Open this article in your favourite AI assistant
If you’ve been quoted for a generative engine optimisation audit and you’re not entirely sure what you’re paying for, you’re not alone. Many agencies are now repackaging existing SEO audit templates under an AI-sounding name, without actually testing how a business appears in tools such as ChatGPT, Gemini, Perplexity or Google’s AI Overviews. A genuine GEO audit looks at different signals to a standard SEO audit, tests different systems, and produces a different kind of output. This guide breaks down exactly what should be included, how each component works, and how to read a proposal critically before you commit any budget to it, so you can tell the difference between a proper audit and a relabelled technical SEO review.
How a generative engine optimisation audit differs from a standard SEO audit
A standard SEO audit is built around how Google’s search crawler and ranking systems interpret a website. It checks indexation, site speed, internal linking, keyword targeting, backlink profiles and on-page optimisation, all aimed at improving position within the traditional ten blue links. A generative engine optimisation audit asks a different question: can AI systems access, understand and confidently cite this business when generating an answer to a user’s query, rather than simply listing a link to it.
This distinction matters because generative AI tools don’t rank pages in the way Google Search does. They synthesise an answer from multiple sources, often without showing a full list of results, and the business either gets mentioned within that answer or it doesn’t. An audit designed for this environment needs to test retrievability, comprehension and citation behaviour, not just crawlability and rankings.
What a standard SEO audit typically checks
Most SEO audits focus on technical health, on-page signals and link authority. They’ll typically cover site architecture, canonicalisation, Core Web Vitals, title tags and meta descriptions, keyword cannibalisation, and backlink quality. These checks remain relevant for traditional organic visibility, but none of them confirm whether an AI model can actually extract a usable, accurate answer from a page.
What changes when AI systems become the audience
When the audience shifts from a search crawler to a large language model, the audit needs to test things that never mattered for classic SEO: whether AI crawlers are blocked at server or CDN level, whether content is structured in a way that supports extraction rather than just readability, whether the business is actually mentioned when you ask an AI tool questions it should be able to answer about that business, and whether structured data correctly identifies who the business is, what it does and how it relates to other entities online.
| Area checked | Standard SEO audit focus | GEO audit focus |
|---|---|---|
| Crawl access | Googlebot indexation and crawl budget | Access for GPTBot, Google-Extended, PerplexityBot and similar AI crawlers |
| Content format | Readability and keyword relevance for ranking | Structure that supports extraction into an AI-generated answer |
| Visibility testing | Rank tracking against target keywords | Direct prompt testing to see if the business is cited by name |
| Structured data | Rich result eligibility (FAQ, review, product snippets) | Entity clarity and consistency across schema, web and third-party sources |
| Success signal | Position and organic traffic | Frequency and accuracy of citation within AI-generated answers |
The four pillars a proper GEO audit should cover
Regardless of the agency delivering it, a thorough audit should be built around four interconnected pillars. Leaving any one of these out produces an incomplete picture, because each pillar tests a different point of failure between your content and an AI-generated answer.
- AI bot crawlability – confirms whether AI crawlers can physically reach and retrieve your content in the first place.
- Content structuring – assesses whether the content, once retrieved, is written and formatted in a way a model can parse and reuse accurately.
- Citation testing – checks whether the business is actually being mentioned or recommended when real prompts are run against major AI platforms.
- Schema and entity review – verifies that structured data and cross-platform information consistently identify who the business is and what it’s known for.
A credible audit treats these as a diagnostic sequence rather than four unrelated checklists. If a crawler can’t access a page, the structuring review is largely academic. If structuring is poor, citation testing will simply confirm the business is invisible without explaining why. The order matters for interpretation, even if the testing itself happens in parallel.
Step-by-step: how auditors test AI bot crawlability
Crawlability testing for AI systems is more involved than checking a single robots.txt file, because different platforms use different crawlers for different purposes, and some sites block them unintentionally at server or CDN level rather than through robots.txt alone.
- Identify which AI crawlers are relevant to the business, based on the platforms customers are likely to query (commonly GPTBot and ChatGPT-User for OpenAI, Google-Extended for Gemini and AI Overviews, and PerplexityBot for Perplexity).
- Review the robots.txt file line by line to check whether any of these user agents are explicitly disallowed, either deliberately or by inheritance from a broad disallow rule.
- Check server and firewall logs, or request a log sample from the hosting provider, to see whether these crawlers have actually attempted to access the site and what response code they received.
- Test key pages directly using a user-agent switcher or server-side log filtering to confirm the content returned matches what a normal browser sees, ruling out cloaking or bot-specific blocking at CDN or WAF level.
