Measurement & Analytics

    AI Visibility Metrics: What to Track and How

    How to measure your brand in ChatGPT, Perplexity, Gemini and AI Overviews: mention rate, share of voice, sentiment, citations and AI referral traffic.

    Saarthak Srivastav

    Saarthak Srivastav

    · 6 min read

    Key takeaways

    • Rank tracking doesn't see AI answers. AEO measurement is built on a fixed set of buyer prompts, run repeatedly across assistants.
    • The core metrics are mention rate, share of voice, position, sentiment and accuracy, citation share, and AI referral traffic.
    • AI answers vary between runs, so report rates over many prompts and runs, never single answers.
    • Segment everything by prompt type and platform. Averages hide where you're winning and losing.
    • Connect visibility to business outcomes with AI referral traffic, branded search trends and self-reported attribution.

    AI visibility metrics measure how your brand shows up inside AI-generated answers: how often you're mentioned, how prominently, how favorably, and which sources the assistant relies on. Traditional rank trackers can't see any of this. AEO measurement works differently: you define a fixed set of buyer questions, run them repeatedly across assistants, and track rates over time. This guide explains each metric, how to calculate it, and how to report it.

    Why AI visibility needs its own measurement

    Three things make AI answers hard to measure with SEO tools:

    1. There's no list to rank in. An answer names a handful of options in prose. Your "position" is whether you're named, and where in the text.
    2. Answers vary. The same question can produce different brands on different runs, days, platforms and locations.
    3. The answer is the outcome. Many users never click. A mention that shapes a buyer's shortlist has value whether or not it sends a visit.

    So the foundation is a prompt set: a fixed list of questions you track consistently.

    Building the prompt set

    A good prompt set reflects how your buyers actually ask:

    • Category prompts: "What are the best [category] tools?"
    • Use-case prompts: "Best [category] for [industry / team size / budget / constraint]"
    • Comparison prompts: "[You] vs [competitor]", "[competitor] alternatives"
    • Decision prompts: "[You] pricing", "Is [you] worth it?", "[You] reviews"
    • Problem prompts: "How do I [problem your product solves]?"

    Tag each prompt by stage, product line, use case and region. Write them in natural language, the way someone would type into ChatGPT, not as keywords. Start with 50 to 100 and keep the core set stable so trends are comparable.

    The core metrics

    1. Mention rate

    What it is: The share of answers that mention your brand.

    How to calculate: Answers mentioning you ÷ total answers, across all prompts and runs.

    Why it matters: It's the headline number for presence. If you're mentioned in 25% of answers to category prompts, three out of four buyers asking those questions never hear your name.

    Segment by: prompt type, platform, product line. A 60% mention rate on branded prompts and 5% on category prompts tells a very different story than the 30% average.

    2. Share of voice

    What it is: Your mentions as a share of all brand mentions in your competitive set.

    How to calculate: Your mentions ÷ total mentions of you and your tracked competitors.

    Why it matters: Mention rate tells you about you; share of voice tells you about the market. If everyone's mention rate rises because answers name more brands, your share of voice shows whether you actually gained ground.

    Benchmark: Compare share of voice with your market share. A big gap in either direction is worth investigating.

    3. Position (prominence)

    What it is: Where in the answer you appear: first, in the middle, or last, and whether you're the primary recommendation or an "also consider."

    How to calculate: Average order of mention when mentioned, or the share of mentions where you're named first.

    Why it matters: Readers weight the first options heavily, and assistants often frame the first option as the default choice.

    4. Sentiment and framing

    What it is: How you're described: positive, neutral or negative, and with what qualifiers.

    How to track: Classify each mention, and also keep the actual phrases. "Powerful but expensive," "best for enterprises," "limited integrations" are actionable; a sentiment score alone isn't.

    Why it matters: A mention with a damaging qualifier can do more harm than no mention. Recurring phrases usually trace back to specific sources you can update or counter.

    5. Accuracy

    What it is: Whether facts about you are correct: pricing, features, availability, integrations, founding details, leadership.

    How to track: Keep a fact sheet and check answers against it. Flag errors by type and source.

    Why it matters: Outdated pricing or a feature you shipped a year ago but the AI says you lack can lose deals silently. See how AI assistants decide which brands to recommend for tracing errors to sources.

    6. Citation share

    What it is: How often your own URLs are cited as sources, and which third-party domains are cited most for your prompts.

    How to calculate: Citations to your domain ÷ total citations, plus a ranked list of cited domains.

    Why it matters: Citations show what the assistant trusts. Your domain gaining citation share means your content work is landing. The third-party list is your outreach roadmap: those are the reviews, roundups and communities to be present in.

    7. AI referral traffic

    What it is: Visits from links in AI answers.

