AI visibility metrics: how to measure your AI visibility (the KPIs that matter)

The AI visibility metrics at a glance

Five metrics do most of the work, and they double as the AI search visibility KPIs a team can report on. Here is what each one tells you and how to measure it, before the sections below go deeper on each.

What is an AI visibility score?

An AI visibility score is a single number that summarises how often AI assistants name your business when customers ask them for a recommendation. Most tools build it from presence (the share of answers that name you), weighted by position or by how many assistants agree, and some fold in sentiment. There is no industry standard, so a 62 in one tool and a 62 in another do not mean the same thing.

Peekl deliberately reports a rate rather than a composite score: the share of 135 answers that named you, per assistant, with every answer behind it. A rate can be checked, compared month to month and read next to a competitor's. A blended score cannot, and it hides the reading that matters most, which is that a business can sit at 30 percent on Gemini and zero on ChatGPT while the average looks fine.

If a tool gives you a score, ask three things: how many answers it is built from, how many runs per question, and whether the per-assistant split is visible. A score on ten prompts asked once is a coin flip with a decimal point.

AI visibility tracking success metrics: what to report each month

If you report to a manager or a client, five numbers cover AI visibility tracking without a dashboard. Each row shows how the KPI is computed, what good looked like in our September 2026 scans of 504 local answers, and the reading that should trigger action.

What is presence, or mention rate?

Presence is how often an assistant names your business across many answers. If an assistant names you in 20 of 100 relevant questions, your presence is 20 percent.

Measure it by asking the same real customer questions several times, across assistants, and counting the share of answers that include you. A single answer is noise; the rate across many is the signal. In our own audits that means 15 questions asked repeatedly across ChatGPT, Gemini and Claude, 135 answers per business.

What good looks like depends on your market. Being named in a clear majority of relevant answers is strong; single digits mean AI rarely brings you up. In one of our own checks, an established clinic sat at 19 of 135 answers, about 14 percent.

Peekl reports this as your headline number, the share of answers that named you, with the per-question detail behind it.

How many questions do you appear in? (coverage)

Presence measures the share of all answers that name you. Coverage measures something related but distinct: across how many different questions, and buying intents, you show up at all.

A business can post a decent presence by being named often for one popular question. Yet it can miss entirely on the others customers ask. Coverage catches that blind spot.

Measure it by grouping your questions by intent (best in your area, near a landmark, for a specific need). Then check where you appear and where you never do. The empty intents point at the demand you are missing.

Good is broad, even coverage across the intents that matter to you, not a single question carrying your whole score. Peekl shows which questions name you and which name only competitors, so the gaps are obvious.

What is share of voice in AI search?

Share of voice in AI search is your share of the business mentions in the answers to your customers' questions: your mentions divided by your mentions plus your competitors', times 100. Presence asks how often you are named; share of voice asks how often it is you rather than a rival, when a business could be recommended.

To calculate it, take one set of answers, count how many times each business is named, and divide your count by the total. In the check above, assistants named the business 19 times and its top competitor 70. Against that one rival, its share of voice is 19 ÷ (19 + 70), or 21%. Add the next two or three competitors to the total and the share falls further, so good means first or second in your set rather than any fixed percentage.

Tools do not all calculate it the same way, so compare numbers within one tool, never across two. Peec AI divides your brand mentions by all brand mentions in the answers, yours included. Ahrefs Brand Radar counts the percentage of AI responses in your topic set that mention or cite you, against competitors. One counts mentions, the other counts answers, and the same business can score differently on each.

Peekl shows this as a you-versus-top-competitor split, plus the full ranked list of who gets named and how often. When one rival dominates that list, here is why AI recommends a competitor instead of you and how to compare yourself with it.

What tools measure share of voice in AI search?

Most AI visibility trackers, including the ChatGPT rank trackers we compared, report it. Checked on each vendor's own site in September 2026, share of voice is a named metric in Otterly, Peec AI, Ahrefs Brand Radar, the Semrush AI Visibility Toolkit, Profound, LLM Pulse and Rankscale, which breaks it down by assistant.

These are built for marketing teams watching a trend: a fixed list of prompts, run daily or weekly, drawn as a chart. If you want the number for your own trade and city, with the pages behind it, Peekl's $99 audit counts who is named across 135 answers on ChatGPT, Gemini and Claude and sets you against the business named most often. The AI visibility tools comparison covers what each tracker costs and which assistants it checks.

