AI Visibility Metrics Explained: 7 to Track in 2026

AI visibility metrics show how often AI answers in ChatGPT, Perplexity and Google AI Overviews mention your brand, cite your pages or send visitors. The 7 that matter are mention rate, citation rate, share of voice, position, sentiment, AI referral traffic and prompt coverage.

Updated November 2026. This guide is for marketers, SEO leads and founders who want to know whether AI engines name their brand, and which of the numbers to trust.

AI visibility is whether AI engines such as ChatGPT, Perplexity, Gemini and Google AI Overviews include your brand when a buyer asks a question. A metric turns that into a number you can follow from week to week. In this guide a prompt is one question put to an AI engine, an engine is the assistant that answers it, and an answer is one reply from one engine.

Choosing the software that reports these numbers is a separate job. For that, see the best AI visibility tools. This guide explains what each number means, how it is counted and when it misleads.

Which AI visibility metrics matter most?

Mention rate, citation rate and AI referral traffic matter most, 3 of the 7, because they show whether AI names you, links to you and sends visitors.

Share of voice adds the competitor view once you have named the rivals to track. Position, sentiment and prompt coverage explain why the first numbers moved.

What does each of the 7 AI visibility metrics mean?

Each of the 7 metrics is a ratio or a count taken from the same pile of AI answers to your tracked prompts.

Mention rate

Mention rate is the share of AI answers that name your brand. Divide the answers that name you by all the answers you ran, then multiply by 100. A brand counts once per answer, however often the answer repeats the name.

Example: 40 prompts run on 3 engines make 120 answers. If 30 of them name you, your mention rate is 25%.

When it misleads: a prompt set full of questions that already contain your brand name pushes the rate up. Keep a mix of branded questions and category questions such as "best CRM for a small team". Tools also count differently: in the Semrush AI Visibility Toolkit a mention is a prompt that includes your brand, not each repeat of the name. Source: Semrush Knowledge Base.

Citation rate

Citation rate is the share of AI answers that link to a page on your domain as a source. A mention is your name in the text. A citation is your link in the source list. The two are not the same thing.

In one study of 3,981 domain appearances across 115 prompts and four engines, 61.7% were cited without being named, 13.2% were cited and named, and 25.1% were named without a link. Source: Semrush (June 2026).

Example: if 18 of your 120 answers link to your site, your citation rate is 15%. Track it beside mention rate, because one can rise while the other stays flat. Also note which of your pages get cited, since that is the page AI already trusts.

Share of voice

Share of voice compares your visibility with the rivals you track. Two formulas are common, and they give different numbers for the same answers.

Example: you and two rivals, B and C, are tracked across 5 answers. Answer 1 names you and B. Answer 2 names you. Answer 3 names B. Answer 4 names you, B and C. Answer 5 names nobody. Counting mentions, you have 3 of 7, which is 43%. Counting answers that name any tracked brand, you are in 3 of 4, which is 75%.

A third method weights each prompt by how often people ask it, so a share built on rare questions falls. Ahrefs uses that weighted method for its AI Share of Voice. Source: Ahrefs (October 2026). Write down the prompts, engines, rivals and dates behind any share of voice, because change any one and the percentage changes.

Position

Position is where your brand or your link appears inside an answer. In a list answer it is your place in the list. In a source list it is the order of the citations.

Some tools average it across tracked prompts. The Semrush Average Position metric is where a citation of your domain typically appears, for example first or second. Its Visibility view runs from 0%, never among the top citations, to 100%, the first citation on every tracked prompt. SE Ranking also collects average positions inside generated answers.

When it misleads: many answers are paragraphs with no list, so there is no place to hold. Answers also change between runs, so position moves too. Use it for lists and first citations, and never compare it with a Google ranking.

Sentiment

Sentiment is how your brand is described in AI answers: positive, neutral or negative. Tools sort each mention into those groups. BrightEdge AI Catalyst gives the reason for each label, Similarweb lets you filter AI mentions by the three groups, Semrush gives an overall balance and the themes behind it, and Moz AI Explorer includes sentiment analysis.

Example: 25 mentions, with 18 positive, 6 neutral and 1 negative, is 72% positive.

When it misleads: a label does not check facts. An answer can be positive and still give a wrong price or an old feature. Read the answers behind the label, and use brand monitoring in AI answers to catch the wrong ones.

AI referral traffic

AI referral traffic is the visits that arrive from AI assistants. It is the one metric of the 7 counted from real visits. In Google Analytics 4 the AI Assistant channel groups visits from sources such as ChatGPT, Gemini, DeepSeek, Copilot and Grok, and it leaves out Google AI Overviews and AI Mode. Source: Google Analytics Help.

Google AI Overviews and AI Mode have their own view. The generative AI performance report in Search Console counts impressions: how many times links to your site were shown in those two features. Source: Search Console Help. Similarweb, Ahrefs (through Web Analytics) and the Semrush Traffic & Market Toolkit also report AI traffic.

When it misleads: it undercounts. A buyer who reads your name in an answer and types your address a day later lands as direct traffic. Treat the figure as a floor.

Prompt coverage

Prompt coverage is the share of your tracked prompts where you appear at all, as a mention or a citation, on at least one run. Mention rate counts every answer. Coverage counts each prompt once.