    How to track: In your analytics, look for referrers such as chatgpt.com, perplexity.ai, gemini.google.com, claude.ai and copilot.microsoft.com. ChatGPT often appends utm_source=chatgpt.com to links. Create a custom channel group for AI assistants so it's visible in every report.

    Caveat: Some AI-driven visits arrive with no referrer (copied links, app-to-browser handoffs) and show as direct. Google AI Overviews and AI Mode clicks are counted as Google organic. Treat AI referral numbers as a floor.

    Handling variability

    AI answers are probabilistic. To get stable numbers:

    • Run each prompt several times per measurement cycle. Three to five runs is a reasonable start.
    • Measure on a schedule, such as weekly, rather than ad hoc.
    • Report rates and trends, not individual answers. "Mention rate up from 18% to 26% over six weeks" is meaningful; "ChatGPT didn't mention us this morning" isn't.
    • Keep conditions consistent. Same prompts, same platforms, same locations where relevant, and note model changes when platforms ship them.
    • Watch for step changes. A sudden shift across many prompts on one platform usually means a model or retrieval update, not something you did.

    Building a report

    A useful monthly AI visibility report fits on one page:

    Section What to show
    Headline Mention rate and share of voice, this month vs last, by platform
    Segments Mention rate by prompt type (category, use case, comparison, decision)
    Competitors Share of voice for you and your top three to five competitors
    Framing Top positive and negative phrases used about you; accuracy issues found
    Sources Top cited domains for your prompts; your domain's citation share
    Outcomes AI referral sessions and conversions; branded search trend
    Actions What was done last month, and what's planned next

    The "Actions" row matters most. Each metric should point to work: a missing comparison page, a review platform to get onto, a pricing fact to correct.

    Connecting visibility to revenue

    Executives will ask what AI visibility is worth. Build the case from several signals, since no single one is complete:

    • AI referral conversions: sessions and conversions from AI referrers, tracked as their own channel.
    • Branded search demand: AI mentions often lead to people searching your name later. Watch branded impressions in Search Console.
    • Self-reported attribution: add "ChatGPT / AI assistant" as an option to "How did you hear about us?" on signup and demo forms. It often surfaces far more influence than referral data shows.
    • Sales conversations: ask reps to note when prospects mention asking an AI assistant.

    Tools

    You can start with a spreadsheet: a list of prompts, a schedule, and columns for mention, position, competitors, sentiment and citations. It works for 30 prompts on two platforms. It stops scaling quickly once you add platforms, repeated runs and competitors.

    AEO-Rank runs your prompt set across ChatGPT, Perplexity, Gemini, Claude and Grok on a schedule and tracks mention rate, share of voice, sentiment, cited sources and competitor positions over time, so you can spend your time on the actions instead of the data collection.

    For a shorter take, see our guide to measuring AI visibility.

    The bottom line

    AI visibility is measurable if you measure the right thing: a stable set of real buyer questions, run repeatedly across platforms, tracked as rates and trends. Watch mention rate and share of voice for presence, position and sentiment for quality, citations for what AI trusts, and referral traffic and attribution for business impact. Then make every metric point to a next action.

    For context on overall AI usage, see AI search statistics.

    FAQ

    Frequently asked questions

    What is AI visibility?

    AI visibility is how often, how prominently and how favorably your brand appears in answers from AI assistants such as ChatGPT, Perplexity, Gemini, Claude and Google AI Overviews, measured across a set of questions your buyers ask.

    How many prompts should I track?

    Start with 50 to 100 prompts covering your main categories, use cases, comparisons and buying questions. Fewer than about 30 makes rates noisy. Large brands with many product lines often track several hundred, grouped into segments.

    How often should I measure AI visibility?

    Weekly is a good cadence for most brands, with each prompt run several times per measurement to account for variability. Monthly reporting on trends works well for stakeholders.

    Can Google Analytics track traffic from ChatGPT?

    Yes, partly. Visits from links in ChatGPT, Perplexity, Gemini, Claude and Copilot usually arrive with those domains as referrers, and ChatGPT often adds a utm_source parameter. Some AI traffic arrives without a referrer and shows as direct, so treat analytics numbers as a floor.

    What's a good share of voice in AI search?

    It depends on how crowded your category is. A useful benchmark is your share of voice compared with your market share: if you hold 20% of the market but only 8% of AI mentions, AI answers are underrepresenting you.

    • metrics
    • measurement
    • share of voice
    • analytics
    • reporting
    Saarthak Srivastav

    Written by

    Saarthak Srivastav

    AEO & GEO, AEO-Rank

    Saarthak Srivastav writes and reviews articles on the AEO-Rank blog, drawing on what the AEO-Rank team sees tracking AI answers across ChatGPT, Gemini, Claude, Perplexity and Grok.