How to Measure AI Visibility: Mentions, Citations and Share of Voice

To measure AI visibility, ask a fixed set of questions your customers ask on each AI assistant, on a schedule, and record whether each answer names your business, links to your site, and which competitors it names instead. From those records, track four numbers: mention rate, citation rate, share of voice and a weighted visibility score.

Below: each number with its formula, the traps that make AI visibility reports look better or worse than they are, our own scoring method as BlogAmplify runs it in October 2026, and a spreadsheet you can start with today.

What does AI visibility actually measure?

Every answer an AI assistant gives to one of your questions lands in one of four states:

1. Named. The answer mentions your business by name, often in a list of options.
2. Linked. The answer cites a page on your site as a source. This is a citation.
3. Left out. The answer names other businesses or cites other sites, and not yours.
4. No answer. The engine didn't produce an AI answer at all. Google often shows no AI Overview for a search.

Named and linked are different wins, and an answer can be both. Being named puts your business in front of the reader. Being linked means the engine used your page as evidence, which is the only one of the two that sends a click directly.

The fourth state matters more than it looks. A search with no AI Overview is not a loss: there was nothing to appear in. If you count it as "left out", your numbers fall every time Google decides not to show an overview.

The four numbers worth tracking

In each case, "answers given" means answers that exist. Leave out checks where the engine gave no AI answer, and always report the count next to the rate: "named in 14 of 40 answers" says far more than "35%".

Two more things are worth reading, though they don't reduce well to a number:

• What the answers say about you. Is the description accurate, current and fair? A wrong price or a discontinued product in an AI answer usually traces back to a page somewhere, often your own pages disagreeing with each other.
• Where the answers get their information. The sites AI cites for your questions (review sites, directories, media articles, competitors' pages) are the places worth being on. Our post on how to rank in ChatGPT covers what to do with that list.

How do you calculate AI share of voice?

Classic share of voice is your brand's portion of all the attention in a market: ad impressions, press mentions or search clicks. AI share of voice applies the same idea to AI answers.

The calculation:

1. Decide which brands you track: you plus the competitors your customers actually compare you with.
2. For every answer, note which tracked brands it names. Count each brand at most once per answer, even if the answer repeats the name.
3. Add up all the mentions of all tracked brands across all answers.
4. Divide your mentions by that total.

Example: across 50 answers, you are named in 12. Your four competitors are named 48 times between them. Your share of voice is 12 ÷ (12 + 48) = 20%.

Two cautions. First, share of voice depends on the competitor list: add a big brand and your share drops even if nothing about the answers changed, so keep the list stable and note when you change it. Second, share of voice and mention rate answer different questions. Your mention rate can sit at 24% while your share of voice is 20%, and both are correct. Don't put them on the same chart.

Our AI visibility score (our method as of October 2026)

Rates treat every answer alike. In practice, being named first in ChatGPT for a buying question matters more than being named fifth in a smaller engine for a side question. Our score weights for that. This is the method BlogAmplify's AI visibility tracking runs as of October 2026; we will publish changes when we make them.

Step 1: score each answer

Position counts only the brands you track (you plus your listed competitors), in the order the answer first names them. So an answer that lists four brands you don't track and then you still counts as naming you first.

Being named at all is worth most of the value (0.5). Position adds at most 0.3, because the order of a list is the least stable part of an answer (more on that below).

Step 2: weight engines by reach

Each engine's answers are weighted by an estimate of how many people use it. These are our October 2026 estimates, not measured market shares:

The weights are re-scaled over the engines that actually answered. A question's score is the weighted average of its answers, times 100.

Worked example. One question, five engines:

• ChatGPT names you first: 0.8
• Google shows no AI Overview: not counted
• Gemini leaves you out: 0
• AI Mode links to your site without naming you: 0.3
• Perplexity names you fourth and links to you: 0.5 + 0.2 = 0.7

Weighted sum: 0.30 × 0.8 + 0.15 × 0 + 0.10 × 0.3 + 0.08 × 0.7 = 0.326. The engines that answered weigh 0.30 + 0.15 + 0.10 + 0.08 = 0.63. The question scores 0.326 ÷ 0.63 = 0.517, or 51.7 out of 100.

Step 3: weight questions by importance

Your overall score is the average of your question scores, weighted by how much each question matters. Importance rises with the question's intent (buying and local questions count most, then category, comparison and problem questions) and with how often people ask about its subject in AI search. In practice, the most important questions count roughly twice as much as the least important ones.

Why are questions that name you scored apart?

A question like "Is [your business] worth it?" or "[your business] vs [competitor]" gets answered with your name almost by definition. Mix those into your score and you measure the question, not your visibility.

We flag any tracked question that names the business and score it separately. It's left out of the main score, the trend, the mention and citation rates, the per-engine figures and share of voice. Those questions still count for the things they're good at: what AI says about you, which sources it uses, and whether it links to your pages.

If you track by hand, do the same: put branded questions on their own tab.

How do you read a trend without fooling yourself?

The trap: you add five new questions this month, three of them easy, and your score goes up. Nothing about your visibility changed; the question set did.

The fix is a like-for-like change. Compare this check with an earlier one only over the questions both checks scored. Questions you added or retired since then don't move the trend, and branded questions stay out of it.

Keep your question set mostly stable for the same reason. In our tracking, the monthly question research changes at most a fifth of the questions the system chose, and never retires questions the owner pinned or questions an article is currently targeting.

