One run tells you whether a gap exists. It cannot tell you how big it is, because these answers are probabilistic and personalised. Use the fast check to find the question worth tracking, then track it properly.
What question should you ask?
Use the kind of question a buyer asks before they know which business to choose. A good test includes the category, a meaningful constraint, and, when relevant, the location.
"What are the best [service providers] for [type of buyer] who need [important outcome or constraint] in [market]?"
Do not ask for your company by name. That tests whether the engine can retrieve you, which is a different and much easier question than whether it recommends you during genuine discovery. Both are worth knowing, and confusing them is the most common way people conclude their AI visibility is fine when it is not.
"Best commercial cleaning companies for office blocks in Guwahati." Tests whether you are in the shortlist at all.
"Compare [you] and [competitor] for [use case]." Tests whether the engine knows your differences accurately.
"We need X but our constraint is Y. Who should we talk to?" This is how buyers actually type and it is where full-sentence queries land.
"What does [your business] do?" Tests accuracy of the facts about you, which is a separate failure from absence.
How do you run the 60-second test?
- Open the tool in a fresh conversation, signed out if you can, so your own history does not shape the answer.
- Ask one representative buyer question.
- Note every company named in the initial answer, in order.
- Open the cited sources and see what actually supported the shortlist. This is the step people skip and it is the one that produces the plan.
- Repeat the exact question in a second engine, because they retrieve differently.
Four stages, and they fail for different reasons
"Are we in AI" is not one question. It is four, and knowing which one you are failing tells you what to do. Most businesses fail at the third, and almost everyone misdiagnoses it as the fourth.
| Stage | What it means | If you fail here |
|---|---|---|
| Retrievable | The engine's crawler is allowed to fetch your pages | A robots.txt or CDN problem. Cheapest fix in the practice and nothing else works until it passes |
| Retrieved | You appear in that engine's underlying search results | An ordinary ranking and indexation problem in whatever index it uses |
| Cited | A page of yours is linked as a source in the answer | Usually a passage problem. The answer exists on your page but is buried, hedged, or written as prose where a procedure was wanted |
| Named | Your business appears in the answer text as a recommendation | A corroboration problem. The engine has no third-party evidence that you are a credible option |
One of our own pages was cited in a Google AI Overview, once, for a sentence stating that causal attribution is difficult. A competitor was cited five times, carrying every step of the answer's method. They had written the procedure as a numbered list. We had written the same procedure as prose. We passed "cited" and lost the substance of the answer, which is a failure mode invisible to any tool that only counts citations.
What should you record?
Was your business named at all, and where in the list?
Was it described correctly? A wrong description is worse than absence.
Which domains supported the answer? This list is the placement plan.
Without both, the result cannot be compared to anything later.
Keep mention and citation in separate columns. An answer can cite your page without naming your business, and can name your business while citing a directory. If they share a column the data cannot answer either question.
Reading the result
The useful output is not yes or no. A missing brand, an inaccurate description and an uncited mention are three different problems with three different next steps, and the cited-instead list usually points at the cheapest one.
If directories and review platforms dominate the citations, your gap is a profile you can fix this month. If trade publications dominate, it is an outreach problem measured in quarters. If a competitor's own comparison page is cited repeatedly, they have written something you have not.
What can this quick test prove?
It can reveal a visible gap. It cannot establish a stable score, and treating one run as a measurement is the main way people waste money here.
- Answers vary by engine, time, location, account history and exact wording. Two runs of the same prompt can name different companies.
- Signed-in sessions carry personalisation. A result you like may be your own history reflected back.
- One query is not a market. A business can be strong on the questions it wrote pages for and absent everywhere else.
- Citation is not traffic. Being in the answer and being visited are now separate outcomes, and the first is far more common than the second.
Use the fast check to find the question. Use a tracked prompt set to prove whether the answer is changing.
A proper baseline uses a fixed prompt set, fresh sessions, several runs per prompt, and rates rather than single answers. The full method is published here, and if the gap turns out to be real, whether it is worth acting on is a separate question with its own test.
Want a second pair of eyes? Send us your site and we will run a small prompt set and email you what came back, including who was recommended in your place.