- Check for JavaScript-dependent content that may render fine for a browser but return as empty or incomplete HTML to a crawler that doesn’t execute scripts.
- Document which crawlers are blocked, which are allowed, and which pages return errors or incomplete content, as the basis for remediation recommendations.
| Crawler | Associated platform | What to check |
|---|---|---|
| GPTBot | OpenAI training and retrieval | Robots.txt directives and server-level blocking rules |
| ChatGPT-User | Live browsing within ChatGPT | Whether real-time fetch requests succeed on key pages |
| Google-Extended | Gemini and AI Overviews | Separate disallow rules distinct from standard Googlebot |
| PerplexityBot | Perplexity AI | Access to product, service and FAQ-style pages specifically |
A site can rank perfectly well in traditional search while being quietly invisible to one or more of these systems, usually because a security plugin, CDN rule or legacy robots.txt line was configured years before AI crawlers existed and nobody has reviewed it since.
What a content structuring review actually involves
Once access is confirmed, the next question is whether an AI system can accurately extract and reuse the information on the page. This isn’t about writing for robots instead of people; it’s about removing ambiguity so a model doesn’t have to guess what you mean.
Structural signals that genuinely help extraction
- Clear, descriptive headings that state the actual question or topic rather than a vague or clever phrase.
- Direct, self-contained answers near the top of a section, before supporting detail and caveats.
- Consistent terminology for the same concept, rather than switching between synonyms that a model may treat as different entities.
- Defined facts such as pricing ranges, service areas, opening hours or eligibility criteria stated plainly rather than implied through marketing language.
- Logical heading hierarchy (H2 then H3) that reflects the actual structure of the content rather than being decorative.
As an illustrative example, a page explaining “same-day appliance repair” that only says the service is “fast and reliable” gives a model nothing concrete to cite. A version that states the typical response window, which postcodes are covered, and which appliance types are included gives the model specific, quotable facts it can safely reuse in an answer. This is a demonstration of the principle rather than a documented result from a real client.
How citation testing across platforms works in practice
Citation testing is the part of the audit most people assume is simple, but doing it properly requires a structured, repeatable method rather than a handful of casual prompts.
- Build a list of realistic prompts a potential customer might type, covering direct brand queries, category queries (“best [service] in [area]”), and comparison queries against named competitors.
- Run each prompt across the platforms most relevant to the business, typically ChatGPT, Gemini, Perplexity and Google AI Overviews where available.
- Record whether the business is mentioned, how it’s described, whether the description is accurate, and whether a link or citation is included.
- Repeat the same prompts after a short interval, since AI-generated answers can vary between sessions, to check whether mentions are consistent rather than a one-off.
- Compare results against two or three named competitors using identical prompts, to establish relative visibility rather than viewing the business in isolation.
For example, a local accountancy practice might test the prompt “who are reliable accountants for small businesses in Leeds” across each platform. If a competitor is consistently named with an accurate description and the practice itself never appears, that’s a specific, actionable finding, not a vague visibility problem.
| Platform | Test method | What a reasonable outcome looks like |
|---|---|---|
| ChatGPT | Direct and category prompts with browsing enabled | Accurate mention with correct service description |
| Google AI Overviews | Search queries likely to trigger an AI summary | Inclusion in the summary or linked source list |
| Perplexity | Comparison and recommendation-style prompts | Citation with a working link to the correct page |
| Gemini | Conversational and location-based prompts | Accurate description consistent with the website |
Schema and entity review: the practical checks that matter
AI systems lean heavily on structured signals to disambiguate who a business is, particularly when the business name is generic or shared with other organisations. This part of the audit checks whether schema markup and entity information are complete, accurate and consistent across the sites the business doesn’t fully control, not just the website itself—a prerequisite that assumes your site’s technical SEO foundations are already in place.
This typically includes validating Organisation, LocalBusiness, Service and FAQ schema against current specifications, checking that the business is described identically across its website, Google Business Profile, LinkedIn and relevant directories, and confirming that any Knowledge Panel or third-party database entries don’t contradict the primary site. Because this area blends technical markup with broader entity consistency, it’s often the part of the process most agencies underinvest in, which is why it’s worth asking specifically how a provider approaches it. If you’re comparing proposals, a generative engine optimisation audit should set out exactly which schema types will be validated and which third-party platforms will be checked for consistency, rather than offering a generic technical review.
Common schema and entity inconsistencies found during audits
Mismatched business names between schema and directory listings, outdated address or phone details carried over from a previous office location, missing sameAs links connecting the website to verified social and business profiles, and service schema that lists offerings the business no longer provides are all common findings. Each of these creates ambiguity that makes it harder for an AI system to confidently attribute information to the correct entity.
