing
- https://shopify.dev/docs/apps/build/checkout/thank-you-order-status/add-survey
- https://help.shopify.com/en/manual/checkout-settings/customize-checkout-configurations/upgrade-thank-you-order-status
- https://apps.shopify.com/fairing
- https://apps.shopify.com/zigpoll
- https://otterly.ai/blog/chatgpt-reddit-citations/
Questions this article answers
How many prompts do I really need?
Enough to cover your buying stages and hero products without being noisy: 30 to 50 for a single-category brand, up to 150 for a multi-category catalog. Fewer, sampled five times each, beats hundreds sampled once. Add prompts from real shopper questions as you find them.
Is "share of voice" comparable across engines?
Only within an engine. Each engine cites a different mix of sources and returns a different number of brands per answer, so report share of voice per engine and track the trend inside each one. A combined number hides which engine moved.
Can I just check ChatGPT myself once a month?
You can, but one logged-in check is a single noisy sample. Use a fresh session, repeat the prompt several times, and write down the result every time. A spreadsheet with dates, prompts and outcomes is a legitimate starting panel.
What should I do when a competitor appears and I do not?
Read the answer and note which page it cites. Most product recommendations draw on third-party lists, videos and reviews rather than brand sites, so the fix is usually to earn a place on that page, plus closing any attribute or FAQ gap on your own product page.
Does the free Merchant Center report replace a paid tool?
For Google AI Mode and AI Overviews in supported countries, it is the best number available because Google computes it from its own data. It says nothing about ChatGPT, Perplexity or Copilot, so you still need sampling or a tool for those.