GEO audit — check your AI visibility yourself in five steps

A GEO audit is a systematic check of how a company appears in AI search and what is blocking that visibility. The good news: a basic audit can be done yourself in a single afternoon without paid tools. Here is the framework we use, five steps after which you know where you stand and what to fix first.

Step 1: Test what the models say about you

Write 15 to 20 questions your customers would ask while buying: recommendations (”what is a good X”), comparisons (”A or B”) and brand questions (”what is Company Ltd”, ”Company Ltd reviews”). Run them in ChatGPT and Perplexity and record in a table: is the company mentioned, which competitors are named, which sources are cited and are there errors in the answers.

This is the most important step of the audit, because it produces two things at once: a baseline to compare later progress against, and a list of the cited sources, which is a direct map of the listings and sites you need to get into.

Step 2: Check whether the AI can reach your site

Three technical checks, fifteen minutes in total:

• robots.txt: open yourdomain.com/robots.txt and look for blocks on AI crawlers (GPTBot, ClaudeBot, PerplexityBot, Google-Extended). A block can be left over from a security plugin without anyone remembering it.
• Bot protection: some firewalls return an error page to ”unusual” crawlers even when robots.txt allows them. Test by fetching your page with curl using an AI crawler’s user agent, or ask a technical partner to check.
• Index coverage: search site:yourdomain.com in both Google and Bing. Bing matters because several AI services draw on its index. If the page count is clearly lower than reality, indexing goes on the fix list.

Step 3: Assess how citable your content is

Open your five most important pages and ask three questions about each:

• Which customer question does this page answer, and does it answer it in the first paragraph or only at the end?
• Does the page contain verifiable facts: prices, timescales, comparisons, definitions? Or only general promises?
• Would any paragraph work as it stands as part of an AI answer?

A typical finding on Finnish company sites is that service pages describe activity (”we offer comprehensive…”) but answer no question at all. In that case the top of the fix list becomes producing question-shaped content: pricing articles, comparisons, FAQ sections.

Step 4: Check the machine-readable basics

Put your home page and a service page through a schema validator. Check whether there is an Organization markup with correct details, whether FAQ sections have FAQPage markup and whether articles carry dates. At the same time check the consistency of your basic details: the same name, address and description on the site, in the Google Business Profile and in the main directories. Contradictions go on the fix list; they are the cheapest thing to correct and they directly affect whether a model dares to mention you.

Step 5: Map your mentions elsewhere

Search for your company name in quotation marks and see where you are mentioned: listings, directories, media, discussions. Compare against the source list from step 1. Do the models cite places where you are absent but your competitor is present? Every such place is a concrete line on the work list.

The output: a prioritised fix list

After the five steps you have findings that almost always fall into the same order of importance:

Priority	Typical fix	Effort
1	Incorrect information in answers and contradictory basic details	hours
2	Crawler blocks and indexing gaps	hours
3	Lack of content that answers questions	weeks
4	Missing mentions in cited sources	months

The order is not accidental: first fix what blocks or distorts, then build what is missing. And repeat the whole audit with the same question bank each quarter, so progress shows in numbers.

What an audit is trying to establish

An audit is not a list of everything wrong with a site. It answers one question: at which point does the chain from crawler to citation break for this company.

The chain has four links. The service has to be able to fetch the page. It has to be able to read the content. The content has to answer a question someone asks. And the source has to be trustworthy enough to cite. A break at link one makes work on links three and four pointless, which is why the order of the audit matters as much as its contents.

Most audits find the break in the first two links, which is good news: those are the cheapest to fix.

Step one: can the services reach you

Open the server log and filter for the AI crawlers. Three things need checking, and only the third surprises people.

Whether they visit at all. Which pages they fetch, which shows what they consider worth reading. And what response code they receive. A 403 or 429 means something is turning them away before robots.txt is even consulted, which is common on sites behind a CDN or bot protection.

If no AI crawler appears in the log at all, stop the audit here and fix that. Everything else is theory until this is working.

Step two: can they read what is there

Fetching is not reading. Two failures are common and both are invisible to a person using a browser.

Content that requires JavaScript. Some crawlers do not execute scripts, so text rendered by the browser after load is simply absent for them. Test by disabling JavaScript and reloading, or by reading the raw page source.

Content locked behind an interaction. Tabs, accordions and load-more buttons are fine when the text is in the source and merely hidden. They fail when the text is fetched on click. The check is the same: search the source for a sentence you can only see after clicking.

A third, smaller item: check that important pages are not excluded in robots.txt by a rule someone added years ago for a reason nobody remembers.

Step three: does the content answer anything

This is where an audit turns from technical to editorial, and where most of the real work sits.

Take the twenty questions customers ask before buying. For each, find the page on your site that answers it. Three outcomes are possible: a page answers it directly, a page mentions it inside a longer text, or nothing on the site addresses it. Count the three.

A typical first audit finds four or five direct answers, ten buried mentions and five gaps. The buried ones are the fastest gain, because the material already exists and needs restructuring rather than writing.

Step four: is the source credible

The last link is about identity, and it is checked page by page rather than site-wide.

Does the page name an author with a reason to be believed. Does it carry a date and has it been updated since publication. Are claims sourced where they are not your own observation. Does the site state plainly who the company is, where it is and how to reach it.

These take minutes to check and are the most commonly skipped part of an audit, because they feel like formalities. They decide which of several equally good candidate pages gets cited.

Turning findings into an order of work

An audit that ends in a list of forty items gets ignored. Sort the findings into three groups instead.

Blocking issues first: anything preventing pages from being fetched or read. These are usually fewer than five and often fixable in a day.

Quick gains second: existing pages where the answer exists but is buried. Restructuring one of these takes an hour and produces a quotable passage.

New content third: the genuine gaps, which take real writing time and should be prioritised by how close the question sits to a purchase decision.

Then set a repeat date. Six months is enough for a site publishing regularly, twelve for one that is not, and the second audit is far faster than the first because the technical checks are already known good.

Frequently asked questions

What is a GEO audit?

A systematic check of AI visibility: what the models say about the company, whether crawlers can reach the site, whether the content works as a source for answers and where the company is mentioned elsewhere. The output is a prioritised fix list.

Can a GEO audit be done in-house?

The basic level can: question tests, robots.txt, site searches and mention mapping all work without special tools in an afternoon. An agency is usually used for technical testing of bot protection, competitor comparison and ongoing monitoring.

How often should the audit be repeated?

A light question test monthly, the full audit quarterly. AI services and their sources change fast enough that a picture from a year ago says nothing about today.

What does a GEO audit cost from an agency?

Typically €500–2,000 depending on scope. The price usually includes question tests across several models, a technical check, a competitor comparison and a written action list.

How long does a GEO audit take?

The technical checks take a few hours for a normal site. The editorial part, matching customer questions against existing pages, is what takes the time and depends on how many pages the site has. A first audit for a small company fits into two days.

How often should the audit be repeated?

Every six months for a site publishing regularly, annually otherwise. The repeat is much faster than the first pass, because the technical findings are already known and only the content coverage needs re-checking.

You can start the first step of the audit right now: the AI visibility test runs the brand question for you. The full audit as a service is on the GEO service page.