How to Perform a GEO Audit Step by Step

Most brands running traditional SEO audits are measuring the wrong thing. They check rankings, crawl errors, and backlink profiles – and those metrics still matter – but they tell you nothing about whether your content is being cited by ChatGPT, Perplexity, Gemini, or Claude. You can hold a top-three position on Google and be completely invisible inside every AI-generated answer your prospects actually read.

That gap is what a GEO audit closes. Generative Engine Optimization (GEO) auditing is the systematic process of evaluating whether your content is structured, authoritative, and technically accessible enough to be extracted and cited by large language models. Research from Aggarwal et al. found that GEO techniques can boost content visibility in generative engine responses by 30 to 40 percent. The question is not whether you need a GEO audit. The question is whether you know how to run one.

This guide walks through the full process – from baseline visibility testing to technical infrastructure checks to content restructuring – using a framework built from the most current practitioner research available.

Why GEO Auditing Is Different from Traditional SEO Auditing

Before running a GEO audit, it helps to understand why the two disciplines diverge so sharply.

Traditional SEO audits ask: can this page rank? They evaluate crawlability, indexation, keyword density, page speed, and backlink authority. Google’s Web Rendering Service processes JavaScript, follows internal links, and builds a searchable index from rendered content.

GEO audits ask: can this paragraph be extracted? AI crawlers – GPTBot, ClaudeBot, PerplexityBot, Google-Extended – operate differently. Most do not execute JavaScript. They read raw HTML on the first server response and move on. A site that ranks well on Google can be completely invisible to ChatGPT and Perplexity if its content loads via client-side JavaScript or if its key facts are buried in poorly structured prose.

There is also an architectural distinction. AI engines use Retrieval-Augmented Generation (RAG), which means they retrieve chunks of content from indexed sources and synthesize them into answers. For your content to appear in those answers, it must pass two tests: it must be technically accessible to AI crawlers, and it must be structured in a way that makes individual paragraphs and sentences extractable as standalone, trustworthy statements.

Owen Steer of Fifty Five and Five, writing in March 2026, notes that AI crawl volume now sits at roughly 40 to 50 percent of Googlebot-level activity across the web – but the traffic return is asymmetric. Anthropic’s crawlers generate approximately 38,000 crawl requests for every single referral back to a site. The value is not in click-throughs. It is in citations.

Phase 1: Baseline Visibility Assessment

The first phase of any GEO audit is establishing where you currently stand across AI platforms. You cannot optimize what you have not measured.

Step 1: Select the AI platforms you will track. Michael Lamp, Chief Digital and Social Officer at Hunter, recommends starting with ChatGPT, Claude, Gemini, Perplexity, and Copilot. These five cover the majority of AI search traffic and represent different underlying models with different citation behaviors. Assign one team member to each platform to run consistent daily or weekly checks.

Step 2: Build a fixed prompt set. Develop five to ten prompts that mirror the questions your target audience actually asks. These should be conversational and intent-driven, not keyword-stuffed. Example prompt structures include: “What are the best tools for [your category]?”, “Who are the leading companies in [your industry]?”, “How do I choose a [your product type]?”, and “What should I know about [your core topic]?” The prompts must remain identical across each testing session so you can track changes over time.

Step 3: Document responses systematically. Screenshot every response. AI models update constantly, and small shifts in training data or retrieval logic can change what gets cited. Store screenshots in a shared folder where the full team can review them. Record which sources get cited, whether your brand appears, and how competitors are positioned.

Step 4: Identify citation gaps. After running your prompt set across all five platforms, map the results. Note which platforms cite you, which ignore you, and which cite competitors instead. Pay attention to the type of content that earns citations – is it listicles, how-to guides, original research, or product comparison pages? These patterns tell you where to focus your optimization effort.

Phase 2: Technical GEO Audit

Once you have a baseline, the next phase addresses the technical infrastructure that determines whether AI crawlers can access your content at all.

Step 5: Audit your robots.txt for AI crawler policies. Your robots.txt file is the first gate AI crawlers encounter. There are two distinct categories of AI crawlers, and most teams conflate them.

Training crawlers – GPTBot (OpenAI), ClaudeBot (Anthropic), Google-Extended (Google/Gemini), Applebot-Extended (Apple Intelligence), and CCBot (Common Crawl) – harvest content to build and update model knowledge bases. Retrieval crawlers – ChatGPT-User (OpenAI), Claude-SearchBot (Anthropic), and PerplexityBot (Perplexity) – fetch content in real time when a user asks a question.

