The methodology, fully transparent
How to run an AI visibility audit.
Ninety minutes. Four assistants. Your real paragraph.
The complete methodology for auditing how ChatGPT, Claude, Gemini, and Perplexity describe your brand. Six structured sections, ninety minutes of focused work, and a baseline measurement you can run again every month. Nothing proprietary — the same methodology Lynceus runs for clients, documented for self-service.
Last reviewed: May 2026
Before you start
Three things you need ready.
A spreadsheet
Rows for prompts. Columns for ChatGPT, Claude, Gemini, Perplexity. Plus columns for response text and citation URLs. Google Sheets, Excel, anything. The spreadsheet IS the audit's primary artifact.
Access to all four assistants
ChatGPT (chat.openai.com or chatgpt.com), Claude (claude.ai), Gemini (gemini.google.com), Perplexity (perplexity.ai). Free tiers work for the audit; you don't need paid plans. Use a fresh browser session to avoid prior-conversation context leakage.
Sales-call notes or customer interviews
The prompt panel needs to come from real buyer language, not your marketing team's preferred phrasing. If you don't have customer voice data, the audit's prompt panel will be biased toward what you wish buyers ask rather than what they actually ask.
The six audit sections
The 90-minute workflow.
Build the prompt panel (30 min)
Write down 20-40 buyer-intent prompts your category's buyers actually run in AI assistants. Pull from sales-call notes, customer interviews, and competitor SERP queries. Each prompt should be phrased the way an actual buyer types it, not the way an SEO would. Mix head terms ('best [category] software'), comparison prompts ('[your brand] vs [competitor]'), and long-tail intent ('[category] for [specific use case]').
Run the prompts against all four assistants (45 min)
Execute every prompt against ChatGPT, Claude, Gemini, and Perplexity. Capture the verbatim response and any citation URLs the assistant surfaces. A spreadsheet works — one row per prompt, one column per assistant, plus columns for response text and citation URLs. Tedious, but the spreadsheet is the audit's primary artifact.
Compute your share of paragraph (15 min)
For each response, calculate what percentage of the AI's text references your brand specifically, what percentage references each named competitor, and what percentage references third-party sources (Reddit, Wikipedia, review sites). Average across the prompt panel. This is your current share of paragraph — your baseline metric.
Categorize each prompt response (15 min)
Tag each response as one of: 'You win' (your brand is the primary recommendation), 'You appear' (you're mentioned but not primary), 'You're absent' (no mention at all), 'You're described wrong' (mentioned with factual errors), or 'You're cited from competitor' (mention sourced to a competitor's URL). The distribution across these categories tells you exactly what type of intervention you need.
Audit the citation sources (15 min)
For prompts where your brand is mentioned, examine which URLs the AI cited as sources. Are they your own domain, a competitor's site, a third-party review platform, Reddit, or Wikipedia? Sources you don't control are sources that can change your paragraph without warning. Sources on your own domain are leverage points you can improve directly.
Identify your three highest-leverage interventions
Based on the previous five sections, identify the three changes that would move the most prompts in the panel. Most commonly: publish an entity-defining homepage paragraph (moves 'absent' to 'appears'), publish category comparison pages (moves 'cited from competitor' to 'cited from your domain'), and publish cluster content on the head topic (moves 'appears' to 'wins'). Specifics depend on your audit findings.
Interpreting your findings
What each failure pattern means.
If you're absent from most responses
Root cause is almost always missing entity-defining content on your own domain. The AI has no authoritative source to cite. Highest-leverage fix: publish a clean factual paragraph on your homepage defining what your brand is, who you serve, and why someone buys from you. Deploy Organization + Product schema. Expect movement in 4-12 weeks.
If you appear but get described wrong
The AI is summarizing a stale or inaccurate source. Identify the cited URL — that's the source you need to outrank. Highest-leverage fix: publish authoritative current content on your own domain on the same factual claims, with entity-rich structure and schema markup. The AI re-ingests the newer source over 3-6 months and the paragraph updates.
If your competitor gets the citation when describing you
Your competitor wrote the comparison content first. The AI cites their URL because that's where it found the comparison. Highest-leverage fix: publish your own comparison pages — your-brand-vs-competitor — on your domain. The AI redirects comparison-intent citations to your URLs within 4-12 weeks.
If long-tail prompts win but head terms lose
Common pattern for new entrants. You're winning the specific, narrow queries but losing the contested category prompts. Highest-leverage fix: build a comprehensive pillar page on the head topic plus 8-12 cluster pages on subtopics. The pillar + cluster architecture is what signals topical authority on contested queries. Expect 3-6 months for measurable movement on head terms.
