Measurement guide
How to measure AI visibility.
Four metrics. One monthly loop.
AI visibility measurement isn't traffic analytics with extra steps. It's a different discipline — prompt panels instead of keywords, share of paragraph instead of rank position, monthly cadence instead of weekly. This page is the complete measurement framework.
Last reviewed: May 2026
The core framework
Four metrics. The first one matters most.
AI visibility decomposes into four measurable metrics. They're related — movement on one usually predicts movement on others — but each captures a distinct dimension of how your brand appears in AI responses. Track all four; weight share of paragraph highest.
Share of paragraph (the primary metric)
For each prompt in your panel, calculate what percentage of the AI's response references your brand specifically. Aggregate across the full panel to get an average share. This is the AI-search analog to share-of-voice in traditional media. A brand at 25% share of paragraph in its category is occupying meaningful real estate in buyer evaluations; a brand at 3% is structurally absent. Track monthly.
Mention frequency (the coverage metric)
Out of the prompts in your panel, what percentage produced a response that named your brand at all? A brand mentioned in 80% of relevant prompts is in the conversation; a brand mentioned in 15% is missing from the conversation entirely. This metric is simpler than share of paragraph and useful as an early indicator — coverage gaps appear here first.
Citation source quality (the leverage metric)
When the AI cites your brand, which URLs does it surface as sources? Citations from your own domain are leverage points you can control. Citations from third-party sources you don't control (Reddit, competitor sites, stale review platforms) are exposure points where your paragraph can shift without warning. Track the ratio of own-domain citations to off-domain citations — the higher, the more durable your AI visibility.
Competitive paragraph share (the relative metric)
For each prompt, compute share of paragraph for every named competitor in your category. Your absolute share of paragraph matters less than your share relative to the leading competitors. Going from 5% to 8% while the leader holds 30% is different from going from 20% to 25% while the leader drops from 30% to 22%. Competitive share captures relative momentum, not just your own movement.
The measurement loop
What the monthly cadence actually looks like.
Week 1 — Re-run the panel
Execute every prompt in the panel against ChatGPT, Claude, Gemini, Perplexity. Capture verbatim responses and citation URLs. Time required: 60-90 minutes for a 30-prompt panel.
Week 1 — Compute the four metrics
Calculate share of paragraph, mention frequency, citation source ratio, and competitive paragraph share. Log to your tracking spreadsheet or dashboard alongside the prior month's values. Time required: 30-45 minutes.
Week 2 — Diagnose movement
Which prompts moved up? Which moved down? Which competitor's share narrowed; which widened? Movement diagnostics drive the next month's content priorities. Time required: 60 minutes including stakeholder review.
Weeks 2-4 — Ship the response
Publish 2-4 new authoritative pages addressing the prompts where you're losing or absent. New cluster page, refreshed pillar section, new comparison content, schema fix — whatever the diagnostics surfaced. The publishing cadence IS the optimization layer; without it, measurement is just observation.
The tooling landscape
Four options. Pick by team capacity.
Free / DIY measurement
A spreadsheet, manual prompt execution against ChatGPT/Claude/Gemini/Perplexity, and a 90-minute time investment. Captures all four metrics for a 20-30 prompt panel. Sustainable for the first 2-3 audits; most teams abandon by month 3 because the manual work compounds. Best for category exploration and pre-vendor evaluation. See /ai-visibility-audit-guide for the full methodology.
Free tool measurement (Lynceus)
The free Lynceus AI Visibility report at /tools/ai-visibility runs three buyer-intent prompts against all four major AI assistants in three minutes. Captures share of paragraph and citation sources but only for the three prompts. Good for monthly spot-checks and initial baseline. Not sufficient for full prompt-panel coverage; suitable for early-stage measurement and pre-engagement discovery.
Tooling-only SaaS (Profound, Peec AI, OtterlyAI)
Dedicated AI visibility dashboards with prompt-panel tracking, scheduled re-runs, and analytics layered on top. Captures all four metrics across larger prompt panels (typically 50-500 prompts depending on tier). Requires your team to act on findings — the tools surface what's broken; you fix it. Pricing ranges from $29-489/mo (OtterlyAI) to $4-15K+/mo (enterprise Profound). See /lynceus-vs-profound, /lynceus-vs-peec-ai, /lynceus-vs-otterly for vendor comparisons.