These are separate systems with separate access controls. Blocking GPTBot prevents your content from entering OpenAI’s training data but does not affect ChatGPT-User, which is what fetches your content during live queries. HTTP Archive data from 12.15 million sites shows that 21 percent of the top 1,000 websites currently block GPTBot. For most brands in 2026, the right policy is to block training crawlers while allowing retrieval crawlers – preventing your content from being used to train models while keeping it visible in AI search answers.

Step 6: Check JavaScript rendering. If your content loads via client-side JavaScript, AI search engines cannot see it. GPTBot, ClaudeBot, and PerplexityBot do not execute JavaScript. They do not run your scripts, wait for API calls to return, or interact with single-page applications. What they see is the raw HTML your server sends on the first request. Use seoClarity’s rendering tests or a simple curl command to check what your pages look like without JavaScript execution. Any content that only appears after JavaScript runs is invisible to most AI crawlers.

Step 7: Evaluate your structured data implementation. Schema markup is not optional for GEO. AI engines rely on structured data to understand entities, relationships, and content types. Prioritize these schema types for AI visibility:

Use Google’s Rich Results Test to validate JSON-LD syntax. Ensure schema content matches visible page content – AI engines penalize mismatches.

Step 8: Assess llms.txt implementation. The llms.txt standard is a structured file that tells AI engines what your site is about and how to interpret it. However, SE Ranking’s analysis of nearly 300,000 domains found no measurable correlation between having an llms.txt file and being cited by AI engines. Implement it as optional scaffolding after fixing robots.txt, JavaScript rendering, and schema markup – not before.

Step 9: Check page speed and Core Web Vitals. Target mobile page load times under 1.8 seconds. Prioritize Interaction to Next Paint (INP), which replaced First Input Delay as a Core Web Vital in March 2024. While AI engines do not directly measure page speed, they tend to cite content from technically sound websites.

Phase 3: Content Structure Audit

Technical access is necessary but not sufficient. Even if AI crawlers can reach your pages, they will not cite content that is poorly structured, vague, or difficult to extract as standalone statements.

Step 10: Audit for answer density. Directive Consulting recommends tracking “Answer Nugget Density” – the number of direct, one-to-three sentence answers per 1,000 words of content. Aim for at least six direct answers per 1,000 words. Every section should open with a clear, quotable statement that answers the implied question of that heading. If a section requires reading three paragraphs before the key point appears, AI engines will skip it.

Step 11: Evaluate content extractability. Each section of your content should be able to stand alone if extracted out of context. Ask Mona’s GEO audit methodology recommends testing this by reading individual paragraphs in isolation – if a paragraph requires surrounding context to make sense, it is not extractable. Rewrite sections so that each one delivers a complete, self-contained insight.

Step 12: Check heading structure and formatting. Use semantic HTML with a clear H1 to H3 hierarchy. Structure H2s as questions that mirror real user prompts. Limit paragraphs to under 120 words. Use numbered lists for processes, bullet points for quick facts, and comparison tables for feature or option comparisons. Keep key facts in text, not only in images or PDFs.

Step 13: Assess content freshness. Research has found that AI assistants unfavorably weight older content. Build a spreadsheet tracking published dates, modified dates, and citation performance. Proactively update high-value pages quarterly. Make last-updated dates visible on the page – AI engines use timestamps as a freshness signal.

Phase 4: Authority and Trust Signal Audit

The third dimension of GEO is authority. AI engines do not cite sources they do not trust. Trust is built through a combination of E-E-A-T signals, external validation, and entity clarity.

Step 14: Audit author and brand credentials. Every piece of content should have a visible byline with a linked author bio that states credentials and practical experience. Include an About page that clearly describes your organization’s expertise. Add Organization schema with verified social profiles linked via sameAs properties – connecting your brand to LinkedIn, Wikipedia, Crunchbase, and other authoritative directories reduces NLP ambiguity and increases the likelihood that AI engines will confidently cite you.

Step 15: Evaluate citation and source quality. AI engines lean toward sources that cite primary research. Audit your content for claims that lack supporting data. Add statistics with publication years, reference recognized research institutions, and link to first-party data studies. Profound’s GEO framework recommends earning citations from at least 20 high-authority domains per quarter as a benchmark for citation authority.

