What is AI visibility tracking?

Why rank tracking doesn't cover this

A rank tracker tells you where a page sits in a list of links. An assistant writes one reply, and your brand is either in it or it isn't. The two are also more loosely connected than most people assume.

When Ahrefs studied 15,000 prompts in 2025, only 8.0% of ChatGPT's in-text citations and 6.1% of its references also ranked in Google's top 10 for the same prompt. Gemini sat at 8.6% and Copilot at 8.2%. Perplexity was much higher at 28.6%. So a strong Google position tells you little about ChatGPT, and a good result in one assistant tells you little about another. You have to look at each one.

Prompts, mentions and citations are different things

These three words get used loosely, but each one measures something different. It helps to see them together.

A prompt is what gets asked

A prompt is the text someone types into an assistant, such as "best running shoes for flat feet". It's the one part of tracking you control, so choosing well matters. We cover that in how to write prompts worth tracking.

A mention is your name in the reply

A mention means the assistant says your brand in its written reply, with or without a link. It's the closest thing to a recommendation, because it's what the reader sees first.

A citation is a link to a source

A citation is a page the assistant links to as a source. That page might be yours, but it could just as easily be a review site, a news story or a rival. Citations show you which pages shape the reply, and they give the reader a route to follow.

What each combination tells you

Because the two are separate, a brand can earn one without the other. Each result points to different work.

• Mentioned and cited. This is the strongest place to be. AirOps found that brands earning both are 40% more likely to resurface across consecutive runs than brands with a citation alone.
• Mentioned, not cited. The assistant knows you, but the reader has no link to your site. Look at which pages it cites instead.
• Cited, not mentioned. Your page helped shape the reply, yet the reader may never see your name. Check that the page says clearly who you are and what you do.
• Neither. Someone else is being suggested. Those rivals, and the pages cited for them, are where to start.

Why one check tells you very little

The natural first step is to open ChatGPT, type a prompt and see what comes back. The trouble is that the next attempt may look quite different. In the same AirOps report, published in December 2025, only 30% of brands stayed visible from one reply to the next, and just 20% remained present across five consecutive runs of the same prompt.

Tone moves around even more. Pratyush Kumar's study of more than 100,000 prompt responses found that whether a brand is framed positively or negatively flips about 6.7 times more often than whether it's mentioned at all.

So a single screenshot is an anecdote. Useful tracking runs each prompt several times and reports a rate: mentioned in three of five runs, say, rather than a plain yes or no. Small changes from one week to the next are usually noise, so look at the trend over at least four weeks before you read anything into it.

What a useful tracker reports

Whichever tool you use, or if you do this by hand, these are the things worth recording for each prompt:

• How often you're mentioned, as a rate across repeated runs.
• Which pages get cited, yours and everyone else's. Other people's pages show you where you could be listed.
• Who gets mentioned instead. Rivals you didn't expect are often the most useful finding.
• How you're described, including anything out of date or wrong.
• Each assistant on its own, and each place you sell if results differ by location. A single blended score hides the gaps.
• What to do next. A number you can't act on won't change anything.

What to expect if your brand is small

Don't be alarmed by a low starting figure. In Kumar's data, covering 100+ brands between March and May 2026, global household names appeared in 73% of relevant replies on their first runs. Established mid-market and regional brands appeared in 44%, and niche and small brands in just 11%.

That gap reflects how well known a brand is more than how well its pages are written, so compare yourself with your own earlier numbers and your nearest rivals rather than with the largest names in your market.

Where to start

Write a small set of prompts your buyers really would use, run them repeatedly across the assistants that matter to you, and read the results before you change anything. When you know where the gaps are, turn to how to earn more citations from AI assistants.

Suggestible does the repeat work for you. It asks AI assistants the prompts your buyers use, then reports your mentions, your citations, your rivals and what to fix first.