How to measure and improve your brand's visibility in AI answers (2026)
Measure your brand's visibility in AI answers with mention rate and share of voice across ChatGPT, Perplexity, and Gemini, then the levers that improve it.
Buyers now ask ChatGPT, Perplexity, and Gemini the questions they used to type into Google, and the answer they get names a handful of brands. If yours is not one of them, you are invisible at the exact moment a decision is being made — and unlike a search ranking, there is no results page to scroll, so second place is often no place. The first job is to find out where you stand: whether the engines mention your brand at all, how often relative to competitors, and how they describe you. The second is to move that number.
The reason this needs its own workflow is that AI visibility does not behave like SEO. Answers are non-deterministic, so the same prompt returns different brands on different runs; being cited is not the same as being recommended; and the levers that improve your mention rate sit mostly off your own site, in the entity signals and third-party mentions models trust. This is the measure-then-improve loop end to end — build a repeatable measurement, read the three numbers that matter, and work the levers that actually raise them. For the metric definitions and formulas underneath, see the companion guide [AI visibility measurement](/guides/ai-visibility-measurement); for why AI can know your brand yet never name it, [the AI brand visibility gap](/guides/ai-brand-visibility-gap-in-search).
The steps
Separate the two questions and define your metrics. You are answering two different things, so name them before you count. "Does AI mention us?" is your mention rate — the share of relevant answers your brand appears in, sometimes called brand visibility. "How much, versus everyone else?" is your AI share of voice — your mentions as a fraction of all brand mentions across the same prompt set. Layer three more on top: citation rate (were you named with a source link, not just described), prominence (were you first or buried), and sentiment (how were you characterized). Mention rate tells you if you exist to the model; share of voice tells you whether you are winning the category.
Build a stable prompt set from real buyer questions. Your measurement is only as good as the questions behind it, so write the ones a real prospect would ask — not keywords. Cover the three types that matter: category questions ("best tools for X"), comparison questions ("X vs Y"), and problem questions ("how do I solve Z"). Include a few branded prompts ("what is [your brand]") to check how you are described, but weight the set toward unbranded category queries, because that is where a mention is actually earned. Twenty to fifty prompts is enough to be meaningful. Freeze the list — the value is in comparing the same questions over time, not in a one-off snapshot.
Run the set across the engines your audience uses. Run every prompt through the surfaces your buyers actually use — ChatGPT, Perplexity, Gemini, Google's AI Overviews and AI Mode, and Microsoft Copilot cover most of the market, with Claude and Grok as needed. Standardize the conditions: use logged-out or fresh sessions so personalization and memory do not skew the result, and keep location and language constant. Crucially, run each prompt several times per engine on the same day. Answers are non-deterministic, so one run is a coin flip; you are measuring a rate, and a rate needs repetition.
Score every response on the same rubric. For each answer, log five things in a spreadsheet: was your brand mentioned (yes/no), was it cited with a clickable source, roughly where it appeared (first, mid, last), how it was described (positive, neutral, wrong), and which competitors showed up instead. That last column is the most useful one you will keep — the competitors an engine names in your absence are your real benchmark and your outreach target list. Keep the rubric identical every cycle so the numbers are comparable; a scoring change mid-stream invalidates your trend.
Compute the numbers and benchmark them. Now do the arithmetic. Mention rate is the count of answers naming you divided by total answers, times 100. Share of voice is your mentions divided by the total brand mentions (yours plus every competitor's) across the set. A single figure means little on its own — one 2026 industry report (AthenaHQ's State of AI Search) put the average brand mention rate around 17%, so context is everything. Read your numbers next to two things: the competitors from your rubric, and your own prior period. "Cited in 20% of answers" only becomes a story when it sits beside where rivals land and which direction you are trending.
Decide: manual tracking or a dedicated tool. For a handful of priority queries you can run this by hand in a spreadsheet, or automate the runs with each engine's API for consistency. To track a large prompt set across many engines with competitor share-of-voice built in, use a dedicated platform — Profound, Peec AI, Otterly, AthenaHQ, Ahrefs' Brand Radar, and Semrush's AI visibility tracking all monitor mentions across the major answer engines and benchmark you against named competitors. The tool is a time-saver, not a strategy; whichever you pick, the discipline is the same fixed prompt set on a fixed schedule.
Improve the entity layer first. To be named, a model has to be confident about who you are. Make your brand an unambiguous entity: use one exact name and one consistent description everywhere you appear, claim your Google Knowledge Panel, and — where you genuinely qualify — get into Wikipedia and Wikidata, which act as identity anchors models cross-reference. Add Organization schema on your site so the machine-readable facts match the prose. Inconsistency is what keeps a brand in the "AI knows us but won't name us" trap; when every source tells the model the same true thing about you, it can cite you without hedging.
Work the off-site and video layer, then re-measure. Most of the lift is off your own domain. Earn third-party mentions where AI systems already look — Reddit threads, review sites like G2 and Capterra for software, and earned editorial that quotes your experts by name and title. Publish claims in extractable form so a model can lift a clean sentence and attribute it. And treat video as a first-class channel: a large-scale Ahrefs study of 75,000 brands found YouTube mentions were the single strongest signal of AI visibility, ahead of backlinks — with the honest caveat that this is correlation, not a switch you flip, so it reflects broad cross-surface presence rather than one channel. Then close the loop: re-run the same set monthly, watch the trend, and when a competitor owns a query you should own, publish the answer that is missing.
