How we measure AI search visibility

1. Buying prompts, not keywords

We track the questions shoppers actually ask AI before they buy: "best trail running shoes under $150", "waterproof hiking boots for wide feet", "<your brand> vs <competitor>". The prompt assistant suggests them from your catalog and categories; you keep, edit or add your own.

Plans set how many prompts each store tracks (from 40 on Starter). Prompts are tagged by product, category and market so every metric can be split the same way.

2. Five AI engines, asked like a shopper

We ask ChatGPT, Google AI Mode, Google AI Overviews, Gemini and Perplexity. Answers are collected through two data providers, cloro and DataForSEO, as a logged-out user in the country you sell in. If one provider fails, the other takes over where it supports the engine.

Personalised answers for logged-in users can differ. We say so wherever it matters.

3. Repeated runs

Every prompt runs on every selected engine on a fixed schedule: daily from Growth, every 2 days on Starter. One answer is noise; thirty answers start to show a pattern. Ranges narrow as answers add up.

Each account has a hard daily cap on data costs. If a run does not fit, it is skipped, shown on your dashboard, and those prompts go first next time. Nothing is silently dropped.

4. Reading each answer

From every answer we extract the brands and products named, their order, product cards with price, the sources cited, and the tone of each mention. Products are matched to your imported catalog. The raw answer is stored so you can check any data point yourself.

5. The formulas

Same definitions as the rest of the category, so numbers are comparable. We add the uncertainty.

6. Ranges and sample size

Every rate is shown with a 95% Wilson interval and n, the number of answers behind it. Example: 3 mentions in 10 answers is 30%, but the true rate could be anywhere from about 11% to 60%. With 100 answers, 30% narrows to about 22–40%.

Under 30 answers we flag a small sample. If your range and a competitor’s overlap, treat the gap as possible noise, not a result.

7. Measured vs estimated

• Measured: real AI answers we collected, and real orders from your connected Shopify store.
• Estimated: the revenue potential calculator. Every assumption is visible and editable, and the result is labeled "Estimated". It never appears in the same chart as measured results.
• Source-backed assumption: a default from a published study, with the source shown. Replace it with your own numbers any time.

8. AI revenue from Shopify orders

With Shopify connected (Growth plan and up), we classify each order by referring site, landing page and UTM source: chatgpt.com, perplexity.ai, gemini.google.com, Copilot and other assistants. Some AI apps strip the referrer, so this likely undercounts AI orders. We never present it as complete. Shopify connect is coming soon.

9. Before and after a fix

When you mark an action done, we compare the same prompts in the 14 days after with the period before. That is a measurement, not proof that this one action caused the change: seasons, promotions and model updates happen too, and we say so next to the result.

10. What we cannot do

• No one outside the AI companies knows their ranking logic. We show signals that correlate with being recommended, not a formula that forces it.
• Nobody can guarantee a recommendation in AI answers, including us.
• Engine names describe what we query. We are not affiliated with or endorsed by their makers.