How to Run a GEO Audit: Step-by-Step Practical Guide (2026)

Why run a GEO audit?

Conversational AI systems — ChatGPT, Claude, Gemini, Perplexity — now mediate a growing share of how customers discover brands. Unlike traditional search engines that display a ranked list of links, LLMs synthesize a single answer from the sources they consider most relevant. Being cited or omitted by these models directly affects brand awareness, traffic, and revenue.

A GEO (Generative Engine Optimization) audit measures where your brand stands today across AI models and gives you an actionable plan to improve your visibility. This guide walks through the practical steps, from scoping to reporting.

Step 1: Define your scope

Select the AI models

• ChatGPT (OpenAI) — the most widely used conversational AI. It cites brands in answers, generates product recommendations, and can hallucinate URLs. Monitor both chatgpt.com and the GPT Store integrations.
• Claude (Anthropic) — strong in professional and technical contexts. Claude tends to cite authoritative sources and is growing quickly in enterprise settings.
• Gemini (Google) — deeply integrated with Google's ecosystem. Gemini's answers appear in Google Search (AI Overviews) and the Gemini app, combining traditional ranking signals with generative output.
• Perplexity — the first AI-native search engine. It explicitly cites sources with numbered footnotes. Perplexity users are research-oriented and click through to cited URLs at high rates.
• Mistral (Le Chat) — widely used in French and European markets. Mistral emphasizes data sovereignty and multilingual accuracy.

Define your market and language

AI responses vary significantly by language. A brand strongly visible in English may be entirely absent in French or German queries. Define the languages and markets that matter to your audience.

Choose your benchmark queries

Three categories of queries to test:

• Category queries — "best CRM for SMBs", "enterprise cloud storage providers"
• Problem queries — "how to reduce churn in SaaS", "solutions for GDPR compliance"
• Comparison queries — "HubSpot vs Salesforce", "AWS vs Azure pricing"

Prepare 10–30 queries per market segment. These become your benchmark set for all future audits.

Step 2: Build your prompt list

For each query, create a standardised prompt that you can reuse across models and audit cycles. The key is to test spontaneous, unprompted visibility — do not mention your own brand or any competitor in the prompt.

Example prompt format: "List the top [product category] providers for [target audience]. Explain what each one does well."

A well-constructed prompt list is the foundation of a reproducible audit. Document each prompt with its expected output category and the models it will be tested against.

Step 3: Collect the data

Manual collection

For each prompt, record:

• Was your brand mentioned?
• If mentioned, where in the response (first, middle, last)?
• If not, which brands were cited instead?
• Was the mention positive, neutral, or negative?
• Did the AI cite a specific source (URL, article)?
• Was the URL real or hallucinated?

Automated collection

For regular monitoring, automated tools can run the same prompt batch across multiple models and aggregate the results. This is especially useful for audits that need to run weekly or monthly, or for agencies managing multiple client brands.

Key features to look for in an automated solution:

• Multi-model support (ChatGPT, Claude, Gemini, Perplexity, Mistral)
• Multi-language query execution
• Hallucination detection (distinguishing real URLs from invented ones)
• Exportable results for client reports

Step 4: Calculate your metrics

Mention rate

The percentage of queries where your brand is cited at least once. This is the baseline metric for GEO. Industry benchmarks vary by sector — technology brands typically see higher mention rates than niche B2B services.

Citation position

Where in the response your brand appears. First-third mentions carry more weight because users read the beginning of AI answers most carefully.

AI Share of Voice (SoV)

Your brand's proportion of total brand mentions across your query set. This tells you whether the AI ecosystem mentions you as often as it mentions your competitors.

Mention sentiment

Classify each mention as positive, neutral, or negative. Positive mentions include recommendations and quality endorsements. Neutral mentions are factual listings. Negative mentions include warnings, errors, or hallucinated content about your brand.

Hallucination rate

The percentage of citations attributed to your brand that link to non-existent pages on your domain. A high hallucination rate suggests AI models have incomplete or incorrect knowledge about your content.

Step 5: Identify gaps and competitors

For queries where your brand is absent, note which competitor brands appear instead. Common patterns include:

• Absolute gaps — your brand never appears for entire query categories
• Position gaps — your brand appears but always at the bottom of the response
• Sentiment gaps — your brand appears but with neutral or negative framing while competitors get endorsements
• Source gaps — competitors have their content cited from authoritative sources while your citations link to thin pages

Step 6: Analyse technical readiness

AI models retrieve your content through a combination of crawling, indexing, and retrieval processes. Six technical factors directly influence whether your content gets cited:

Structured data (Schema.org)

Schema markup helps AI understand your content structure. FAQ schema, HowTo schema, and Organization schema are particularly relevant for LLM retrieval. Google's AI Overviews favour pages with well-structured schema.

Robots.txt and llms.txt

The llms.txt standard, introduced in 2025, gives site owners a way to provide AI models with a curated summary of their content. It complements robots.txt by telling LLMs which pages are most relevant to their use case, rather than just blocking or allowing access.

Content freshness

AI models have knowledge cut-off dates. For time-sensitive topics, regularly updated content has a higher chance of being cited. Date-stamp your articles and review them quarterly.

Internal linking

Well-structured internal linking helps both crawlers and language models understand site architecture. A clear hierarchy from pillar pages to topic clusters improves the probability of content being retrieved.

Multi-language handling

If you serve multiple markets, hreflang tags and separate language-specific URLs help AI models serve the right content to the right audience. A French user querying in French should receive your French content, not your English page.

AI bot detection and tracking

Identifying which AI crawlers (GPTBot, ClaudeBot, Google-Extended, etc.) visit your site, how often, and which pages they access helps you understand what AI models know about your content. Some tools offer dedicated AI bot tracking to close this visibility gap.

Step 7: Build an action plan

1. Fix critical gaps first — create or improve content for query categories where your brand is entirely absent.
2. Address brand safety issues — any negative mentions or hallucinated content should be addressed immediately. If an AI model is associating your brand with incorrect information, establish authoritative sources that contradict the error.
3. Improve technical readiness — implement schema markup, review robots.txt, and add an llms.txt file. These low-effort changes often produce quick visibility improvements.
4. Create citable content — AI models favour content that is original, well-structured, and authoritative. Research reports, data studies, and comprehensive guides are cited more frequently than thin marketing pages.
5. Monitor continuously — GEO is not a one-time project. AI model behaviour changes with each update, competitor content evolves, and new models emerge. Schedule audits at least quarterly.

Recommended tools

Several platforms offer GEO audit capabilities, from free basic checkers to comprehensive enterprise solutions. The right choice depends on your scope, budget, and whether you need manual one-off audits or automated recurring monitoring.

AI Labs Audit provides automated multi-model, multi-language GEO audits with hallucination detection, competitive SoV analysis, and scheduled reporting. A free radar pre-diagnosis is available to test your baseline.