Samples and noise: why one answer proves nothing

An AI answer is one sample from many possible answers, not a record you look up. In a study reported by Search Engine Journal, SparkToro and Gumshoe ran 2,961 prompts and found that ChatGPT and Google's AI Overviews each returned the same list of brands less than 1% of the time across repeated runs of the same prompt, and that the same list in the same order appeared less than 0.1% of the time.

What that means for measurement:

• Rates over many answers are meaningful; one answer isn't. Being named in 30% of answers is a real signal. Being named in today's answer is a coin toss.
• Position is the noisiest part. That's why our score gives position a small bonus and gives being named the bulk of the value.
• Group questions. Report by topic group (for example "pricing questions" or "comparison questions") as well as overall, since single-question results bounce.
• Check on a fixed schedule. Same day, same time, same questions, in your customers' country and language.

What should count as a win?

Rates tell you how often; a win tells you whether a specific gap closed. We define a win per question, because that's the level at which you act (write an article, fix a page, get listed somewhere).

As of October 2026, our tracking works like this:

1. Visible means named or linked on at least half of the engines that answered, and on at least one.
2. A goal opens when competitors are named or other sites cited and you're not visible. An important question (a high-intent question people ask often in AI search) also opens one when you're missing on ChatGPT or Google AI Overviews while that engine names a competitor or cites another site, even if other engines name you.
3. A win needs the gap closed in two checks in a row. For an important question, you must also no longer be missing on ChatGPT or AI Overviews where they name a competitor or cite another site. One lucky answer never counts as a win.
4. A won question that slips is watched. Not visible on the latest check marks it as being defended. Not visible on two checks in a row reopens the goal, at a higher priority.

If you track by hand, borrow rule 3 even if you skip the rest. It removes most of the false celebrations and false alarms.

What do Google and Bing show you for free?

Neither search engine will tell you whether an AI answer named your business, but both now show when your pages appear in their AI features.

• Google Search Console. Google says pages appearing in AI Overviews and AI Mode are included in the overall search traffic, reported in the Performance report within the "Web" search type. Since June 2026 there is also a Search Generative AI report that shows how often your URLs appeared in those features, by page, country, device and date. It reports impressions, and Google says it has been available to all websites worldwide since August 31, 2026. See our guide to connecting Google Search Console.

• Bing Webmaster Tools. The AI Performance report, introduced as a public preview in February 2026, counts how often your pages are cited in Microsoft Copilot, AI summaries in Bing and some partner integrations. It shows total citations, average cited pages per day and the grounding queries the AI used to find your content. Bing's announcement lists no click data. Our guide to Bing Webmaster Tools covers connecting it.

• Google Analytics. Visits that arrive from AI assistants have their own AI Assistant channel. Google's channel definitions describe it as visits "from sources like ChatGPT, Gemini, Deepseek, Copilot, or Grok" and say it excludes Google's AI Overviews and AI Mode, so visits from those don't show up there. To see these visits:

  1. Open Reports, then Acquisition > Traffic acquisition.
  2. In the table grouped by channel, find the AI Assistant row.
  3. To see which assistant sent each visit, switch the table's dimension to Session source / medium and search the table for "ai-assistant", the medium Google gives these visits. Add Landing page as a secondary dimension to see which pages they arrive on.
  4. Google gives examples, not its full list of assistants. If one you care about, such as Perplexity, isn't under AI Assistant, search the same report for its domain (perplexity.ai).

  The numbers are usually small, but they are visits you can tie to a page.

One check before any of this: make sure the engines can read your site. OpenAI, for example, says OAI-SearchBot is used to surface websites in ChatGPT's search features, separately from GPTBot, so blocking OAI-SearchBot in robots.txt opts your site out of being surfaced in ChatGPT search.

These reports cover your pages, not your name. Use them alongside question tracking, not instead of it.

How to measure AI visibility with a spreadsheet

You can start today with a spreadsheet and an hour a week.

1. Write 20 to 30 questions. Use the words a customer would type into a chat box, not keywords. Mix four kinds:

• Category: "What are the best [category] for [audience]?"
• Comparison: "[Competitor A] vs [Competitor B]: which is better for [need]?"
• Problem: "How do I fix [problem your product solves]?"
• Buying: "How much does [service] cost in [city]?"

Add a few questions that name you on a separate tab.

2. List the competitors you track. Five is plenty. Keep the list fixed.

3. Ask each question on each engine. Use a logged-out or fresh session so past chats don't steer the answer, and ask from the country your customers are in.

4. Record one row per answer. Position is your place among the tracked brands the answer names, in the order it first names them.

5. Calculate weekly:

• Mention rate: rows with Named = Yes ÷ rows with AI answer = Yes
• Citation rate: rows with Linked = Yes ÷ rows with AI answer = Yes
• Share of voice: your Named count ÷ (your Named count + every competitor mention in "Others named")
• Optionally, the score: apply the answer values and engine weights above with a lookup column

6. Compare like for like. When you add or drop a question, compare weeks only on the questions both weeks share.

7. Act on the gaps. Sort by questions where competitors are named and you're not, look at which pages the answers cite, and decide: write an article that answers the question directly, fix a page of yours that says something different, or get listed on a page the answers already use.

The spreadsheet becomes tedious past a few dozen questions and a weekly check. That's the point where tracking software earns its keep. In BlogAmplify, the Growth + AI Visibility plan asks up to 50 questions on ChatGPT, Gemini, Perplexity, Google AI Overviews and Google AI Mode every day, scores them with the method above, and plans articles for the questions where AI leaves you out. Our methodology page explains how those articles are researched and written.