How to Measure AI Perception: 6 Metrics Beyond Mentions and Citations in 2026

Mentions and citations tell you whether AI names your brand. They don't tell you if AI recommends you, describes you correctly, or keeps citing you. Here are six metrics that measure AI perception, with formulas and tool picks.

Key takeaways

  • Mentions and citations measure presence. AI perception is about position, tone, accuracy, and whether the visibility lasts.
  • Six metrics cover it: position in answer, share of recommendations, narrative accuracy, sentiment by theme, citation half-life, and prompt space coverage.
  • Segment everything by platform, topic, and prompt type. Brand-specific prompts behave very differently from organic ones.
  • Treat any metric as a trend, not a snapshot. Citations decay, and one prompt tells you very little.
  • Prompt coverage is the honesty check. A great score on a thin prompt list is a great score on nothing.

Why mentions and citations are not enough

Most AI visibility dashboards start with the same two numbers: how often your brand is mentioned, and how often your domain is cited. Both are useful. Both are also blunt.

A mention can be the first brand in a recommendation or the fifth in a list of alternatives. It can be glowing or lukewarm. It can describe your product correctly or place a premium tool in the budget bucket. A citation can point at your pricing page, or at a three-year-old review that says something you fixed long ago.

So what is AI perception? It is how AI engines position, describe, and recommend your brand when a buyer asks a question, and whether that picture stays stable over time. Measuring it takes a layer of metrics on top of presence.

I've pulled together six, drawing on frameworks from Brainlabs, Adobe, and Cognizo, plus first-party citation data from Promptwatch. Brainlabs is upfront that its KPIs are a proposed framework rather than settled industry standards, and that is the right way to treat all of this. Some of these metrics are still being defined.

Brainlabs article on five KPIs beyond mentions and citations for AI search

The baseline you still need

Before the six, get the basics in place. Visibility Score (the share of tracked prompts where you appear), share of voice, and citation share are the foundation. Cognizo's guide describes Visibility Score as the AI equivalent of impressions: 60 appearances across 200 tracked prompts is 30%. Report it as a trend and segment it by platform, topic, and funnel stage. A brand can lead on ChatGPT and trail on Perplexity for the same prompts.

The six metrics below sit on top of that baseline. If your baseline is noisy, they will be too.

Cognizo guide to the AI visibility metrics that matter in 2026

Metric 1: Position in answer

Being named is not the same as being named first. Position measures where your brand lands in the order of brands within an answer. Lower is better.

This matters more than it first appears because of how crowded AI answers are. Promptwatch Data found that 47.6% of ChatGPT responses with citations cite 10 or more unique domains, and 66.1% cite at least 8 (Sep 28 to Oct 4, 2026; see the unique domains per response report). You will share almost every answer. The question is whether you are the lead recommendation or a footnote.

How to measure it

  • Record the rank of your brand among all brands named in each response.
  • Report average position and the share of responses where you land in the top two.
  • Segment by prompt type. Comparison prompts and organic prompts produce different lists.

A healthy pattern is a falling average position next to a stable Visibility Score. A worrying one is a rising Visibility Score with a worsening position, which means you are being added to more lists but at the bottom of them.

Metric 2: Share of recommendations

This one comes from Brainlabs' framework, and I like the question it asks: when you appear, does AI actually recommend you? Many AI answers describe a category and name brands without endorsing any of them. Others say "for small teams, X is the best fit."

Share of recommendations separates neutral mentions from endorsements.

How to measure it

  1. Classify each response where you appear as a recommendation, a neutral mention, a comparison, or a caution.
  2. Divide recommendations by total responses that mention you.
  3. Compare against competitors on the same prompt set.

The gap between share of mentions and share of recommendations is the useful signal. A brand with 30% of mentions and 8% of recommendations is a well-known name that AI does not trust to pick. That is a content and proof problem, not an awareness problem.

Metric 3: Narrative accuracy

Adobe's list includes answer accuracy, narrative consistency, hallucination or misinformation rate, and outdated information rate. Cognizo calls its version positioning accuracy. They are the same idea: does AI describe you correctly?

A model can mention you positively and still get you wrong. It might attribute features you do not have, miss ones you do, or frame a premium product as a budget option. That misframing shapes buyer expectations before anyone visits your site.

How to measure it

  • Write down 10 to 20 facts that must be right: category, pricing model, key features, who it is for.
  • Pull the individual claims AI makes about you from tracked responses.
  • Mark each claim accurate, inaccurate, or outdated.
  • Report accuracy as a percentage of claims, and track the inaccurate ones until they disappear.

