What Is a GEO Audit, and How to Recognize a Trustworthy Methodology

A GEO audit determines whether and how a brand, website, or product shows up in generated answers, and investigates the reasons behind that outcome. A trustworthy audit doesn't rely solely on a technical checklist or a single opaque score. It combines testing actual answers, crawling the website, and analyzing content quality, entities, authority, and competitors.

What the audit should answer

• Does the AI know the brand and describe it correctly?
• Does the brand appear for non-branded queries?
• Is it merely mentioned, or actively recommended?
• Which pages are being cited?
• For which topics do competitors win out?
• Is the website accessible to crawlers?
• Does it contain specific information usable in an answer?
• Are the company, authors, products, and locations unambiguous?
• Do trustworthy external confirmations exist?
• Which changes have the highest expected impact?

Two layers of the audit

1. Observed outcomes

The audit runs prepared prompts against individual systems and records the answers, sources, mentions, and recommendations.

2. Website and brand readiness

The audit checks the possible underlying causes:

• crawling and indexing;
• robots.txt and specific bots;
• HTTP status codes and rendering;
• content coverage;
• structure and accuracy of information;
• schema and entities;
• authorship and trustworthiness;
• external mentions and reviews.

Technical readiness alone doesn't prove visibility. Conversely, real answers without diagnostics won't tell you what to fix.

How the prompt set should be built

Prompts should be derived from:

• products and services;
• customer problems;
• stages of the decision journey;
• locations;
• specializations;
• competitive alternatives;
• search and customer support data.

Split them into branded, informational, comparison, recommendation, purchase, and local categories. Disclose at least the structure and examples used; the client needs to understand what the score actually represents.

Repetition and variability

Run every important prompt repeatedly. Record the exact model, mode, and date. The result should be an occurrence rate, not a claim of "the AI recommends you" based on a single run.

For a larger audit, you can report:

• the number of unique prompts;
• the number of repetitions;
• the total number of answers collected;
• the breakdown by platform;
• the variability interval.

Technical review

The minimum scope should cover:

• Googlebot and indexing;
• OAI-SearchBot, GPTBot, and other relevant crawlers;
• robots.txt, meta robots, and X-Robots-Tag;
• canonical tags;
• the sitemap;
• 3xx, 4xx, and 5xx responses;
• server-side content vs. JavaScript;
• the mobile version;
• WAF/CDN blocking;
• structured data;
• product or local feeds.

llms.txt can be checked as a supplementary item, but it must not be given disproportionate weight.

Content review

This evaluates:

• intent coverage;
• original information;
• specificity;
• sources and verifiability;
• clarity of answers;
• authorship;
• freshness;
• logical structure;
• duplication;
• multimedia.

An automated score should serve as a diagnostic aid, not as objective truth.

Entities and authority

The audit compares:

• name, address, phone number, and business ID;
• company profiles;
• authors and their professional profiles;
• sameAs links;
• reviews;
• relevant directories;
• industry mentions;
• links and citations.

Quantity without quality is misleading. Ten duplicate directory listings may hold less value than a single authoritative reference.

How to read a 0–100 score

Ask:

• What's the formula?
• What are the weightings?
• Is visibility separated from technical readiness?
• How is missing data handled?
• Is the score relative to competitors?
• Can you click through to the evidence?
• Is a three-point change larger than normal variability?

Without answers to these questions, a score is just a visual shortcut.

Audit deliverables

A quality report includes:

1. an executive summary;
2. methodology and limitations;
3. raw metrics;
4. sample answers;
5. platform differences;
6. technical findings with URLs;
7. content gaps;
8. entity and reputation issues;
9. competitor comparison;
10. priorities ranked by impact and effort;
11. a plan for repeated measurement.

Warning signs

• guaranteed placement;
• secret "proprietary AI metrics";
• confusing GPTBot with a search crawler;
• heavy weighting of llms.txt without evidence;
• a single run of each prompt;
• unspecified models and dates;
• no raw answers provided;
• hundreds of automated recommendations with no priority order;
• competitor comparisons using a different methodology.

Sources

• OpenAI: Overview of OpenAI crawlers
• Google: Generative AI optimization guide
• GEO research paper