Example: you track 40 prompts and appear on 18 of them. Your coverage is 45%, even if your mention rate is 25%.

Split coverage by buyer stage: research questions, comparison questions and decision questions. BrightEdge maps prompts to Research, Consideration and Decision for this reason. When it misleads: a coverage figure only means something against a fixed prompt set, because adding prompts mid-period inflates mentions without any real gain. Source: Semrush (May 2026).

How accurate are AI visibility metrics?

AI visibility metrics are directional: six of the 7 are samples of AI answers, and AI answers change from run to run.

Four things move a number without any change in your real visibility:

• Answers vary. The same prompt can name different brands on two runs, so one answer is an anecdote. Groups of prompts, run more than once, give a steadier number.
• The prompt set is a sample. The number of possible questions is effectively infinite, so every tool tracks a slice of the likely ones.
• Some numbers are modeled. Estimated impressions weight mentions by search demand, which is a modeling choice, and the figures are directional indicators, not exact traffic counts. Source: Ahrefs (February 2026).
• Definitions differ. Two tools can report a different share of voice for the same brand, and both can be right, because the prompts, engines and rivals behind the number differ.

Three habits make any of these numbers more reliable. Fix the prompt set for the period. Start a new baseline whenever you change the prompts, the rivals or the engines. Write the scope beside every figure, for example "24% across 500 prompts on two engines against three rivals in August".

Treat a move of a few points as noise until it holds for three or four checks in a row. The one exact number is the count of visits in your analytics, so use it to check the direction of the sampled ones.

Which AI visibility metrics software is the most accurate?

The most accurate AI visibility metrics software is the one whose method you can check, on 4 points: prompts, runs, engines and the answers behind each number.

Check these before you trust a dashboard:

• Prompts: can you set your own, and how many does the plan track?
• Runs: how often is each prompt run, and how many runs make one reported result?
• Engines: are the engines your buyers use in the plan you are weighing?
• Answers: can you open the actual answers and cited sources behind a number?

Five platforms and what each has for the four points, as of 5 November 2026.

In BrightEdge a report can include the AI responses, Botify has prompt-level insights, and seoClarity ArcAI has competitor benchmarks and topic gaps. Whatever you pick, test it yourself: run 10 prompts by hand in each engine, note whether you are named, and compare with the tool for the same prompts on the same day. Exact matches are not realistic because answers vary, so look for the same direction.

How do you find AI search visibility gaps?

A gap is a prompt where AI names your rivals and does not name you, listed as Missing Topics and Prompts in Semrush.

Use these six steps:

1. Write 20 to 40 prompts across research, comparison and decision questions.
2. Run them on the engines your buyers use, more than once each.
3. Mark every prompt: you named, a rival named, or nobody named.
4. Group the misses by topic and by buyer stage.
5. Open the sources cited in the missed answers. These are the pages AI reads about your rivals and not about you.
6. Fix the pages or earn the mentions, then re-run the same prompts. Do not add prompts until the period ends.

Several platforms automate the middle steps. Semrush has Topic Opportunities, Weak and Missing Topics and Prompts, and Missing Sources. BrightEdge flags the moment a competitor takes over an answer you held. SE Ranking marks where rivals appear and you do not, and seoClarity covers topic gaps. For the fixes, see the steps that improve these numbers.

Which tools report these AI visibility metrics?

Eight established platforms report these metrics: Semrush, Ahrefs, BrightEdge, seoClarity, Botify, Moz, Similarweb and SE Ranking.

Google Search Console and Google Analytics 4 are free and cover the one counted metric, AI referral traffic, plus Google's own AI impressions. Prices and plan details for Semrush, Ahrefs, SE Ranking and BrightEdge are on the AI visibility tools hub.

Which AI visibility metrics should a small team start with?

A small team should start with 3 metrics, mention rate, citation rate and AI referral traffic, using 20 prompts, 2 engines and a weekly check.

1. List 20 prompts a buyer would ask, half with no brand name in them.
2. Run them by hand each week on ChatGPT and Perplexity, and log named, linked or neither in a spreadsheet.
3. Open the AI Assistant channel in Google Analytics 4 reports and the generative AI report in Search Console, which appears once a site has enough AI impressions.
4. After 4 weeks, name 3 rivals and add share of voice. Add position, sentiment and coverage when you need to explain a change.
5. Move to software when 20 prompts become 100 and running them by hand stops being practical.

For a monthly report, four KPIs are enough: mention rate, citation rate, AI referral sessions and share of voice. Put the prompt count, the engines and the rivals beside each one. A free first reading comes from a free check of how ChatGPT and Perplexity answer about you.

Frequently asked questions

In short: which AI visibility metrics should you track?

Start with 3 of the 7 metrics, then add the rest as your questions get harder.

• Track mention rate and citation rate together, because an answer can link to you without naming you.
• Add share of voice once you have named 3 rivals, and write the scope beside the number.
• Use AI referral traffic as the one counted figure to check the direction of the others.
• Treat position, sentiment and prompt coverage as the numbers that explain a change.
• Fix the prompt set for the period, and read a few points of movement as noise until it holds.

Definitions and tool features are as of November 2026. Tools rename and add metrics, so check the current names before you buy. For the platforms behind these numbers, see AI visibility software compared and the free Appscribed checker.