How to Measure Your Brand's AI Visibility: Prompts and Metrics

To measure your brand's AI visibility, put the same buyer questions to each AI engine on a fixed schedule, run every question repeatedly, and score whether each answer names you, where, and beside which competitors.

Search Engine Land describes the target as how often and how credibly your brand appears in AI responses. You can run the method below with a spreadsheet, and OnlyAEO's managed service includes monthly AI visibility analytics.

What does AI visibility measure?

AI visibility measures whether AI engines name your brand when buyers ask about your category. The headline metric is citation share: the percentage of answers to your fixed prompt set in which an engine names your brand.

Citation share doesn't tell you how you were named. As Aperture Insights notes, a brand can show up in nearly every relevant response and still lose, because the answer recommends a competitor.

Report every metric per engine. A blended average can look fine while you're missing entirely from a single engine.

Why can't Google Analytics and Search Console measure it on their own?

Your analytics record visits to your site and Google's own search surfaces. Most AI visibility happens inside answers that send no click, on engines Google does not report on.

What Search Console shows

Search Console counts appearances in AI Overviews and AI Mode inside the regular Performance report, under the Web search type, mixed in with the rest of your search data.

Google's Generative AI performance reports add dedicated views of your impressions within generative AI features on Search and Discover. They say nothing about what ChatGPT, Claude, Perplexity or DeepSeek tell buyers about you.

What Google Analytics shows

Google Analytics sees clicks. OpenAI's publisher FAQ says ChatGPT automatically adds a UTM source parameter, chatgpt.com, to referral URLs, so visits from ChatGPT search arrive with a named source.

But referral metrics only measure the last step, the click, as Search Engine Land puts it.

Pew Research Center found that Google users who saw an AI summary clicked a traditional search result in 8% of visits, against 15% when no summary appeared.

A buyer can read your name in an answer and visit days later, and nothing links that visit to the answer. Track the clicks you can attribute, as in tracking AI referral traffic in GA4 through to pipeline, and measure answers directly.

How do I build the prompt set?

Build the prompt set from the questions your buyers actually ask, across personas and buying stages, then lock it with a version label.

Aperture Insights suggests starting a baseline panel with 20 to 50 queries that reflect customers' real questions across their main markets and personas.

• Write for each persona. A head of engineering asks different questions from a procurement lead, so tag every prompt with its persona.
• Cover every buying stage. HubSpot describes awareness-stage prompts as category-level, consideration-stage prompts as comparative, and decision-stage prompts as specific.
• Mix branded and unbranded prompts. Use both, and include head-to-head comparisons against the competitors you track, as AirOps recommends in a LinkedIn post on measuring AI search visibility.
• Leave out vanity prompts. Asking the engine to describe your company always gets a mention and says nothing about whether a new buyer would find you.
• Freeze and version it. Record each prompt's wording, persona, stage and version. When you change the set, publish a new version and mark the break on the trend line.

Which AI engines should I measure, and how do I run the prompts?

Measure every engine your buyers use, and record for every answer whether the engine searched the web. HubSpot advises running each prompt through each engine separately, in logged-out or temporary sessions, so personalisation does not skew results.

1. Open a clean session. Use a logged-out or temporary chat.
2. Paste the prompt exactly as worded in the current prompt set version.
3. Flag the search. Note whether the engine searched the web, and which URLs it cited.
4. Save the answer word for word with prompt, engine, date and prompt set version, so you can re-score it later.
5. Repeat in a new session until you reach your run count.

The search flag matters: a searched answer can cite your pages, while an unsearched answer reflects older material. Without it, a jump in citation share might only mean the engine started searching.

Why does a single AI answer prove nothing?

A single AI answer tells you about that answer and little else, because engines answer the same prompt differently almost every time.

SparkToro tested this at scale: 600 volunteers ran 12 prompts through ChatGPT, Claude and Google's AI a combined 2,961 times.

If you ask ChatGPT or Google's AI the same prompt 100 times, SparkToro put the chance that any two responses give the same list of brands at less than one in 100. Order is less stable still: SparkToro estimated it is more like 1 in 1,000 runs before two lists come back in the same order.

How many runs is enough?

• For a single prompt's brand list. SparkToro concluded that to know an AI's set of recommendations for a single prompt you usually need to ask at least 60 to 100 times and average the results.
• For a practical routine. HubSpot advises capturing responses 3 to 5 times per engine per prompt, with a new temporary chat each time.

Our reading: citation share across a whole prompt set samples prompts multiplied by runs, so it needs fewer runs per prompt. Report the average of repeated runs, and treat a small move as noise until it holds into the next period.

How do I score each AI answer?

Score every answer with the same written checklist, so any reviewer records the same result.