The measuring part is easy to describe and tedious to do by hand. Tools like Promptwatch pull the claims out of every tracked response through Brand Narrative, and its Fact Check feature then finds which pages on your own site state the wrong claim or its opposite, so you know where AI is learning it from.

Favicon of Promptwatch

Promptwatch

Track and optimize your brand visibility in AI search engines
View more
Screenshot of Promptwatch website

If a claim is wrong, look at third-party pages too. Promptwatch's own advice from the unique-domains data is to check which review and forum domains ChatGPT cites for prompts about your brand, since those shape how it describes you.

Metric 4: Sentiment by theme

Overall sentiment is a weak number. A score of 60 out of 100 could mean everyone is mildly positive, or that people love your product and hate your support. Break it down by theme: pricing, features, support, reputation.

Adobe frames this as share of favorable mentions and comparative sentiment. Compare your theme scores against competitors, because AI sentiment is relative. Neutral is a perfectly normal result for many categories.

How to read it

PatternLikely causeWhere to look
High visibility, weak sentimentReputation problemReview sites, Reddit threads, news
Strong sentiment, low visibilityCoverage problemPrompt set, content gaps
Weak sentiment on pricing onlyOutdated or unclear pricing contentYour pricing page, comparison pages
Sentiment differs by modelDifferent source mix per engineCitation sources per platform

Do not read sentiment in isolation from accuracy. A positive but wrong description is arguably worse than a neutral and correct one.

Metric 5: Citation half-life

This is the metric most teams skip, and it may be the most useful for planning. Citation half-life asks how long the visibility you earn actually sticks around.

Promptwatch Data tracked every page ChatGPT cited on July 6, 2026 for 90 days (see the content half-life report). The results vary a lot by content type:

Content typeStill cited after 30 daysStill cited after 60 days
Landing pages78%69%
Docs76%67%
Comparison pages72%63%
Listiclesn/a60%
Product pages, reviews, how-tosn/aabout 55%
Newsn/a41%

Product pages, reviews, and how-tos reach their half-life at 74 to 77 days. News is the fastest to fade: half of news pages were gone after 36 days, and almost a quarter were never cited again after day 0. One caveat from the report: landing pages, docs, comparisons, and listicles only crossed 50% in the last week because the window ended, so their real half-life is likely close to or beyond 90 days.

How to use it

  • For each important page, record the first cited date, the peak, and the date it falls below half its peak citations.
  • Set refresh cadences by content type. News-style content needs frequent updates. Docs and landing pages need less.
  • Treat a falling citation rate on a page as an action item, not a mystery.

This is also why one-off snapshots mislead. A spike in citations after a launch is not the same thing as durable visibility.

Metric 6: Prompt space coverage

Every other metric depends on one assumption: that your tracked prompts represent what people actually ask. Brainlabs calls this prompt space coverage, and it is the honesty check for the whole dashboard.

People phrase the same question dozens of ways, add context, name competitors, and combine needs into one request. A list of 100 prompts that a team brainstormed in a meeting can miss entire demand themes. In Brainlabs' analysis, the gaps included 66.5 million impressions around oil and energy prices and 50.5 million around currency exchange rates, among other themes.

Their suggested reporting style is worth borrowing. If your prompts cover only 61% of the demand themes you have identified, then "38% Share of Mentions at 61% prompt coverage" tells a much more honest story than 38% alone.

How to measure it

  1. List the demand themes for your category from Search Console queries, People Also Ask, sales calls, and support tickets.
  2. Map your tracked prompts to those themes.
  3. Coverage is the share of themes with at least one tracked prompt, ideally weighted by demand.
  4. Add prompts of every type: organic, brand-specific, and competitor comparison.

Prompt type matters because source behavior differs. In the Promptwatch report, organic prompts cited 10 or more domains 52.5% of the time, while brand-specific prompts did so only 21.3% of the time, with 5 domains the most common count. Blend them into one average and you hide both.

Measure across many organic prompts. One prompt tells you little.

Putting the six together

The metrics diagnose different problems. Read them as a set.