Done-for-you services (Lynceus, traditional SEO agencies)
Vendor-run measurement plus content engineering as one engagement. The vendor runs the prompt panel, identifies gaps, and ships content to fix them on a monthly cadence. Lynceus runs ~$4-7K/month; traditional SEO agencies retooled for AI run wider. The right choice when your team can't sustain monthly measurement plus the content velocity required to act on it. See /lynceus-vs-seo-agency for the SEO-agency comparison.
For a deeper look at the AI citation tracking tools landscape including feature-by-feature comparison see AI citation tracking tools.
Frequently asked questions
How do I measure AI visibility?
Run a locked panel of 20-40 buyer-intent prompts against ChatGPT, Claude, Gemini, and Perplexity every month. For each response, compute four metrics: share of paragraph (% of response devoted to your brand), mention frequency (% of prompts that named your brand at all), citation source quality (ratio of own-domain citations to off-domain), and competitive paragraph share (relative position vs named competitors). Track all four over time; share of paragraph is the primary signal, the other three are leading indicators and leverage metrics.
What's the most important AI visibility metric?
Share of paragraph across a locked prompt panel — the percentage of each AI response devoted to your brand, averaged across the panel. This metric captures both whether you're mentioned AND how much of the AI's reasoning your brand occupies. A brand at 25% share of paragraph in its category is occupying meaningful real estate; a brand at 3% is structurally absent. Mention frequency, citation source quality, and competitive paragraph share are supporting metrics; share of paragraph is the primary one to track monthly.
How often should I measure AI visibility?
Monthly for the locked prompt panel. AI assistants update training and grounding data on irregular schedules, and share-of-paragraph movement happens on 30-90 day cycles. Weekly measurement over-reacts to AI response variance and creates noise. Quarterly measurement misses the signal between cycles. Monthly is the right cadence for actionable measurement — frequent enough to detect movement, infrequent enough to avoid noise.
Can I measure AI visibility with Google Analytics?
Only partially. GA captures the click-through layer — visits sourced from chatgpt.com, claude.ai, perplexity.ai, gemini.google.com hostnames. This is a fraction of total AI impact because most AI brand impressions don't produce clicks. Use GA for AI referrer tracking as a supporting signal; use prompt-panel measurement for the primary share-of-paragraph metric. Relying on GA alone systematically underestimates AI impact. See /ai-traffic-vs-google-traffic for the deeper analysis.
How big should my prompt panel be?
20-40 prompts for SaaS and most ecommerce; 60-100 for B2B enterprise (because of multi-persona × multi-industry coverage requirements). The panel should mix head terms ('best [category]'), comparison prompts ('[your brand] vs [competitor]'), use-case prompts ('[category] for [specific use case]'), and decision prompts ('what's the difference between X and Y'). Smaller panels miss long-tail coverage; larger panels create measurement overhead without proportional insight gain.
How do I track AI citation sources?
For each AI response that mentions your brand, capture the URLs the AI surfaces as citation sources. Categorize each as: own-domain (you control it), competitor-domain (they control it), third-party authority (Wikipedia, .edu, .gov, established publication), social/community (Reddit, Quora, forums), or review platform (G2, Capterra, Wirecutter, etc). The distribution tells you where your AI paragraph is sourced from — and which sources you need to either improve, displace, or build on.
Should I use a tool or DIY measurement?
DIY for the first 1-3 audits to understand the methodology and prove the value internally. Move to a tool or vendor when manual measurement becomes a bottleneck — typically by month 2-3 when your team's discipline starts to slip. Free tools (Lynceus's free report) cover narrow prompt panels for spot-checks; paid SaaS tools (Profound, Peec AI, OtterlyAI) handle larger panels with scheduled re-runs; done-for-you vendors handle measurement + content engineering as one engagement. Choose based on team capacity to act on findings, not just measurement bandwidth.
What if my prompt panel shows no movement over 4-6 months?
Three likely causes. First: your prompt panel is the wrong panel — too generic, not aligned to actual buyer-intent queries, or missing the specific phrasings your category's buyers actually use. Second: your content output is too slow — share-of-paragraph movement requires consistent monthly publishing of authoritative content, not one-time site refreshes. Third: your content lacks the structural discipline (semantic chunks, schema, entity richness) AI assistants reward. Audit which of the three applies before changing strategy.
Start your measurement loop
Three prompts. Four assistants.
Your baseline in three minutes.
The free Lynceus AI Visibility report captures the four core metrics for a three-prompt sample across all four major AI assistants. Use it as the start of your monthly measurement loop, then expand the panel.