How to Measure Presence and Accuracy in AI Search
Summary
Presence and accuracy are the first two stages of measuring AI search. Presence tells you whether AI includes your brand when buyers ask about your category. Accuracy tells you whether AI describes your brand correctly once it does. Each stage has one metric for leadership and one for the team that does the work. @Why Referral Traffic Is the Wrong Measure for AI Search explains why these stages replace referral traffic as the measure, and @How to Measure Selection and Pipeline Impact from AI Search covers the third stage.
Why the answer itself matters
Buyers often act on the AI answer without visiting any source. A Citation Labs usability study of Google AI Mode, published in 2026, followed 48 US participants through 185 major-purchase tasks and found that participants clicked nothing in 64% of AI Mode tasks. The study describes its results as directional rather than population-level, but the pattern is clear enough to plan around: what the answer says about you is, for many buyers, the only thing they read.
Presence metric for leadership: share of voice
Share of voice measures how often AI names your brand.
How to measure it: answers that name your brand, divided by all tracked answers to your buyers' questions.
Leadership already understands share of voice from media and advertising, so it needs no new explanation. Report it against your three to five closest competitors, because the number only has meaning relative to the brands buyers are comparing you with.
Presence metric for the team: fan-out inclusion rate
When an AI model answers a question, it often runs several background searches to gather sources. This is called query fan-out. Some AI tools show these searches. If the model searches for your brand by name, for example your pricing page, your brand was on its list before it read a single page.
How to measure it: fan-out searches that name your brand, divided by all fan-out searches behind your buyers' questions.
This metric tells your team whether the model considers you a candidate at all. It is limited to tools that expose their fan-out searches, so treat it as directional.
Accuracy metric for leadership: message pull-through
Message pull-through measures whether AI tells your story in your terms.
How to measure it: answers that repeat your key messages, divided by answers that mention your brand.
AI models often describe a brand in the language of review sites, comparison articles, and forums rather than the brand's own positioning. A low pull-through rate means third parties are writing your description, and leadership should know that before it approves messaging spend that AI is not repeating.
Accuracy metric for the team: brand accuracy
Brand accuracy measures whether AI gets the facts right.
How to measure it: answers that state your facts correctly, divided by answers that mention your brand.
Check the facts buyers use to decide: pricing, plan limits, features, integrations, contract terms, and who the product is for. An error in any of these can remove you from a shortlist, and the fix is usually on pages you or third parties control, such as pricing pages, documentation, and review profiles.
How to collect the data
Three practices keep these metrics reliable.
1. Track real buyer questions
Use the questions buyers actually ask, with their constraints included, such as company size, budget per seat, and required features. A generic category term such as "best CRM software" produces a generic answer that does not reflect what your buyers see.
- Collect questions from sales call recordings and notes, CRM records, support tickets, and the longer, question-style queries in Google Search Console. These sources show the exact wording and constraints buyers use.
- Rewrite each one as a full prompt in the buyer's voice, for example: "I run marketing at a 40-person B2B company. What is the best CRM under $100 per user per month that includes marketing automation?"
- Group the prompts by buying stage, such as early research, vendor comparison, and final decision. Your brand may appear often in one stage and rarely in another, and the grouping shows where to focus.
- Keep the core list fixed from month to month so the results can be compared. Add new questions as a separate set rather than swapping them into the core list.
2. Run each question several times a month
AI answers vary from one run to the next, and a single run will most likely undercount the brands a model names. Repeated runs give you a stable picture of who appears and how often.
- Use the same prompts in the same AI tools on a set schedule, such as weekly.
- Run each prompt in a fresh session or logged-out window where possible. Saved memory and chat history can change the answer, and buyers will not share your history.
- Log each run in a spreadsheet with one row per run: date, tool, prompt, full answer text, brands named, the brand picked if any, facts stated about you, and sources cited.
- Calculate each metric for the month as a share of all runs. For example, if your brand appears in 12 of 40 runs, your share of voice for that prompt set is 30%.
3. Read the results as direction, not as a score
The useful question is whether each metric moved up or down since last month, and against which competitors. A single month's number on its own says very little.
- Calculate the same four metrics for your three to five closest competitors, so you can see whether a change is yours alone or shared across the category.
- Compare each month with the month before and with the same month in the prior quarter, which smooths out short-term swings.
- Note any AI model updates during the month. A sudden shift may come from a change in the model rather than from your work.
- Match each change against the work your team shipped, such as updated pricing pages, new documentation, or corrected review profiles. This shows which actions moved the metrics and which did not.
Related
- @ChatGPT Query Phase Flow: how ChatGPT moves through the phases of answering a query.
- @ChatGPT Cross-Checks Your Site Against Your Reviews: why third-party reviews shape how AI describes a brand.
- @This Is Not a Permanent State. AI Models Change.: why model updates shift AI search results over time.
- @Personalization vs. Customization in Search: how user history and settings change what a searcher sees.
Sources
About the author
@Ryan Edwards is the founder of @CAMINO5, an SEO and AI search advisory firm based in Burbank, California. He has worked in search for more than 20 years.