Common gotchas
A single run is noise, not a measurement. AI answers are non-deterministic, so the same prompt names different brands run to run — always average multiple runs over a fixed set and report the trend, never a one-shot reading.
Mention, citation, and recommendation are three different things. Being described, being linked, and being actively recommended are separate outcomes; collapsing them into one "we're visible" number hides where you are actually losing.
A visibility score with no competitor benchmark says nothing. "Cited 20% of the time" only lands next to where competitors land — the gap, not the absolute number, is the insight.
Logged-in sessions and personalization skew results. Memory, history, and location change what the engine returns, so run from clean or logged-out sessions with location and language held constant, or you are measuring your own footprint back at yourself.
Measuring only branded prompts flatters you. Of course the model describes you well when you name yourself; mention rate matters on the unbranded category questions where a prospect has not decided yet.
Improvement is prospective and lagged. Models refresh their understanding on their own cadence, so entity and off-site work will not show up in your numbers the same week — hold the measurement steady and give it cycles.
Where Kompozy fits
Almost every AI-visibility measurement ends at the same diagnosis: the model is not confident enough about you, and you are not present on enough of the surfaces it reads. Both are production problems dressed up as measurement ones, and no tracker fixes them — a dashboard tells you the mention rate is 12%; it does not raise it. Two of the levers above are exactly where Kompozy earns its place, and neither is generic "post more."
The first is entity consistency, the single thing that pulls a brand out of the "AI knows us but won't name us" trap. Kompozy governs every output through a [Persona Brief](/glossary/persona-brief) — one source of truth for how your brand is named, defined, and positioned — so the blog, the carousels, the text posts, the newsletter, and the on-camera video all tell an answer engine the same true thing about you, which is precisely the cross-source agreement that makes a model confident enough to cite. The second is format spread with video treated as first-class, which maps directly to the finding that video mentions correlate most strongly with AI visibility: from one brief Kompozy generates a [Persona Short](/glossary/persona-shorts) or Persona HeyGen where an avatar explains the topic on camera, alongside the rankable blog anchor and the liftable carousels and quote graphics — 18 formats in all.
[Autopilot](/glossary/autopilot) then schedules and fans that library across eight social platforms plus blog and email behind a per-post review gate, so the consistent, multi-surface presence the measurement said you lacked actually gets built and maintained rather than promised. Keep Profound, Otterly, or Ahrefs Brand Radar pointed at the scoreboard; Kompozy is what you point at the gap it exposes. Creator ($49/mo, 2,500 credits) suits a solo brand raising its mention rate on a focused query set; Pro ($299/mo, 18,000 credits) fits a team building presence across every channel weekly; Enterprise is custom for running AI visibility across many brands. The tool tells you the number; Kompozy is how you change it before the next cycle.
Frequently asked questions
What is a good AI share of voice or mention rate?
There is no universal pass mark, and a raw number is close to meaningless without a benchmark. One 2026 industry report put the average brand mention rate around 17%, so anything in that range is ordinary and the leaders sit far higher. Judge yourself two ways: against the competitors your own tracking surfaces in the same answers, and against your prior period. The goal is out-mentioning rivals on your priority queries and trending up over time, not hitting an arbitrary percentage.
What's the difference between mention rate and share of voice?
Mention rate — also called brand visibility — is the share of relevant AI answers that name you at all: your appearances divided by total answers. Share of voice is relative: your mentions as a fraction of every brand mention (yours plus competitors') across the same prompt set. Mention rate tells you whether the model knows to include you; share of voice tells you whether you are winning the category against the specific competitors it names instead of you.
Which AI engines should I track for brand visibility?
Cover the surfaces your buyers actually use. ChatGPT, Perplexity, Gemini, Google's AI Overviews and AI Mode, and Microsoft Copilot account for most of the volume; add Claude and Grok if your audience skews that way. Track each separately rather than blending them — they retrieve from different sources and will name different brands — and keep the engine list stable across cycles so your trend line stays comparable.
How do I actually get mentioned more in AI answers?
Work three layers. First, entity clarity — one consistent name and description everywhere, a claimed Knowledge Panel, and Wikipedia/Wikidata presence where you qualify, so the model is confident who you are. Second, off-site corroboration — mentions on Reddit, review sites like G2 and Capterra, and earned editorial that names your experts. Third, format spread, with video treated as first-class since YouTube mentions correlate most strongly with AI visibility. Publish extractable, consistent claims across all of it.
How often should I measure AI brand visibility?
Monthly is the standard cadence for a full read, run against the same frozen prompt set and the same engines so the numbers compare cleanly. A lighter weekly spot-check on your top few queries catches a large swing early. The one non-negotiable is consistency — same questions, same engines, same conditions — because with non-deterministic answers the signal lives in the trend across cycles, not in any single measurement.