Which sources get cited, and what is your share?

Citations, sometimes called AI citations, are the websites an assistant draws on to build its answer: directories, best-of lists, review sites. Your source share is how much of that citation activity points to places that carry you.

Measure it by capturing the cited sources behind each answer and checking which ones list you and which do not. The gaps are your to-do list.

Good means the dominant sources in your market cite pages that include you. If the top few directories never mention you, that is often the whole reason competitors get named instead.

Peekl records the sources behind the answers and flags the ones that feed answers where you are absent.

How does AI describe you? (sentiment)

Sentiment is the tone an assistant uses about you when it names you: positive, neutral or flagged. Being named is good; being named with a lukewarm or cautionary description is a softer problem worth catching.

Measure it with sentiment analysis across many mentions, with a confidence on each read, rather than reacting to a single quote. It is a secondary signal and a decision aid, not fact.

Good is a consistently positive tone with enough mentions to be reliable. Peekl reports a sentiment read with real example phrases, and only when there are enough mentions to trust it.

What normal looks like: benchmarks from 504 AI answers

A metric only means something against a baseline, so here is one. In September 2026 we put six customer questions per assistant to ChatGPT and Gemini for seven local trades in six cities, and recorded every business each answer named: 504 answers, 807 business mentions counted city by city. The figures below describe what an ordinary local market looks like to an AI assistant right now. They are a one-month snapshot from a small number of questions per city, not a standard, and the detail behind each row sits on the trade's own page: dentists, aesthetic clinics, physiotherapists, plumbers, hair salons, personal trainers and real estate agents.

• Names per answer: about 1.6 businesses. An answer is a short list, and rarely more than three names long.
• Concentration: the most-named business in a city appeared in 33% of that city's answers on average, from 29% for hair salons and personal trainers to 42% for plumbers. Even the leader is missing from two answers in three.
• Agreement: ChatGPT and Gemini named at least one business in common in 22 of the 42 city-and-trade pairs, 52%. For the same question, the two assistants disagree about as often as they agree.
• Sources: booking platforms and marketplaces made up 29% of the top-cited sources, directories and review sites 16%, official and professional bodies 14%, media and community sites 9%. Individual business websites accounted for 31%, mostly national chains and a few practices whose pages answered one specific question well.

How to read your own numbers against these benchmarks

Start with presence. If you are named in more than a third of the answers in your city, you are the leader or close to it, because that is what the leader averaged across these trades. Being named in more than half, the bar our reports use for holding your market, was rare in every trade we scanned. Between roughly 10% and 33% you are on the list sometimes, the overlooked case, where the businesses ahead of you are simply better represented on the sources the assistants read. Below 10% you are effectively absent. That is where the London clinic in our audit sat with ChatGPT: 0 of 45 answers, despite a 4.9 rating and 189 Google reviews.

Then read share of voice, not just presence. In a fragmented trade a 25% presence can be first place, because nobody holds more, while in plumbing the same 25% is a distant second behind a franchise. Finally, read each assistant on its own. With agreement at 52%, a healthy blended number can hide a zero on the assistant your customers actually use, which is the single most common surprise in the audits we run.

How do the metrics differ by assistant?

AI visibility is not one number. One assistant can name the same business often while another ignores it, so a per-engine breakdown matters.

These assistants weight sources differently, so the same business can score well on one and barely appear on another. In one analysis, only about 11 percent of domains were cited by both ChatGPT and Google's AI answers for the same query, so strength on one engine tells you little about another. That is why an average across all three can mislead, and why the per-engine split matters.

Measure each assistant separately, then compare. A strong overall number can hide a zero on the one your customers use most.

Peekl breaks every metric down by assistant, so you can read your Gemini visibility next to your ChatGPT and Claude results, not blended into a single average.

AI visibility metrics for Gemini specifically

AI visibility metrics for Gemini are the same core metrics, presence, share of voice, citations and sentiment, read against Gemini on its own rather than a blended average, because Gemini behaves differently from ChatGPT and Claude.

The difference is where Gemini looks. As Google's assistant, it leans on Google's own signals, so your Google Business Profile, your Google reviews and your place on Maps tend to move your Gemini numbers more than they move your ChatGPT ones. A strong Gemini score next to a weak ChatGPT one is itself a useful reading: it usually means your Google footprint is solid but your wider web presence is thin.