1. Named? Yes or no.
2. Where? At the start, in the middle, at the end, or only in a link list.
3. Recommended? A recommendation or a passing mention.
4. Accurate? Quote any error about you.
5. Competitors? Which are named, and in what position.
6. Searched? Whether the engine searched the web, and which URLs it cited.

Position matters because being named and being picked are different results. In SparkToro's data, City of Hope hospital in Los Angeles appeared in 69 of 71 ChatGPT answers, a 97% visibility rate, yet was the top mention in only 25 of them.

Then tag each mention by position. Every tag counts toward citation share, but only hero counts toward hero citation share.

We recommend reporting hero citation share as its own number, since a board can read it without a legend. If you use a weighted score internally, never change the weights mid-trend.

How do I benchmark competitors on the same prompts?

Score named competitors on every answer in the same run, with the same checklist, so the numbers compare like for like.

Share of voice follows Search Engine Land's formula: your citations divided by total citations in the competitive set, multiplied by 100.

Without a competitor line, a loss can look like progress, because your citation share can climb while a competitor's climbs faster. The figure the board needs is the gap to your closest tracked competitor, in percentage points, per engine.

The detailed method is in benchmarking AI citation share against a named competitor.

How often should I measure AI visibility?

Run the full, fixed prompt set every month. Add a lighter weekly check only if you need early warning on a handful of prompts.

• Monthly for the trend. HubSpot calls monthly a baseline for spotting trends, with a fixed prompt set on the same cadence so movement stays comparable.
• Weekly for early warning. Aperture Insights suggests that for most brands a weekly baseline check is a good rhythm. Use it on the prompts closest to revenue to catch a sudden drop or a wrong fact, and keep it off the trend line.

What goes in the monthly AI visibility report?

A monthly AI visibility report shows your core metrics per engine against last month, the gap to your closest tracked competitor, and the prompts behind every change.

• Headline metrics: you and each tracked competitor, per engine, against last period.
• Prompt-level detail: every prompt, every run, the raw answer kept word for word, and its score.
• Source citations: the URLs engines cited, yours and others'.
• Accuracy log: wrong statements, quoted, with engine and date.
• Prompt set version: the version used and any changes since last month.
• Actions: the prompts you lost, ranked by buyer value, with the page or listing that should answer each.

The board slide can stop at citation share, hero citation share and the gap to your closest tracked competitor. What counts as good depends on your category, as covered in good AI citation share benchmarks and targets.

In OnlyAEO's New York edibles retailer case study, measurement tracked the brand from 1.5% AI visibility, ranked #50, to 25% visibility, #4 of 374 brands, and showed that its own website had become the most-cited source in the category, ahead of New York State's official site.

Should I measure it myself or use a tool or service?

You can measure AI visibility yourself with a spreadsheet and temporary chat sessions. A tool or service is worth paying for when prompts multiplied by engines multiplied by runs exceed the answers your team can read and score each month.

For a first pass, running an AI visibility audit in an afternoon walks through the manual version.

What to ask any tool

Any tool has to sample engine outputs much as you would by hand. Google states that no third-party tool has access to its internal ranking or AI systems. Check for:

1. Engine coverage, with each engine reported separately.
2. Run count per prompt per engine, and how runs are averaged.
3. Raw answers kept word for word and exportable.
4. Position tags separating hero, supporting and trailing mentions.
5. Competitors scored on the same prompts in the same runs.
6. A versioned prompt set you control, built from your buyers' questions.
7. A search flag and the URLs each engine cited.

What OnlyAEO provides

OnlyAEO is an Answer Engine Optimization (AEO) and Generative Engine Optimization (GEO) agency. The free AI visibility audit is delivered within 48 hours and covers ChatGPT, Claude, Gemini and DeepSeek.

The managed service includes monthly AI visibility analytics alongside researched, fact-checked articles. Every month a fresh report shows where the brand is named, and the client and OnlyAEO review what worked and plan the next month.

The method is laid out in how OnlyAEO measures AI visibility.

How we tested this guide

On 13 October 2026, OnlyAEO's content engine put this article's main question to ChatGPT and Claude. ChatGPT answered without searching the web.

Claude ran a web search and cited trade-press guides and AI visibility vendors' blogs. Neither answer named OnlyAEO.

A single run like this proves little, which is why this method repeats every prompt.

Frequently asked questions

Should I count the links in AI answers or the brand names?

Count the names first. Pew Research Center found that Google users who saw an AI summary clicked a link inside the summary itself in just 1% of visits to pages with a summary.

How common are AI answers in Google search?

Pew's analysis of the browsing data of 900 U.S. adults found that 58% ran a Google search in March 2025 that produced an AI-generated summary.

Do mentions on other websites affect AI visibility?

They move together. Ahrefs' brand visibility study found that branded web mentions correlated highly with AI visibility, at 0.66 to 0.71, across ChatGPT, AI Mode and AI Overviews. Correlation does not prove cause.