What Is a GEO Audit and What Does It Check?

Generative Engine Optimization starts from a simple idea: AI can't cite what it can't read. A GEO audit is how you find out what a model can actually read, understand and lift out of your pages. It is not a ranking report and it is not a list of guesses — it is a set of findings about your own site.

How a GEO audit differs from an ordinary SEO audit

• SEO audit: relevance, authority, user satisfaction, technical health, rankings and traffic.
• GEO audit: all of the above as a baseline, plus explicit AI-crawler access, answer-first structure, question-phrased headings and visible freshness signals.
• The difference in one line: SEO asks "can this page rank?" — GEO asks "can this page be quoted?"

The two are not rivals. AI answer engines still run web searches to get current information, read the top results and compose an answer with citations, so ranking is a prerequisite for being cited. GEO builds on SEO rather than replacing it.

The quickest way to see what is blocking citation is a GEO audit, which is exactly what a studio specializing in this work does first: audit, fix, publish and track.

Most GEO audits move through the same checks in roughly the same order, from access to content to freshness. Here is what each one is looking for.

The subtle ones are the crawler rules. Search and citation access are governed separately from training access, and different bots do different jobs. Google-Extended governs Gemini and Vertex grounding and does not control classic Google AI Overviews, so the two should never be treated as the same switch. A single line in robots.txt can quietly remove a site from AI answers without touching its Google rankings.

Structure is the other half. Models split a question into one to three searches, then run follow-up grounding queries, often brand or entity names, to validate what they found. A page that opens with a 40-60 word answer and uses the questions customers actually type is far easier to lift and quote than a page that makes the model dig.

A GEO audit is only useful if it ends in a fix list and a schedule. That order matters, because publishing better content onto a site AI crawlers cannot read wastes the work.

1. Start with access. Read robots.txt line by line and confirm which AI search bots are allowed. Separate training-bot decisions from search and citation decisions, and document both.
2. Clear the technical blockers. Address page speed and anything else consuming crawl budget, then tighten meta descriptions so each page states its purpose plainly.
3. Rebuild the content architecture. Add a direct answer at the top of each page and under each section heading. Rewrite headings as the questions people actually ask. Replace decorative FAQ blocks with real FAQ data and matching schema.
4. Track and refresh. Record a starting point, check citation behavior on a schedule, and refresh content on a cadence so pages keep their freshness signals.

Two honest notes belong at the end of any audit. First, no one can guarantee citations — citation behavior moves day to day even when a site has not changed. What you can do is build the foundation these engines reward and measure it consistently. Second, freshness is not a one-time task. Content refreshed within roughly a month tends to be cited more, while content left untouched for a quarter or more tends to lose citations, so the work continues after launch.

Findings, not guesswork, is the standard to hold an audit to. If a report cannot tell you which specific element on which specific page is stopping a model from reading or quoting you, it is not finished.

A GEO audit report is a working document, not a scorecard. It should give you a clear picture of what is broken and what to do about it. Here is what belongs in it.

Findings

Each finding names the page, the element and the consequence. For example: "robots.txt blocks OAI-SearchBot sitewide, so OpenAI search cannot retrieve any page" or "the services page buries its answer in paragraph six, making it hard to lift." Findings are evidence-based, not opinion-based.

Gap analysis

The report maps your current state against the four requirements for being AI-citable: direct-answer content, question-phrased headings, explicit crawler access, and freshness signals. Gaps are the distance between where you are and where those requirements are met.

Prioritized fix list

Fixes are ranked by impact and effort. A single blocked crawler rule can outrank a dozen content edits because nothing else can get through until it is fixed. The list should be actionable by a developer, a writer or both, with enough detail to hand off.

Tracking plan

Because citation behavior is volatile, the report includes a schedule for checking whether the fixes worked. That means recording a baseline across ChatGPT, Perplexity, Gemini and Grok, then rechecking on a cadence. Trend history matters more than a single snapshot.

Refresh schedule

Since freshness signals decay, the report ends with a cadence for updating content. Pages refreshed within roughly a month tend to be cited more; pages left for a quarter or more tend to lose citations. The schedule keeps that from happening by accident.

GEO audits and SEO audits share the same foundation. Relevance, authority and user satisfaction still decide whether a page ranks, and ranking is a prerequisite for being cited because AI answer engines run web searches to gather current information. The GEO audit adds the layers that classic SEO never had to consider.

What carries over from SEO

• Technical health: site speed, crawlability and clean URLs still matter.
• Keyword relevance: pages still need to target the language people use.
• Meta discipline: titles and descriptions still shape how pages are understood.

What GEO adds

Explicit AI-crawler permissions, answer-first formatting, question-phrased headings, and visible freshness signals. These are the four requirements for being AI-citable, and none of them are standard SEO checklist items.

How the audit feeds ongoing work

The audit does not end with a report. Its findings become a content plan: pages to rewrite with direct answers, headings to rephrase as questions, schema to add, and a publishing cadence. From there, the work becomes a cycle of publishing, tracking citations and refreshing pages on a schedule.

This is where an in-house engine such as Iris, Vision's in-house content engine, fits into a GEO engagement. Iris writes pages to be quoted, checks drafts against a quality bar, publishes them at a paced rate, and refreshes them on a cadence with version history and rollback. The audit identifies what is broken; the publishing and refresh cycle keeps the site in a citable state.

For a business with no content at all, the audit findings can be the starting brief: the gaps tell you what pages need to exist. For a business that already publishes, the audit restructures what is there before adding anything new.

Audits tend to surface the same handful of problems, and most of them are invisible from a normal browsing session. Here are the mistakes that show up most often.

• Blocking the wrong bots. A robots.txt rule added to stop training crawlers often blocks search and citation crawlers too. OAI-SearchBot, PerplexityBot, ClaudeBot and Google-Extended are not the same as GPTBot, CCBot or Bytespider. Training and citation access are separate decisions, and treating them as one switch is the single most damaging error an audit finds.
• Confusing Google-Extended with AI Overviews. Google-Extended governs Gemini and Vertex grounding. It does not control classic Google AI Overviews. Blocking one does not block the other, and assuming it does leads to wrong conclusions.
• FAQ schema as decoration. Many sites include an FAQ section for looks, with questions nobody asks and no matching structured data. Real FAQ data tied to schema is what makes those answers liftable.
• Headings that label instead of ask. "Our Services" and "Why Choose Us" tell a model nothing about what question the page answers. Question-phrased headings match how people type into a chat assistant.
• Answers buried under introductions. A page that takes six paragraphs to reach its point is harder to quote than one that opens with a complete answer. A 40-60 word capsule at the top fixes most of this.
• Stale pages. Content left untouched for a quarter or more tends to lose citations. Freshness signals are weighted heavily, and refreshing within roughly a month keeps them strong.
• Weak meta descriptions. Auto-generated or duplicated descriptions leave a page's purpose unclear to both search engines and models.

None of these require new technology to fix. They require knowing they are there, which is the point of the audit.