The numbers bear this out. In our September 2026 scans, 378 answers per assistant across seven trades and nine cities, Gemini named 3.0 businesses per answer against ChatGPT's 2.2 and cited 3.8 web pages against 2.4. Its favourite business in a city took 57% of that city's answers, against 40% on ChatGPT, so Gemini is both more generous with names and more concentrated on a leader. It also reaches further into the long tail: 86% of the businesses it named appeared only once. Read on its own, that means a Gemini presence rate swings more from run to run than a ChatGPT one, and needs more repetitions before you trust it.

Industry data points the same way. SOCi's 2026 Local Visibility Index found Gemini recommended about 11% of 350,000 business locations, against 1.2% on ChatGPT, so a typical local business is roughly nine times more likely to be named by Gemini. A strong Gemini score next to a weak ChatGPT one is therefore the normal starting position, not a success.

So measure Gemini separately: run your customers' questions through it repeatedly, track presence and share of voice for Gemini alone, and note which sources it cites. Peekl reports every metric per assistant, so your Gemini visibility sits beside your ChatGPT and Claude results instead of being averaged away. To check what Gemini says about you step by step, see how to check if Gemini recommends your business, or run the free Gemini visibility check.

How to measure AI visibility, step by step

Measuring AI visibility is a repeatable procedure, not a one-off search. The five steps below produce numbers you can compare month to month and across assistants; every metric on this page comes out of them.

• 1. Write down the 10 to 15 questions your customers actually ask, in their words: "best [trade] near me", "a good [trade] in [area]", "[trade] for [a specific need]".
• 2. Ask each question in ChatGPT, Gemini and Claude, and repeat each one several times. Answers change from run to run, so repetition is what turns a guess into a measurement.
• 3. For every answer, record who was named, whether you were, and which sources the assistant cited. That single log gives you presence, coverage, share of voice and citations at once.
• 4. Compare your counts with the competitors that appear most, per assistant. The gap, and the sources behind it, is what tells you what to fix.
• 5. Repeat on a fixed cadence, monthly for most businesses, and watch the trend rather than any single number. You can run this by hand for a handful of questions, or let a tool do it at scale; see the best AI visibility tools compared for the options.

How do you track these metrics over time?

AI answers change from week to week, so a one-off snapshot ages fast. Tracking the metrics above over time turns a lucky or unlucky day into a trend you can trust.

The value of historical data is direction: are you named more often, are new sources starting to cite you, is a competitor pulling ahead? When you compare software for AI search optimization historical data, look for clean per-metric trends over time, not just a single current score.

Peekl's monthly tracking re-runs your check and charts presence and sentiment over time, and alerts you when you appear or disappear from answers. Once you can see the trend, how to improve your AI search visibility covers what to change.

How is this different from traditional SEO metrics?

Traditional SEO metrics answer a different question. Rankings, impressions, click-through rate and conversion rate all describe how you do on a page of links, and tools like Google Search Console report them well.

AI visibility metrics describe something those tools cannot see: whether an assistant names you in its AI generated answers, and how you compare there. You can rank well on Google and still be absent from the short list an assistant reads back.

So treat them as a separate layer, not a replacement. Keep watching your search metrics, and add these to see how you show up when customers ask an assistant instead of searching. For the practice behind these numbers, see being visible in AI search.

What about AI Overviews?

You may also see AI Overviews, the AI answers at the top of Google's results. Ranking in them follows similar logic to the metrics above: strong, well-sourced pages and consistent listings help, and there is no switch that guarantees inclusion.

It is a narrower, faster-changing surface than assistant recommendations, so treat it as one more place the same fundamentals apply, rather than a separate strategy. See how to appear in Google's AI Overviews for that surface specifically.

Common mistakes when tracking these metrics

A few mistakes make the numbers lie. Fixing them matters more than adding another metric.

• Measuring once. AI assistants vary run to run, so a single check is a snapshot, not a trend.
• Checking one assistant. A strong ChatGPT number can hide a zero on Gemini, where your customers may actually be.
• Using generic prompts. Real-world questions your customers ask reveal far more than textbook queries.
• Watching the score, not the sources. A number without the AI citations behind it tells you where you stand, not what to fix.

Which tools measure all this?

You do not need to track these by hand. A good tool runs the questions, records every metric, and shows the sources behind them. For an honest, side-by-side look, see the best AI visibility tools compared, with how to choose one.