What Is AI Visibility and How Do You Measure It?
Key Takeaways
- Measure brand mentions and links to your site separately because an AI answer can include one without the other.
- Use the same buyer prompts, engines, and counting rules when you compare results over time.
- Report mention rate, citation rate, share of voice, and answer position with their denominators and scope.
- Turn gaps in relevant answers into specific page updates or content briefs, then retest the same prompts.
AI visibility is how often and how prominently your brand or content appears in relevant AI-generated answers, through brand mentions or links to your pages. Your traffic dashboard cannot show every time an AI answer names your company. It also cannot tell you whether an answer recommends a competitor instead.
You can measure a useful sample of those answers. The goal is to find the buyer questions where you appear, where you are missing, and what you should publish next.
What AI visibility measures
A brand mention, a source link, and a visit are different events. An AI answer might name your product but cite a third-party review. It might link to your guide without naming your company. A buyer might read either answer without clicking. Keep those outcomes separate rather than rolling them into one visibility score. OpenAI says ChatGPT answers that use web search may include citations, and Pew Research Center found that clicks on source links inside Google AI summaries were uncommon in the searches it studied.
AI search visibility also differs from an organic ranking. Answers can vary between runs, and your place in a list of recommendations is not a stable search position. Measure repeated answers to questions your customers actually ask.
Four metrics to track
Use the same eligible answers for mention and citation rates. Define eligibility before you start. For example, you might count an answer when the engine returns a substantive response to a tracked prompt, whether or not it names any brands. Log unanswered prompts separately.
| Metric | Calculation | What to record |
|---|---|---|
| Mention rate | Answers naming your brand ÷ eligible answers | Count your brand once per answer, even if its name appears several times. |
| Citation rate | Answers linking to a page on your domain ÷ eligible answers | Count an answer once if it links to one or more of your pages. Record third-party links about you separately. |
| Mention-based share of voice across prompts | Answers naming your brand ÷ the sum of answers naming each brand in a fixed competitor set | Use the same prompts, engines, and period for every brand. Count each brand once per answer. |
| Position in the answer | Answers placing your brand in each position bucket ÷ answers naming your brand | Use mutually exclusive buckets: first named option, second or third named option, and fourth or later. Treat these as editorial buckets, not search ranks. |
State which brands you included when you report share of voice. Vendors may calculate it from mentions, citations, or weighted positions, so percentages with different rules are not comparable. LLM Pulse describes these differences.
Here is one hypothetical example. You collect 40 eligible answers for a fixed prompt set on one engine. Twelve name your brand, and six link to your domain. Your mention rate is 12 ÷ 40, or 30%. Your citation rate is 6 ÷ 40, or 15%.
Suppose two competitors appear in 18 and 10 of those answers. Your mention-based share of voice among the three brands is 12 ÷ (12 + 18 + 10), or 30%. An answer that names several brands contributes once to each brand’s total. These figures describe the 40 answers you tested on that engine during that period. They do not describe every conversation buyers have with it.
For position, review the 12 answers that name you. Report how many put you first, second or third, or fourth or later among named options. Divide each count by 12. When your team compares periods, keep those buckets and their denominator unchanged.
Choose prompts from buyer questions
Start with language you already hear. Pull questions from sales calls, demo objections, support tickets, on-site search, and Search Console queries. Rewrite each theme as a neutral question with a use case or constraint.
Group the prompts so you can see where you appear during a buying process:
- Category discovery: Which tools research, write, and publish B2B articles?
- Problem-solving: How can a two-person B2B team publish useful content every week?
- Comparison and alternatives: What should a team compare when choosing an AI content tool?
- Integrations: Which publishing tools connect to an existing CMS?
- Pricing: What should an agency budget for weekly client content production?
Add prompts about quality and trust, such as “How do AI content tools keep articles on-brand?” or “What should an agency check before publishing AI-written client content?”
Keep branded prompts in a separate group. “What does [your company] do?” can reveal an inaccurate description. An unbranded question tests whether the answer introduces you at all. Mixing the two can make your reported mention rate look better than your presence in broader buyer questions.
Start with 25 to 50 prompts if you need a manageable audit. That is a working range, not a scientifically guaranteed sample size. Keep a stable core for trend reporting, balance it across intent clusters, and version any additions rather than mixing them into the old baseline. Prompt-tracking guidance recommends starting with a focused set and expanding as you cover more segments and markets.
