AI Visibility Tracker

Build the prompt set and the tracking sheet you need to measure whether AI assistants name your brand. It does not query the assistants for you, and the first FAQ below explains why that is deliberate.

Why this builds a sheet instead of checking for you

Assistant answers vary between runs of the same prompt. Model updates, sampling and personalisation all change the output, so a single check tells you what happened once rather than what typically happens when somebody asks.

That means a one-shot checker would hand you a number that looks like a measurement and is a single draw. What you actually need is the same prompt run repeatedly over time, recorded as a rate: named in six of ten runs, not named in the other four.

So this tool builds the two things that make that possible. A prompt set derived from your category and your competitors, and a tracking sheet with a column per assistant and a row per run. Our full reasoning is in what to watch in AI search, and whether to buy a tool.

How to use the prompt set

Paste each prompt into the assistants your buyers actually use, which is usually two or three rather than ten. Record whether your brand appears and which competitors appear instead of you, because the second column is the more useful one.

Sample each prompt several times per reporting period rather than once per month, then report the rate. A brand appearing in six of ten runs this month and five of ten last month has probably not moved at all.

Any prompt with no page behind it on your own site is a content brief, and that list is the most valuable output here. For the method we use on our own site, see how many prompts an AEO agency should track, or read about our AI SEO and Answer Engine Optimization work.

Frequently asked questions

Why does this not just check ChatGPT for me?
Because the answer would be misleading. Assistant responses vary between runs of the same prompt, so a single automated check returns one draw from a distribution and presents it as a measurement. A tool that showed you a visibility percentage from one run would be reporting noise with a decimal point on it.
How many prompts should I track?
Enough that each one is a question a real buyer would ask, and no more. For most businesses that is between twenty and a couple of hundred. We track 95 on our own site, each mapped to the page that should answer it.
How often should I sample each prompt?
Several times per reporting period, not once. The reporting cadence and the sampling cadence are different things, and conflating them is how a single draw gets reported as a trend. Monthly reporting on weekly sampling is a reasonable starting shape.
Which assistants should I check?
The ones your buyers use, which is usually two or three. Breadth of coverage is the most marketed feature in this category and the least important, because a prompt checked across ten assistants your market does not use tells you nothing.
Does any of my data leave the page?
No. This tool builds the prompt set and the sheet in your browser with plain JavaScript, so your brand name, competitors and prompts never leave the page. It works with the network disconnected.
What do I do with a prompt I am not named in?
Check whether you have a page that answers it at all. If not, the prompt is a content brief. If you do have one, check whether an assistant crawler has ever fetched it in your server logs, because a page that is never read cannot be cited.