Home/AEO & GEO Guide/Brand Audit Playbook
Tactical Playbook

How to Run an AEO / GEO Brand Audit

A comprehensive, multi-engine audit methodology to benchmark AI visibility share, grade competitor queries, hunt hallucination gaps, and detect technical crawler blocks.

Methodology Note: This audit method is an empirical diagnostic workflow to identify visibility patterns and gaps, not a fixed official standard.

1. Setup (10 Minutes)

  • Turn memory off: In ChatGPT, Claude, and Gemini, disable memory in your settings, or use logged-out incognito windows. You want the impartial answer an un-personalized user would see.
  • Pick the target engines:Test across ChatGPT, Google AI Overviews, Google AI Mode, Perplexity, Gemini, and Microsoft Copilot. They share very few cited URLs, so do not treat “AI” as one thing.
  • Build a prompt coverage matrix:Cover each core product × target customer persona × geographic market across all 4 buyer stages (awareness, consideration, evaluation, decision). Take prompt phrasing directly from real sales calls and customer support tickets.

2. The 5 Core Prompt Categories (With Exact URL Input)

Run prompts across these 5 fundamental categories to test every angle of your brand:

CategoryExample Prompt TemplateWhat It Tests
Brand knowledge“What do you know about [brand]?”Entity recognition, accuracy, hallucinations
Category discovery“Best [category] companies for [use case/location]”Inclusion in generic unbranded recommendations
Comparison“[Brand] vs [competitor]”Comparative strengths, weaknesses, and pricing
Trust & sentiment“Is [brand] trustworthy? Strengths and weaknesses?”Sentiment score, complaints, review sentiment
Alternatives“Alternatives to [competitor]”Whether your brand is recommended as a substitute
Important Competitor Prompting Technique:When prompting about competitors, paste their exact URL(e.g. “Analyze https://competitor.com: what are their primary USPs and key service pillars?”). This stops the model from hallucinating or guessing from the brand name alone.

3. Competitor Grading & The Query Fan Map

Classify your rivals and search queries into clear tiers to discover high-ROI attack vectors:

Competitor Tiers
  • Apex rivals: Established authority, high review volume, consistently cited across all major engines.
  • Active rivals: Growing brands, regularly publishing original content with increasing citation presence.
  • Lucky rivals: Cited on only 1 or 2 isolated queries due to an outdated legacy article.
Query Grading (Query Fan Map)
  • Open queries: Engines give generic, inconsistent, or wrong answers. These are your primary opportunities.
  • Contested queries: Multiple active rivals cited; answers vary between runs.
  • Saturated queries: Dominated by apex rivals with entrenched entity consensus.

Strategic Rule: Start by targeting Open queries and weaker (Lucky) rivals first before competing against saturated queries.

4. Hallucination-Gap Hunting & Entity Consensus

Run deep checks on factual accuracy and model behavior:

A. Hallucination-Gap Check

For each Open query, observe whether the model produces generic answers, refuses to answer, changes the question, or cites completely irrelevant sources. These gaps indicate where the LLM lacks grounded training data. Publishing a direct, factually sourced guide allows you to capture that answer space.

B. Entity Consensus Check

Compare your core USPs, pricing, founder info, and locations across your site, Wikidata, Crunchbase, Trustpilot/G2 reviews, and press mentions. Models build confidence through consensus: when facts are repeated consistently across multiple independent platforms, models cite them accurately.

5. What to Record in Your Audit Sheet

Set up a spreadsheet with one row per Prompt × Engine × Run. Track these core metrics:

  • Mentioned vs Cited: Is the brand only named in text, or does it receive a direct clickable footnote citation?
  • List position: Note where you appear if a list is generated (e.g. #1, #3), but remember it fluctuates.
  • Sentiment: Is the generated tone positive, neutral, or negative?
  • Accuracy & Hallucinations: Are pricing, features, and company details accurate?
  • Competitors cited: Which competing brands appeared, and which tier do they belong to?
  • Source domains: Exact URLs and domains cited (your site, YouTube, Reddit, review portals, press).

6. Run Each Prompt 5–10+ Times (Variance Defense)

SparkToro found the same brand list appeared in under 1% of runs. One run proves nothing because LLMs use non-zero temperature sampling.

Instead of tracking an inconsistent single rank, calculate and record your Visibility Share %:

Visibility Share = (Number of runs your brand appeared ÷ Total runs tested) × 100

Standard reporting metric: “Across N sampling runs, the brand achieved an X% visibility share.”

7. Technical Layer & Crawler Trap Checks

  • Robots.txt inheritance: Ahrefs discovered 5.9% of sites unintentionally block GPTBot due to inherited wildcard disallow rules. Verify permissions for OAI-SearchBot, PerplexityBot, Claude-SearchBot, and Googlebot.
  • Cloudflare AI Scraper toggle: Cloudflare often has its automatic AI bot blocking feature toggled ON by default, which can block legitimate search indexing bots. Check your edge WAF rules.
  • JavaScript rendering: Ensure critical text and entity information can be read without JavaScript execution.
  • Entity Home: Have a clean, facts-first About page with Organization schema and authoritative sameAs links (Wikidata, Crunchbase, social profiles).

8. Attribution Setup (Capturing Dark AI Traffic)

Ahrefs data showed AI search accounted for only 0.5% of website sessions, but drove 12.1% of conversions. Standard GA4 referrer tracking severely undercounts AI visits because many users visit directly or use native apps.

  • Self-Reported Attribution:Add an open-text “How did you hear about us?” field on lead forms and sales discovery calls.
  • UTM & Referrer Tracking: Monitor android-app://com.openai.chatgpt and direct referral channels in GA4.

9. Action Plan Remediation Matrix

Audit FindingRemediation Action
Wrong or outdated factsUpdate pages you control (About page, schema), then fix third-party directory profiles to build consensus.
Absent from category promptsTarget Open queries using hallucination-gap hunting and comprehensive comparison pages.
Competitors cited from missing sourcesDigital PR outreach, YouTube content, listicle inclusion, and industry review sites.
Negative sentimentIdentify the source of complaints and address the root cause publicly.
Blocked or unrenderable pagesFix robots.txt directives, check Cloudflare AI bot settings, and eliminate client-side JS rendering blocks.

10. Re-Test After 4–6 Weeks

Do not expect instant changes. AI models and retrieval caches update with lag. Re-test after 4 to 6 weeks using the exact same prompt set and run count to measure your progress over time.