A GEO audit measures a website’s ability to be understood, cited and recommended by artificial intelligence-based search engines — ChatGPT, Perplexity, Google AI Overview and Gemini. It examines the site’s stated identity, the depth of its content, its technical foundations and the signals that these engines use to recommend one business over another.
This guide describes the method we apply to over 1,500 websites: what to look at, in what order, and — this is the part people rarely read — what a GEO audit cannot tell you.
What is a GEO audit?
Traditional SEO optimises a page so that it appears in a list of links. GEO (Generative Engine Optimisation) optimises a website so that a language model mentions it in a written response. The difference is not merely cosmetic: in a list of ten links, coming eighth still has some value; in a response that names three establishments, coming fourth doesn’t count.
A GEO audit therefore answers two distinct questions, which must never be confused:
- Is my website usable by an AI engine? This is a question of suitability: is the information present, structured, consistent and extractable? It is measured entirely on the website itself, and it is actionable.
- Is my establishment actually listed? This is a question of outcome. It is measured by querying search engines, and it depends on a factor beyond your control: competition in your market.
A proper audit distinguishes between these two measures. Mixing them together produces a single figure that rewards luck as much as hard work.
GEO audit and SEO audit: what really changes
| SEO audit | GEO audit | |
|---|---|---|
| Objective | To appear in a list of results | To be mentioned in a written response |
| Unit of measurement | Position, impressions, clicks | Mention or no mention |
| Relevant position | The top ten results | The two or three names mentioned |
| Featured content | Keywords, backlinks, authority | Verifiable, structured facts that are consistent across sources |
| Role of markup | Useful for rich snippets | Crucial: this is the form that the model reads unambiguously |
| External consistency | Secondary | Central: name, address and telephone number must be identical everywhere |
The two disciplines have a lot in common: a slow website, one blocked to webcrawlers or lacking content, is not good for either. We detail the differences in our article ‘SEO and GEO: five key differences’.
The four pillars to audit
Our scoring system allocates 100 points across four equally weighted pillars. This allocation is not a matter of opinion: it is generated by the code that actually calculates the scores, and each criterion is worth a fixed number of points.
1. Identity — does the search engine know who you are?
This is the most frequently overlooked and most rewarding pillar. A search engine that cannot identify your business, your name and your location with certainty will not recommend you for any local search queries.
- The title tag must specify your business and location. ‘Welcome to our site’ means nothing to a machine.
- The meta description carries significant weight in our scoring system because it is often the only summary sentence a search engine finds on a homepage.
- Just one H1 heading, and make it clear.
- Schema.org markup, with the correct type and fields actually filled in — not just an empty shell.
- NAP: name, address, telephone number, email, and above all, consistency with your external listings. If there are two different numbers between your website and your Google listing, the search engine will be suspicious.
2. Content — do you have enough to be cited?
A language model quotes whatever it can extract. A site with just three hundred words offers nothing to extract, however beautiful it may be.
- The depth of the content, and its uniqueness from one page to the next: duplicate descriptions count as a single page in the eyes of a search engine.
- A properly nested heading hierarchy: an H3 under an H1, with no H2, disrupts machine reading.
- Extractability: lists, tables, short, self-contained answers. A fifteen-line paragraph is difficult to index.
- Signs of expertise (E-E-A-T): who is writing, for how long, and with what references.
- Image alt attributes, each distinct from the others.
- Open Graph tags, internal linking, a dynamic blog.
3. Technical — can the search engine read your content?
This pillar doesn’t win you any customers on its own, but failing to get it right will cost you a lot. A site that’s invisible to crawlers is invisible, full stop.
- HTTPS with no mixed content, and a correct sitemap and robots.txt.
- Server-side rendering: an empty JavaScript shell contains no text for a bot that does not execute the script — and most AI bots do not execute it.
- Speed, in lab tests and, above all, in real-world data.
- Mobile optimisation, canonical tags, hreflang if you have multiple languages, and lazy loading of images.
4. GEO Readiness — the signals that trigger a recommendation
This is the pillar specific to GEO: what a search engine looks for when it has to choose who to recommend.
