What Is a GEO Audit? What It Actually Checks

By Muntazir · Published 11 September 2026

Run two audits of the same homepage on the same afternoon — one free, one paid — and the free one can come back with the higher score. Not because the site has fewer problems. Because the free tool checked less of it, found less to complain about, and still reported a number as though it had looked everywhere the paid one did. That gap, between what an audit checked and what it claims to have checked, is most of what separates a page calling itself a “GEO audit” from one that only sounds like it.

What is a GEO audit?

A GEO audit checks whether a business is positioned to be found and cited by AI search tools — ChatGPT, Perplexity, Google’s AI Overviews — the same way a technical SEO audit checks whether a site is positioned to rank on Google. The name borrows its first half from generative engine optimization, a term researchers at Princeton University and IIT Delhi introduced in a 2024 paper measuring which page-level changes actually moved a source higher in AI-generated answers (arXiv:2311.09735).

“SEO audit” has decades of shared convention behind it: practitioners roughly agree what belongs on the checklist. “GEO audit” doesn’t have that yet — the term is a couple of years old, and nobody has published a standard defining what the name requires. Two tools using it can check almost entirely different things. What tends to show up regardless: crawlability for the crawlers AI systems actually use, structured data, E-E-A-T signals, and — on an audit that goes deep enough — a look at what an AI model actually says when it’s asked about the business.

How is it different from a normal SEO audit?

Less different than the name suggests, for most of the checklist. Crawlability, schema markup, page structure and content clarity are SEO fundamentals a GEO audit inherits rather than reinvents — a site that’s unreadable to Googlebot is unreadable to an AI crawler for the same reasons.

The first real divergence is who’s doing the reading. Google runs one crawler against one set of signals. AI systems run several separate crawlers with separate jobs — one that feeds a model’s training data, one that builds the index a live search feature queries, one that fetches a page on demand when someone asks a model to open a link — and a site’s robots.txt can allow one and block another without anyone noticing, until an AI tool stops citing the site while Google keeps ranking it exactly as before.

The second divergence is model-probe testing: actually asking an AI model, directly, what it knows about a business and whether it would recommend it. Nothing in a conventional SEO audit does this, because Google has no equivalent black box to interrogate. A GEO audit that skips this step is checking the setup for AI visibility without ever checking the result. For how a model reads the rest of a page once it’s past the crawl, the AI SEO guide covers the broader mechanics.

What does a free GEO audit check vs. a deep one?

This is where the score gap from the opening actually comes from, and it’s worth being specific rather than leaving it as a vague asterisk.

A free run skips the crawl, the browser render, page-speed testing and most off-site sources. A small handful of off-site checks — a public knowledge-base lookup, a safe-browsing check, a public backlink-authority score — run on both tiers regardless of what’s paid for. Because a run is scored against the full checklist rather than against whatever it chose to check, a free report that verified less of the site gets its score discounted by the share it actually covered, and carries no letter grade at all — just a provisional number out of 100, capped below the maximum, alongside a note on how many checks it never reached. That’s also why a free report can land above a deep report of the same site: the two numbers aren’t scoring the same thing.

The other gap is the model-probe stage described above. A free scan does not run it, at all — asking a model what it already believes about a business is deep-tier only. A business relying on a free scan for AI visibility gets the setup checks and none of the “what does a model actually say” checks. A report that implied otherwise would be misleading in exactly the direction that matters most to a reader trying to decide whether to trust the score. The full checklist behind both tiers, not just a summary of it, is public in the methodology.

Can you run one yourself?

The setup half, yes, in about fifteen minutes. Check robots.txt for rules that treat AI crawlers differently from Googlebot. View the page source and confirm structured data is present and agrees with what’s visible on the page. Search the business name directly in a model and read what comes back — does it describe the real business, or hedge and guess from whatever the name suggests? A free scan runs the setup half automatically and reports back the same checks in one pass.

The result half is harder to do properly alone. One question to one model on one day is a sample of one, and models don’t answer identically twice — a reliable read needs several models, several phrasings of the question, and several runs of each, specifically to catch the run-to-run variance rather than mistake one lucky or unlucky answer for a stable fact. That sampling cost is exactly why it’s the part a free tool skips and a paid one doesn’t.

How often should a business re-audit?

Quarterly is a reasonable baseline for a stable site — neither Google’s crawl cycle nor an AI model’s training cycle moves fast enough to reward checking much more often than that. Re-check sooner after anything that changes what a crawler sees: a redesign, a schema update, a new robots.txt rule, or a rebrand that touches the business name across the site. Those are the changes most likely to shift a result before the next scheduled check would have caught it.

None of this makes “GEO audit” a settled category yet. It’s closer to where “SEO audit” stood before search engines published enough documentation for practitioners to converge on roughly the same checklist. Until something similar happens here, the more honest version of a GEO audit is the one that states plainly what it checked and what it skipped — not the one that came back with the higher number.