Run checks you can repeat
You can start AI visibility tracking in a spreadsheet. Use a dedicated tracker when you need to cover more prompts, engines, languages, or markets. Either way, save enough detail for someone else to check your work:
- engine, location, language, date, and search mode where applicable
- prompt and full answer, including an answer snapshot
- named brands and their position buckets
- source links, with your-domain links marked separately
- eligibility decision and scoring notes
Run priority prompts more than once. The same question can return different brands, wording, or citations on another run. Compare grouped results by engine and prompt cluster rather than changing a content plan because of one answer. OpenAI also cautions that search results and citations can be incomplete, outdated, or incorrect, so open cited pages before relying on them.
Do not combine engines into one trend unless you can explain the mix. Keep the prompts, settings, schedule, and scoring rules consistent within each comparison. If a Google AI Overview does not appear for a query, log that outcome rather than scoring it as an answer with no mentions. Google says AI Overviews do not appear for every query.
Turn answer gaps into pages worth publishing
Inspect the answer before you brief a page. Then inspect any pages the answer cites. Choose one change, publish it, and retest the same prompt cluster.
| What you see in tracked answers | What to do next |
|---|---|
| Competitors appear in a relevant unbranded cluster, but you do not. | Check what buyers ask and what your existing pages cover. Brief a focused use-case, comparison, or problem-solving page where you have evidence to add. |
| The answer names you but has no source links. | Check whether your product or documentation page answers the buyer’s question with clear, verifiable facts. Keep the mention in your report, but do not count a citation. |
| The answer names you but links only to third-party or competitor pages. | Check whether your product, pricing, documentation, or comparison page states the relevant facts clearly and whether crawlers can access it. |
| The answer links to your site but never names your brand. | Check whether the cited page clearly identifies your company and its role in the answer. Keep the citation and mention results separate when you report them. |
| The answer cites your page but describes your product incorrectly. | Correct or clarify the page, then check relevant external descriptions for outdated claims. Verify that the cited passage supports what the answer says. |
| One page earns citations across related questions. | Review which part of the page gets used. Extend coverage to adjacent buyer questions without duplicating the original page. |
Suppose your agency tracks questions about publishing client articles every week. Competing tools appear in the answers, while your site covers AI writing in general terms. Your team could brief a page on the actual workflow: research inputs, brand review, approval, and publishing to a client CMS. Add concrete examples only where you can support them. After publishing, rerun the same unbranded questions and compare the new answers with your saved set.
A new page does not guarantee a mention or citation. For Google’s AI features, Google says a page must be indexed and eligible for a Search snippet to appear as a supporting link. Google specifies no extra technical requirement for AI Overviews or AI Mode. Keep crawl access, internal links, visible text, and structured data consistent with what the page actually says. Do not treat JSON-LD, FAQ markup, or an llms.txt file as a shortcut to citations. Google says llms.txt is not needed for Google Search.
Use your other reports for different questions
Prompt checks show what a selected set of AI answers said. Your analytics tools cover other parts of the picture:
- Google Search Console: As of August 31, 2026, its dedicated generative AI insights had rolled out worldwide. The report includes impressions for AI Overviews and AI Mode, with page, country, date, and device views. Some properties may lack enough impressions to see it. It does not report the brand mentions and competitor comparisons in your prompt set. Search Console Help
- Bing Webmaster Tools: Its AI Performance preview reports citations, cited pages, grounding-query samples, and trends across supported Microsoft experiences. Those citation counts do not tell you where a page appeared in an individual answer. Microsoft Bing
- Website analytics and CRM: Check referrals, qualified leads, and conversions separately. A source link is not a visit. In a Pew Research Center study of U.S. Google searches from March 2025, users clicked a source link inside an AI summary in 1% of visits to pages with such a summary. That finding does not describe every AI product or predict your conversion rate.
Keep a stable set of buyer questions, review the answers, and publish where your coverage is thin. Evatype can handle the research, writing, and publishing work that follows a content decision. Retest the same questions to see what the answers say next.
Frequently Asked Questions (FAQ)
What is AI visibility?
AI visibility is how often and how prominently your brand or content appears in relevant AI-generated answers. You can track brand mentions, links to your pages, and where your brand appears in an answer.
How do you calculate AI visibility?
There is no single standard score. Start with mention rate (answers naming your brand divided by eligible answers), citation rate (answers linking to your domain divided by eligible answers), and a clearly defined share of voice across a fixed competitor set.
How many prompts should you track?
Start with 25 to 50 buyer questions as a manageable working set. Keep a stable core for trend reporting, and track new prompts separately until you establish a new baseline.
Does a citation in an AI answer mean someone visited my website?
No. A citation shows that the answer linked to a page, but it does not show whether anyone clicked. Check referral traffic and conversions separately.
What should you publish if competitors appear in AI answers and you do not?
Inspect the questions and the pages cited in those answers. If you can add useful evidence, publish or update a focused page that answers the buyer’s question, then retest the same prompt cluster.