- Semantic scope of the business, explicit distinguishing features, thematic pages.
- Identifiable booking or contact details, GPS coordinates, map.
- Displayed customer reviews, certifications and ratings, social media.
- FAQs: the format most readily reusable by a response engine.
And the fifth metric: the AI footprint
Beyond the four pillars, we actually query four search engines and note their responses: are you listed, at what rank, and is your address mentioned? This indicator is separate from the score out of 100, and the following section explains why.
The method, in seven steps
- Check that the page being audited is indeed your website. A firewall that returns an access denied response will result in a disastrous diagnosis for a page that isn’t yours. Start by checking the HTTP response code.
- Explore several pages, not just the homepage. The uniqueness of descriptions can only be measured across a set of pages.
- Identify the page’s key elements: title, description, H1, Schema.org, NAP.
- Cross-check the NAP against external sources — Google listing, official register. Inconsistencies are not visible from the site alone.
- Assess technical aspects under real-world conditions, including real-world data.
- Query search engines with the questions a real prospect would ask, with web search enabled. A response from memory does not measure the same thing as an anchored response.
- Prioritise based on points to effort ratio, not by perceived severity.
How to interpret your score
A GEO score is only useful if you know what it does not measure. Three rules for interpreting it:
A criterion that cannot be measured is not a failed criterion. If no field data exists for your site, or if you have only one page, the corresponding criterion must be removed from the denominator — not given a score of zero. An audit that assigns a score of zero to what it has been unable to measure produces an inaccurate score.
Compare websites, not time periods. As soon as a scoring system changes, two scores separated by that change are no longer comparable. That is why we version our own system and refuse to show progress through a change in the rules.
A good score is a prerequisite, not a guarantee. That is the next point, and it is the most important one.
What a GEO audit won’t tell you
We measured, across 630 websites, the correlation between being cited by an AI engine and the quality of the work done on the site. It is +0.008 — in other words, zero. On the same websites, the correlation between being listed and the density of the local market is −0.55.
Put simply: an impeccable Parisian hotel may not be listed by any search engine, whilst a decent campsite in a small town is listed systematically. This isn’t a fault with the site; it’s down to the number of competitors the search engine has to rank.
Three practical implications:
- Never judge your work solely on the citation rate. This is why our AI footprint is a separate indicator from the score out of 100.
- In a crowded market, the query entered into the search engine matters just as much as the website itself: ‘a hotel in Paris’ yields no useful results, whereas ‘a hotel in the 8th arrondissement’ does.
- The four pillars remain the only lever you control. They do not guarantee a ranking; their absence makes it impossible.
Where to start
In order of observed effectiveness: identity first — it’s quick to fix and determines everything else —, then the technical foundations that prevent the page from being read, then the content, and finally the recommendation signals.
You can obtain this comprehensive analysis free of charge: run a GEO audit of your website. The report details each criterion, its score and the corresponding action.
Industry-specific guides are also available for campsites, hotels, restaurants, holiday cottages and B&Bs, solicitors and estate agents.
Frequently Asked Questions
How long does a GEO audit take?
A full automated audit — site crawl, analysis of the four pillars and search engine queries — takes a few minutes. The human analysis of priorities takes around an hour.
How often should a GEO audit be repeated?
Every three to six months for a stable website, and after every redesign. AI engines, however, change more quickly than that: it is the difference between two measurements that provides insight, not the absolute value of a single one.
Does a good GEO score guarantee that you’ll be cited by ChatGPT?
No, and this has been measured: the correlation between website quality and being cited is virtually zero, whilst the correlation with market density is strong and negative. A good score is a necessary condition, but not a sufficient one.
Should we abandon SEO in favour of GEO?
No. The foundations are the same — content, technical aspects, consistency — and AI engines rely heavily on the web index. GEO adds another layer; it does not replace anything.
Does GEO also apply to businesses without a physical premises?
Yes. The location and booking criteria do not apply, but identity, content depth, signals of expertise and structure count just as much, if not more.