The AI Visibility Check
Updated
The AI Visibility Check asks four AI assistants the questions a real shopper asks when they already know what they want and only need to know where to buy it. It asks about one of your products, from one location, in one language, and reports whether your store gets named, at what rank, and which stores got named instead. Everything it reports comes from what the assistants actually said, kept word for word.
Which AI assistants does it check?
Four, and each one is asked separately:
| Assistant | Who makes it |
|---|---|
| ChatGPT | OpenAI |
| Gemini | |
| Google AI Mode | Google, the AI answers inside Search |
| Claude | Anthropic |
Every question goes to all four, and every result is reported per assistant. That split matters: being named by one assistant and ignored by the other three is a very different problem from being ignored by all four, and an average would hide it.
Which questions does it ask?
The ones people ask at the moment they are ready to buy. "Which running shoes are best" is a different fight, one for the brand that makes the shoes. You compete later in the journey, when the shopper has settled on the product and is only choosing a store.
| Intent | The question, roughly |
|---|---|
| Where to buy | "where to buy [product] in [location]" |
| Best place to buy | "best place to buy [product] online in [location]" |
| Authentic | "where to buy authentic [product] in [location]" |
| Cheapest | "cheapest place to buy [product] in [location]" |
| Free shipping | "where to buy [product] with free shipping in [location]" |
More intents get added over time, so treat this as today's list rather than a fixed set. Each one is asked to each assistant, and the report scores them separately: you can be the obvious answer for "where to buy" and invisible for "cheapest", and those call for different fixes.
How are the questions built?
Not by pasting in your product title. Shop titles are written for search engines, and no shopper types "AMD RYZEN 5 PRO 3500 8GB 256GB SSD 14 Inch, Refurbished Laptop" into a chat window. Ask an assistant that and it matches nothing.
So the check rewrites the title into something a person would say, while keeping what makes it your exact product:
- The brand goes in front. If your title does not name the brand, the vendor name is added, because the brand is the strongest signal an assistant has for matching the right product.
- Identifying parts stay. The product line, the model number, the volume, the version. When in doubt, they stay in.
- Only obvious clutter goes. Repeated category suffixes, part numbers that mean nothing to a buyer.
The middle rule is there because cutting too hard is worse than not cutting at all. Trim "5 PDRN Collagen Intense Vitalizing Serum" down to "PDRN Collagen Serum" and the assistants stop matching the product you sell and start naming a different brand's product instead. Then the check would be measuring the wrong thing entirely, and you would never know.
One cleaned-up product identity is used for every intent and every assistant, so the results can be compared with each other. And you never have to take this on trust: the report shows you the exact question that was sent, under each intent.
How can it check a place I am not in?
The location, language and currency are given to the assistant as context, the same way a local shopper's own question carries them. Nothing is spoofed and no VPN is involved.
One check measures one product, in one location, in one language. Other markets need their own checks, because the answer genuinely differs: the stores recommended in Dubai are not the ones recommended in London.
Why can only live products be checked?
Because a product that is not live has no public page, and an assistant cannot recommend a page that does not exist. The check would return "not visible" every time, which would tell you nothing about your store and everything about the product being unpublished.
Non-live products still show in the picker, greyed out with a Draft badge and a Publish link. Publish, then check.
What if an assistant does not name a price?
It says so. If an assistant recommends a store without stating its price or shipping, the report shows N/A. It does not go looking somewhere else for a number and quietly fill it in, because then you would be reading what the web says rather than what the AI said, and the whole point is the second one.
The same honesty applies to your own result. If no assistant names your store, the report says so plainly. That is not the app failing to find something: it is the finding.
Next step
See Reading your report for what each number in the report means.
