How do I measure my brand's visibility in AI answers?

Last updated October 14, 2026. Written by the Third Coast AI team.

Measure brand visibility in AI answers by writing a fixed set of 60 to 120 prompts the way your buyers ask, running each prompt several times in each engine on a schedule, recording whether you were named and whether your pages were cited, and reporting two rates: the share of runs in which you were named (visibility, or share of voice when compared with competitors) and the share of runs in which one of your pages appeared as a citation (citation rate). A single run is not a measurement, because the engines return different lists on repeated runs, so the rate over many runs is the number, and the trend in that rate over months is the result.

This page sets out the method in enough detail to run it with a spreadsheet or to check that a tool or vendor is doing it properly. It is the measurement behind our audit, and nothing in it is proprietary; the parts we add in practice are the retrieval-on runs in ChatGPT and Gemini and the capture of the hidden searches the engine runs behind a question, which tell you what to write rather than how you are doing.

How should the prompt set be built?

Start from personas rather than keywords. For each buyer type, write the questions they ask at each stage, in their words, including the hedged and badly phrased ones, and sort them into four groups: visibility prompts ("who are the best [category] companies near [place]"), decision-factor prompts ("what should I look for when choosing a [category]"), competitor prompts ("compare [you] and [rival]") and fact-check prompts ("does [you] offer [thing]" and "where is [you] based"). Sixty prompts is a floor for a single brand in a single market; 120 is comfortable; beyond that you are usually adding variants rather than coverage. Keep the set fixed for at least a quarter so the trend is comparable, and version it when you change it.

The fact-check group matters more than its size suggests, because wrong facts are the fastest thing to fix and the easiest to miss when you only track visibility. If the engine names you but says you are in the wrong city or sells a product you discontinued, the visibility number looks fine and the buyer is misled, and the correction guide explains what to do when the log fills up.

Which rates should you report?

Visibility and citation rate are different numbers and both matter. Ahrefs' study of 75,000 brands found third-party mentions correlate with AI citations at 0.664 against 0.218 for backlinks, so a brand can be named often while its own pages are never cited, which means its presence rests on what others say; the reverse, cited but rarely named, usually means the pages answer a sub-question but the brand is not recognized as an entity. The mentions versus backlinks page covers the first case and the entity page the second.

How many runs are enough, and how often should you re-run?

More than one per prompt, and on a fixed cadence. SparkToro and Gumshoe.ai had 600 volunteers run 12 prompts through ChatGPT, Claude and Google's AI a combined 2,961 times and found less than a 1 in 100 chance of the same brand list appearing twice, and roughly 1 in 1,000 for the same order; their recommendation, which matches ours, is to measure visibility as a percentage over dozens to hundreds of prompts run multiple times and to ignore any tool that reports a ranking position in AI. In practice we run each prompt three to five times per engine per cycle, which turns a 100-prompt set into 300 to 500 answers per engine, and we report the monthly rate with the number of runs behind it. Weekly runs are useful for a fact-check set after a correction; monthly is the right cadence for the visibility set, because the engines' indexes and the third-party record do not change faster than that in most categories.

The variance also sets how much change is real. If 100 prompts run four times each produce a visibility rate of 22% one month and 26% the next, that is inside the noise for most categories, and the honest report says so; a move from 22% to 40% over a quarter, with the gain concentrated in the prompts whose pages you rewrote, is a result. We report the rate, the run count and the prompt-level detail so a reader can see which prompts moved, and the page on why answers change between runs explains the mechanism behind the variance.

Should every engine be measured the same way?

The method is the same, and the settings differ. Run ChatGPT with web search on and, separately, off, because the two answer from different sources: with search on, OpenAI's ChatGPT search help page says the query is rewritten into one or more targeted searches sent to search partners, and with it off, the answer comes from training data, where mentions on other sites carry more weight than your own pages. Run Gemini with grounding on. Run Google AI Overviews and AI Mode from a clean browser profile in the location you care about, since both are affected by location and personalization. Run Perplexity and Copilot as a signed-out user. Record the citations in every case, and for Copilot and Bing, pull the first-party numbers from Bing Webmaster Tools, which has reported citations, average cited pages and grounding queries since Bing's February 2026 announcement and added Citation Share in June 2026. Only about 11% of cited domains overlap between ChatGPT and Perplexity, so report the engines separately and resist averaging them into one score.

What does the baseline tell you that you did not know?

• Which competitors the engines name that you did not consider competitors, because the engine's candidate list comes from what is written about the category rather than from your market map.
• Which pages of yours are cited at all, which is often a surprise; Kirsten and colleagues found 53% of the domains cited in AI Overviews were not in Google's organic top ten for the same query, and 40% were outside the top 100.
• Which third-party pages carry your presence, which tells you where a correction or an earned mention would travel furthest.
• What the engines get wrong, which the fact-check group surfaces in the first run.
• Which prompt groups you are absent from entirely, which is usually where the decision-factor pages have not been written.

Once the baseline exists, the measurement runs itself in any of the tools on the comparison page, and the monthly report becomes a short document whose shape the KPI guide describes.

Frequently asked questions