If you appear inconsistently across assistants
ChatGPT, Claude, Gemini, and Perplexity weight signals differently. Strong on ChatGPT but absent from Perplexity often means you're cited in training data but lack the recent authoritative content Perplexity's live search prefers. Highest-leverage fix: publish fresh, dated, schema-marked content quarterly so all four assistants have current authoritative sources to surface.
When to DIY vs hire
The audit is easy. The follow-through is hard.
Running the audit takes 90 minutes. Acting on the findings takes 6-12 months of disciplined content engineering. Most teams run one self-audit, identify the issues, can't sustain the publishing cadence required to fix them, and let the practice lapse.
Run the audit yourself if: your team includes a content engineer comfortable with semantic chunk architecture and schema markup, you can publish authoritative pillar + cluster content on a monthly cadence, and you have someone willing to re-run the audit every 30 days.
Hire a vendor if: you want the measurement plus the content engineering as one engagement, you don't have the in-house content velocity, or you want monthly re-measurement without the team overhead. See Lynceus vs Profound, vs Peec AI, and vs a traditional SEO agency for vendor type comparisons.
Frequently asked questions
What is an AI visibility audit?
An AI visibility audit is a structured measurement of how your brand appears in responses generated by AI assistants (ChatGPT, Claude, Gemini, Perplexity). It runs a locked panel of buyer-intent prompts against each assistant, captures the verbatim responses, computes share-of-paragraph for your brand and competitors, identifies which URLs the AI cites as sources, and categorizes each prompt by failure mode (absent, mentioned-wrong, cited-from-competitor, etc.). The audit's output is a baseline measurement plus a prioritized list of interventions.
How long does an AI visibility audit take?
A self-run audit takes roughly 90-120 minutes for a 20-30 prompt panel across four AI assistants. The free Lynceus AI Visibility report runs an abbreviated version in three minutes — three category prompts across all four assistants, returned with the verbatim paragraphs. Full done-for-you audits with multi-prompt panels, recurring monthly measurement, and category-specific intelligence run 1-2 weeks for the first cycle.
What does an AI visibility audit cost?
Self-run: free, but ~2 hours of time. Lynceus's free audit tool: free, ~3 minutes. Full vendor-run audits across the category range from $500 one-time (small SaaS-only audits) to $50,000+ for enterprise multi-prompt-panel deployments. Lynceus's full audit + 6-12 month engagement runs $4-7K/month and includes monthly re-measurement plus the content engineering to act on findings.
How often should I re-run my AI visibility audit?
Monthly for the locked prompt panel. AI assistants update training and grounding data on irregular schedules; share-of-paragraph movement happens on a 30-90 day rhythm. Quarterly audits miss the signal. Weekly audits over-react to AI response variance. Monthly is the right cadence for actionable measurement.
Can I audit my AI visibility myself or do I need a vendor?
Yes, you can run a self-audit using the methodology above — it's tedious but straightforward. The bottleneck is consistency: most teams run one audit, find issues, can't act on them, and abandon the practice. Vendors solve the consistency problem (monthly re-measurement, prompt-panel discipline) and the action problem (content engineering to act on findings). The decision is whether your team has the discipline to run the methodology in-house. See /lynceus-vs-profound, /lynceus-vs-peec-ai, /lynceus-vs-otterly for vendor comparisons.
What prompts should be in my audit panel?
20-40 buyer-intent prompts that your category's buyers actually run. Mix of head terms ('best [category] software'), comparison prompts ('[your brand] vs [competitor]', '[competitor] alternatives'), use-case prompts ('[category] for [specific use case]'), and decision prompts ('what's the difference between [your brand] and [competitor]'). Pull from sales-call transcripts, search-query data, and customer interviews. The panel is the measurement instrument — generic SEO keyword lists make poor panels.
How is an AI visibility audit different from an SEO audit?
An SEO audit measures Google rank position, technical issues, backlink profile, and on-page optimization. An AI visibility audit measures share of paragraph across AI assistant responses, citation source attribution, and which prompts win versus lose. The methodologies overlap (both reward entity-rich content, schema, topical authority) but the success metrics are different — SEO audits tell you about Google rank; AI audits tell you about what AI assistants say about you. See /lynceus-vs-seo-agency for the deeper comparison.
Skip the spreadsheet
Three minutes vs ninety.
Same four assistants. Same real paragraph.
The free Lynceus AI Visibility report runs an abbreviated audit — three category prompts across ChatGPT, Claude, Gemini, and Perplexity — and returns the verbatim paragraphs in three minutes. Start here, then graduate to the full 90-minute panel methodology if the abbreviated version flags issues worth investigating.