SymptomMetric to checkLikely fix
Named often but late in answersPosition in answerStronger proof, comparison content
Named often but rarely endorsedShare of recommendationsCase studies, third-party validation
Described wronglyNarrative accuracyCorrect pages and third-party sources
Tone is coolSentiment by themeReputation work, review sources
Gains fade fastCitation half-lifeRefresh cadence by content type
Great numbers, no revenuePrompt space coverageExpand and weight the prompt set

Then connect the result to outcomes. Adobe's framework suggests pairing visibility metrics with AI referral sessions, engagement, assisted conversions, and revenue from AI traffic. Be careful with click-based traffic as the main yardstick, since many AI answers include no clickable link. Also note that ChatGPT stopped sending a referrer on October 8, 2026, so tagging links with utm_source=chatgpt.com is now worth doing if you want to count that traffic.

Tools for measuring AI perception

You can run a version of this in a spreadsheet for a small prompt set. It stops scaling quickly. The tools below cover different parts of the job.

ToolBest forNotes
PromptwatchTracking plus fixingReal-UI tracking, Brand Narrative, Fact Check, crawler logs, content agents
ProfoundEnterprise monitoringStrong monitoring, higher price point
Peec AIBasic prompt trackingLightweight, no crawler logs or content generation
Otterly.AIBasic monitoringMentions and citations tracking
SemrushExisting SEO teamsAI tracking added to a traditional suite

A quick tour of the options:

Profound is an enterprise monitoring platform with solid coverage across engines.

Favicon of Profound

Profound

Enterprise AI visibility platform tracking brand mentions across ChatGPT, Perplexity, and 9+ AI search engines
View more
Screenshot of Profound website

Peec AI is a straightforward tracker for marketing teams that want visibility numbers without much setup.

Favicon of Peec AI

Peec AI

AI search visibility tracking for marketing teams
View more
Screenshot of Peec AI website

Otterly.AI tracks brand mentions and citations across ChatGPT, Perplexity, and Google AI Overviews.

Favicon of Otterly.AI

Otterly.AI

AI search monitoring platform tracking brand mentions across ChatGPT, Perplexity, and Google AI Overviews
View more
Screenshot of Otterly.AI website

Semrush adds AI search tracking to an SEO suite most teams already use.

Favicon of Semrush

Semrush

All-in-one digital marketing platform with traditional SEO and emerging AI search capabilities
View more

Most of these report where you stand. Promptwatch is the one in this group that goes on to act: its Action Board prioritizes fixes, the Content Agent writes and publishes pages against the gaps, and Page Tracker plus crawler logs check whether the new pages get crawled and cited. It also collects data from the real interfaces of ChatGPT, Gemini, Perplexity, Claude, and AI Overviews rather than relying only on APIs, which matters because user-facing answers can differ from API outputs. Plans start at $95/mo, with a 7-day free trial; confirm current details on the pricing page, or run the free AI visibility report first. For a wider list of options, the GEO software directory is a good next stop.

A simple measurement routine

Don't try to launch all six at once.

  1. Week 1: build the prompt set and map it to demand themes. Record prompt coverage.
  2. Week 2: capture baselines for Visibility Score, share of voice, and position, segmented by platform and prompt type.
  3. Week 3: tag recommendations and run the accuracy check on your 10 to 20 key facts.
  4. Week 4: add sentiment by theme and record first-cited dates for your key pages.
  5. Monthly: review the set as trends, refresh pages by half-life, and expand prompts to close coverage gaps.

Keep the report short. Six numbers with a one-line read on each beats a forty-widget dashboard nobody opens.

Common mistakes

  • Reporting one blended number. Segment by platform, topic, funnel stage, and prompt type.
  • Treating neutral sentiment as a problem. It is often normal.
  • Judging a single response. AI answers vary run to run, so aggregate across many prompts.
  • Counting a spike as a win. Check whether it survives 30 and 60 days.
  • Ignoring third-party sources. Reviews, forums, and directories feed how AI describes you.

Final word

Mentions and citations answer whether AI knows you exist. These six metrics answer whether it picks you, describes you correctly, and keeps doing so. Start with prompt coverage and narrative accuracy, since a thin prompt set and a wrong description will undermine every other number you report.

If you want the broader picture of how AI engines pull sources, Promptwatch Data publishes regularly updated citation research you can use as a benchmark.

Share:

© 2026 Surferstack · Find the best Marketing tools for your GTM motion · RSS

Surferstack is an affiliate review site. When you click links to vendors or buy through links on our site, we may earn an affiliate commission at no extra cost to you.

Surferstack is a review website based on user reviews on Reddit and G2, and on publicly available information. We keep everything as up to date as possible, but pricing and features can change. Always confirm the details with the vendor before purchasing.