A GEO audit should include five checks: whether the AI crawlers can reach and read your site, what ChatGPT, Gemini, Perplexity, Copilot and Google AI Overviews currently say about your business, whether your pages answer questions directly enough to be quoted, whether the facts about your business agree everywhere they appear, and whether structured data and third-party proof back those facts up. Anything sold as a GEO audit that skips the first or the fourth is a content plan with a new name.
Below is the checklist we work from. Our free GEO scan automates the first check and parts of the third and fifth in about a minute; the rest is done by hand, and in a Deep Dive it sits alongside the SEO, advertising and sales review.
1. Access: can the engines reach the site?
- robots.txt rules for GPTBot, ChatGPT-User, OAI-SearchBot, ClaudeBot, Claude-SearchBot, Claude-User, PerplexityBot, Perplexity-User, Google-Extended, Applebot-Extended, Bingbot and CCBot. Allowed, blocked, or not mentioned.
- Firewall and CDN rules (Cloudflare bot management, security plugins, country challenges) that block or challenge those crawlers regardless of robots.txt. Test with a real fetch, not by reading the config.
- Whether the important content is in the HTML source or loaded by JavaScript after the page renders. Counters that start at zero, tabs that load on click, and headlines injected by a page builder all read as missing.
- Page speed and server errors: a crawler that times out gives up.
- An llms.txt file at the root, and whether the sitemap lists only the pages worth reading.
2. Visibility: what do the engines say today?
- Ten buyer questions, written the way a customer would type them (“who is a good dermatologist in Chattanooga that takes new patients”, not “dermatologist chattanooga”), asked in each engine.
- For each: were you named, were you cited with a link, which page was cited, and who else was named. This is the baseline everything later is measured against.
- Brand questions: “what is [company]”, “is [company] legitimate”, “[company] reviews”. What the engine says about you when asked directly, and whether it is accurate.
- The first-party reports: AI feature impressions in Google Search Console, and Copilot citations and grounding queries in Bing Webmaster Tools.
3. Answerability: can a page be quoted?
- Does each key page answer its title question in the first 60 words, in a sentence that stands alone?
- Are headings real questions or plain statements, and does the first sentence under each answer it?
- Specificity count: how many numbers, names, dates and places versus how many adjectives. “Award-winning” is an adjective. “Silver ADDY, 2018” is a fact.
- Is the business named in the third person at least once per page, so a quoted passage still identifies you?
- One topic per page. A page that tries to rank for six services is quotable for none.
- Dates: does the page say when it was written or updated, and is the content still true?
4. Consistency: do the facts agree everywhere?
- Name, address, phone, website, hours, category and description on Google Business Profile, Bing Places, Apple Maps, LinkedIn, Facebook, Instagram, Yelp, BBB, Clutch and the industry directories, compared against the site.
- Old addresses, old phone numbers and discontinued services still live on third-party sites and in old press.
- Pricing or claims on one profile that the site does not make.
- Whether the site’s own pages agree with each other: the footer, the about page, the contact page and the schema should say the same thing.
5. Proof: structure and third parties
- Structured data present and valid: Organization with sameAs links, LocalBusiness or ProfessionalService per office, Service per offering, FAQPage where questions are answered, Article with author and date on posts, Person for named experts. Checked with Google’s Rich Results Test and the Schema.org validator.
- Review count, rating and recency on Google and the review sites that matter in your industry, against the competitors the engines name instead of you.
- Independent mentions: press, interviews, podcasts, client case studies, partner pages, awards. How many, how recent, and whether they say the same thing the site says.
- Links from other sites, because the engines retrieve from pages that rank, and links still decide much of that.
What a good audit hands you
Not a score. A list, in order: the blockers (a crawler that cannot get in, content that is not in the HTML), the fast fixes (listings that disagree, a homepage that never says what you do), then the slower work (pages rewritten around questions, proof earned from third parties). Each item should say what to change, where, and what it is expected to move. Then the ten questions get asked again in 60 to 90 days, and the baseline from check 2 tells you whether it worked.
What to be suspicious of
An audit that skips crawler access. A score with no list of what to change. A recommendation to publish large volumes of AI-written content, which is the opposite of specific. A promise about a position, since generative answers do not have positions. And any audit that does not ask the engines your actual buyer questions, because that is the only measurement that matters.
Start with the free GEO scan for checks 1, 3 and 5. For the rest, and for how it fits with your SEO, advertising and sales, that is the Deep Dive. The background is in our generative engine optimization guide, and the step-by-step fixes are in how to optimize for generative search.