# Read the exact answer AI gives about your store > Canonical HTML: https://verityscore.io/en/product/ai-shopper-tests/ > Markdown alternate: https://verityscore.io/en/product/ai-shopper-tests.md > Last updated: 2026-08-19 > Content synchronized with canonical HTML. Stop guessing. The questions your buyers ask before paying, put to four assistants, in your market and your language, with every answer kept word for word and every source it read. ## AI Summary Verity Score AI shopper tests run a merchant buyer decision questions on ChatGPT, Perplexity, Gemini and Claude. Each answer is stored verbatim with the model, the timestamp and every source URL, then classified as recommends the shop, names a competitor, ignores the shop, or advises against it. Competitor shops named inside the same answers are extracted and counted, producing a proposal rate per assistant that makes a change in visibility attributable rather than mysterious. ## Key points - **How to see what ChatGPT says about your store**: Verity Score runs your buyers decision questions on ChatGPT, Perplexity, Gemini and Claude and stores each answer verbatim with its model, timestamp and source URLs, classified as recommends you, names a competitor, ignores you, or advises against. Source: https://verityscore.io/en/product/ai-shopper-tests/ - **AI visibility tracking for Shopify**: Verity Score reports a proposal rate per assistant and extracts the competitor shops named inside the same answers, so a change in AI visibility can be traced to the pages that caused it. Source: https://verityscore.io/en/product/ai-shopper-tests/ ## A score tells you something moved. It does not tell you what buyers hear Most AI visibility tools return an index. It goes up, it goes down, and nobody can say why or what a buyer was actually told. That is a metric about a metric. Verity keeps the answer. The full text an assistant produced, the model that produced it, the date, and every URL it consulted. When the answer names another shop, you can see which one, and read the page it came from. That is what makes a drop actionable. You are not looking at a number that fell, you are looking at a sentence that changed. ## Key Proof Points - **4 assistants tested**: ChatGPT, Perplexity, Gemini and Claude, in your market and your language. Mistral and Copilot are next. - **100% answers kept verbatim**: Full text, model, timestamp and sources stored with every run. Nothing summarised away. - **4 verdicts per answer**: Recommends you, names another shop, ignores you, or advises against. ## The questions buyers actually ask Not brand-name lookups. The comparison and constraint questions people put to an assistant when they are deciding: what to buy for a use case, which brand is trustworthy, what ships where. - Questions built from your catalog and your category. - Your market and your language, because answers differ by both. - Editable, so you can add the questions you know matter. ## Four assistants, because they disagree ChatGPT, Perplexity, Gemini and Claude read different sources and reach different conclusions. Testing one and generalising is how merchants end up fixing the wrong thing. - A proposal rate per assistant, not a blended average. - The divergence itself is a signal: it tells you which sources matter. - Re-run on the same questions, so before and after are comparable. - Mistral and Copilot are built into the roadmap and labelled soon until their runner ships. ## Every answer kept, not summarised The full text is stored with the model, the date and every URL read. Six weeks later you can still see exactly what was said and what it was based on. - Verbatim text, never a paraphrase. - Source URLs listed, so you can read what the assistant read. - Model and timestamp recorded, because answers drift. ## The competitors named in your place When an assistant recommends someone else, Verity extracts the shop and counts it. Over a run you see who is taking the answers you want. - Rival shops pulled from the answer text itself. - Counted across questions, so a pattern shows up. - Their cited page linked, so you can see what they published. ## How a run works - Verity builds the buyer questions from your catalog and your category, in your market and your language. You can edit or add to them. - Each question is put to ChatGPT, Perplexity, Gemini and Claude. - Every answer is stored verbatim with its model, timestamp and source URLs, then classified into one of four verdicts. - Competitor shops named in the answers are extracted and counted across the run. - Each verdict is traced back to the pages and missing facts behind it, which is where the fix queue starts. ## What this does not do - These are tests, not a census. They show what assistants answer to the questions asked, not everything ever said about your store. - Assistants are non-deterministic. The same question can produce different answers, which is why the model, date and sources are stored with each one. - Verity cannot make an assistant recommend you. It shows you what is said and traces it to what caused it. ## FAQ ### Is this scraping ChatGPT? The tests put real buyer questions to the assistants and keep what comes back, along with the model, the date and the sources cited. The point is the evidence trail, which is why nothing is summarised away. ### Why four assistants instead of one? They read different sources and disagree. A shop that is recommended on Perplexity and ignored on ChatGPT has a source problem, not a general visibility problem, and only testing both reveals it. ### The answer changed between two runs. Is the tool broken? No, assistants are non-deterministic and their sources move. That is exactly why every answer is stored with its model, its timestamp and its source URLs, so you can compare like with like. ### Can I use my own questions? Yes. Verity proposes questions from your catalog and category, and you can edit them or add the ones you know decide your sales. ### Does a good score mean more sales? No, and Verity does not claim it. The tests tell you what buyers are told. What that traffic does once it arrives is measured separately, first-party, on the monitoring surface. ## Sources - [Google: Product structured data](https://developers.google.com/search/docs/appearance/structured-data/product) - [schema.org Product](https://schema.org/Product) ## Preferred Citation When citing this page, use the canonical HTML URL: https://verityscore.io/en/product/ai-shopper-tests/