How we measure AI visibility
The method is public and the results can be checked. You can follow it yourself or hand it to us.
Three principles
Evidence for every finding
Each issue comes with a screenshot, a link or the exact field text, so you can check it line by line.
Appearance rate, not rank
AI recommendations change every run. Only appearance rate across many runs means anything. We never promise rankings.
Real information only
We never invent specs, reviews or brand mentions, and never build pages meant only for crawlers. Google lists such tactics under its spam policies for AI search.
Step 1
Ask the way buyers ask
Prompts should sound like real local buyers, not like a seller's keyword list. Each key product is covered by four types of question. The examples below are for portable power stations in the US; other markets are tested in the local language.
| Type | Example | What it tests |
|---|---|---|
| Recommendation | What's the best portable power station under $500 for weekend camping? | Whether you make the shortlist |
| Use case | I need something to run a CPAP machine for two nights off-grid. What should I buy? | Whether you match the use case |
| Comparison | Brand A vs Brand B: which is better for RV trips? | How AI sees the two brands |
| Brand | Is Brand B reliable? Where does it ship from and what's the return policy? | Whether AI describes you accurately |
Once the prompt set is final, each prompt is numbered and locked. Re-tests use exactly the same wording.
Step 2
Run many times in the target market
- Platforms: ChatGPT, Google AI Mode and Gemini, Microsoft Copilot and Perplexity, plus the built-in shopping assistant for products sold on Amazon and other marketplaces, depending on what is available in the target market.
- Manual tests in the target market's language and region settings, in a fresh chat each time with personalization off. US tests are run in the US. The report states the test setup for every run.
- Each prompt runs 3 to 5 times per platform. We record whether you appear, in what form, which competitors appear, which pages are cited and exactly how the AI describes your product.
- Every run is saved as a screenshot. No automated scraping, and no getting around any platform's limits.
Step 3
30 readiness checks
Five areas, 20 points each, based on the platforms' published product data specs and public research. No platform publishes its ranking rules, so a high score doesn't guarantee a recommendation, but a low score almost always costs you.
Product identity
Does the title state brand, category, key spec and use? Is the brand name consistent? Is there a GTIN or other unique ID? Are categories and variants structured properly?
Attributes
Are the specs buyers compare complete, specific and in local units? Does the description say who it's for and when to use it? Is the language natural? Are there enough images and videos?
Offer terms
Are price and availability consistent everywhere? Are shipping cost, delivery time and ship-from location stated? Are the return window and warranty clear?
Trust signals
Are there real, machine-readable product and store reviews? Is there an FAQ, certification and company information? Are there videos and reviews off-site?
Machine access
Do robots.txt or CDN settings block AI crawlers by mistake? Is content still there with scripts off? Is the structured markup correct? Is the catalog connected to Shopify Catalog or Google Merchant Center?
OpenAI's product feed spec, Shopify Catalog and Amazon's public material on its shopping assistant all point the same way: complete, accurate, consistent, machine-readable product information.
Step 4
Connect results to causes
| Metric | Meaning |
|---|---|
| Appearance rate | Appearances divided by total runs, by platform and prompt type |
| Share of voice | How often you appear compared with your main competitors |
| Description accuracy | How much of what AI says about price, specs, delivery time and returns is correct. Wrong information causes after-sales problems and is fixed first |
| Citation sources | Whether AI cites your site, marketplaces, review media or videos, which decides whether fixes happen on-site or off-site |
| Readiness score | The total of the 30 checks: the part you can improve directly |
Readiness score and appearance rate are always reported separately. After data is fixed, appearance rate usually takes a few weeks to move, and it fluctuates.
What would your products score today?
The free check covers one target market and 3 to 5 key products.