Step 16: Conduct competitor citation analysis. Run your prompt set and document which competitors appear most frequently. Study their content structure, section length, use of statistics, and formatting choices. Identify patterns in what earns citations in your category. Replicate what works and improve where you can.

Phase 5: Measurement and Iteration

A GEO audit is not a one-time exercise. AI models update continuously, citation patterns shift, and new platforms emerge. Build a measurement cadence into your workflow.

Step 17: Define your GEO KPIs. The core metrics for GEO success differ from traditional SEO metrics. Track the following:

Step 18: Run prompt tests on a fixed cadence. Profound recommends running 20 to 30 unique prompts per core topic, tested daily. Michael Lamp recommends weekly at minimum. The key is consistency – use the same prompts each time so you can detect changes in citation patterns over time.

Step 19: Track GEO Adoption Rate. Directive Consulting defines GEO Adoption Rate as the percentage of audited pages that meet at least eight checklist items out of a defined set. Target 70 percent compliance or higher across your content portfolio. This gives you a single number that represents your overall GEO readiness.

Step 20: Iterate quarterly. Geoptie’s GEO framework recommends quarterly benchmarking cycles: assess visibility scores, share of voice, and sentiment trends; identify which pages gained or lost citations; update content based on findings; and document changes so you can correlate them with visibility shifts.

The GEO Audit Checklist at a Glance

Use this checklist to track your audit progress across all five phases:

Phase 1: Baseline Visibility

• Select 5 AI platforms to track (ChatGPT, Claude, Gemini, Perplexity, Copilot)
• Build a fixed set of 5-10 prompts mirroring real user questions
• Screenshot and document all AI responses
• Map citation gaps and competitor citation patterns

Phase 2: Technical Infrastructure

• Audit robots.txt for training vs retrieval crawler policies
• Test pages for JavaScript rendering issues (curl test)
• Implement and validate FAQPage, HowTo, Organization, and Article schema
• Check page speed and Core Web Vitals (target INP under 200ms)
• Add llms.txt as optional scaffolding

Phase 3: Content Structure

• Audit Answer Nugget Density (target 6+ per 1,000 words)
• Test each section for standalone extractability
• Review heading structure (H1-H3 hierarchy, question-based H2s)
• Check content freshness and update timestamps

Phase 4: Authority Signals

• Add author bylines, bios, and credentials to all content
• Implement Organization schema with sameAs links to verified profiles
• Add primary source citations with publication years
• Earn citations from 20+ high-authority domains per quarter

Phase 5: Measurement

• Define AI Citation Rate, Response Inclusion %, and Query Coverage baselines
• Run fixed prompt set on weekly cadence
• Track GEO Adoption Rate (target 70%+ compliance)
• Benchmark and iterate quarterly

Common GEO Audit Mistakes to Avoid

Over-indexing on llms.txt. As noted above, current data shows no measurable correlation between llms.txt implementation and AI citations. Fix your technical foundations first.

Blocking retrieval crawlers while trying to earn citations. If you block ChatGPT-User or PerplexityBot in robots.txt, those platforms cannot fetch your content in real time. Check your robots.txt carefully – many brands have blanket AI crawler blocks that prevent citation entirely.

Treating GEO as a one-time exercise. AI models update continuously. A page that earns citations today may lose them after a model update. Build a recurring audit cadence into your workflow.

Measuring only traffic. AI-driven citations often generate zero direct clicks but significant brand influence. Track citation rate and response inclusion alongside traffic metrics to get the full picture.

Ignoring the prompt set. The most common failure in GEO auditing is running informal, inconsistent tests. Without a fixed prompt set tested on a regular cadence, you cannot detect citation pattern changes or measure the impact of your optimizations.

Starting Your First GEO Audit

If you are running a GEO audit for the first time, start with your top 20 pages by revenue impact. Apply the Phase 2 technical checklist first – if AI crawlers cannot access your content, nothing else matters. Then move to Phase 3 content structure improvements on the same pages. Run your baseline prompt set before making changes so you have a measurement baseline to compare against.

The goal of a GEO audit is not perfection. It is systematic improvement. Each iteration of the audit cycle – test, fix, measure, repeat – moves your content closer to the standard that AI engines use when deciding which sources to cite. In a search landscape where one synthesized answer replaces ten blue links, being cited is the new ranking.