Evaluating a GEO audit proposal before you commission one
Because this is a relatively new service category, proposals vary enormously in depth and quality. The following framework helps you assess whether a quoted audit is genuinely built for AI visibility or simply an SEO audit with different branding.
- Ask whether the proposal names specific AI crawlers (GPTBot, Google-Extended, PerplexityBot) rather than referring vaguely to “AI bots”.
- Ask whether citation testing involves running actual prompts against named platforms, with documented results, rather than a general statement about “AI visibility”.
- Ask which schema types will be validated and whether third-party listing consistency is included.
- Ask what the final deliverable looks like: a prioritised list of findings with recommended actions, or a generic scorecard with no clear next steps.
- Ask how findings will be measured again after implementation, since a one-off audit without a re-test point makes it difficult to judge whether anything changed.
| Red flag in a proposal | What it usually suggests | Question to ask the provider |
|---|---|---|
| No mention of specific AI crawlers by name | Crawlability testing may not be included at all | Which crawlers will you test access for, and how |
| “AI visibility score” with no methodology explained | Citation testing may be superficial or automated with no manual review | What prompts will be run, and across which platforms |
| Schema check limited to a validator tool screenshot | No cross-platform entity consistency review | Which third-party sources will be checked against the website |
| Deliverable described only as a “report” | Findings may lack prioritisation or practical next steps | Can you see a sample of a previous deliverable |
What to change about your approach after reading this
If you’re currently evaluating proposals, the practical shift is to stop asking “do you do AI SEO” and start asking for specifics against each of the four pillars covered above. Request a short scope document before signing anything, listing which crawlers will be tested, which prompts and platforms will be used for citation testing, which schema types will be validated, and what the final report will contain. If a provider can’t answer these points clearly, that’s a reasonable signal the audit hasn’t been built specifically for generative engines.
It’s also worth setting your own baseline before the audit begins, rather than relying solely on the auditor’s findings. Spend twenty minutes running five or six realistic prompts about your own business across ChatGPT, Gemini and Perplexity, and note down what comes back. This gives you a personal reference point to compare against the audit’s citation testing results, and makes it much easier to judge whether the findings presented to you are accurate and complete.
Frequently asked questions
Is a generative engine optimisation audit a replacement for a technical SEO audit
No. The two serve different purposes and test different systems. A technical SEO audit remains relevant for Google’s traditional search ranking, site health and organic traffic. A GEO audit specifically tests AI crawler access, content extractability and citation behaviour across generative platforms. Most businesses benefit from maintaining both, since traditional search and AI-generated answers currently coexist and often draw on overlapping but distinct signals.
How long does a proper GEO audit typically take
This depends heavily on the size of the website and how many platforms are being tested, since citation testing across multiple AI tools and prompt sets takes time to run and document properly. A small business site with a handful of key pages will generally take less time than a larger site with multiple service lines, multiple locations and a broader competitor set to benchmark against.
Can I run parts of this audit myself before paying an agency
Yes, to a reasonable extent. You can check your robots.txt file manually, run a handful of prompts across ChatGPT, Gemini and Perplexity to see how your business is described, and review your schema markup using a free validator tool. What’s harder to replicate without specialist tools is server log analysis for crawler access, systematic prompt testing at scale, and a full cross-platform entity consistency review, which is where a professional audit adds the most value.
Will fixing the issues found in a GEO audit guarantee my business appears in AI answers
No credible provider should promise this. AI platforms control their own retrieval and generation processes, and citation behaviour can vary between sessions and over time as models are updated. What a proper audit and subsequent remediation work can do is remove known technical and structural barriers, and improve the clarity and accuracy of information available for these systems to draw on, which is a reasonable and achievable goal rather than a guarantee.
Do I need a GEO audit if my business operates purely locally
Local businesses are frequently the subject of AI-generated recommendation queries, such as “best plumber near me” or “reliable accountant in [town]”, which makes crawlability, structuring and citation testing just as relevant as for national brands. The main difference is usually in the schema and entity review, where consistency across Google Business Profile, local directories and the website becomes particularly important for disambiguation.
How often should a GEO audit be repeated
There’s no fixed industry standard for this yet, since generative AI platforms update their models and retrieval methods at their own pace rather than on a public schedule. A sensible approach is to treat the initial audit as a baseline, re-test citation visibility every few months using the same prompt set, and commission a fuller re-audit whenever a significant site change occurs, such as a rebuild, migration or major content restructure.
Next steps before you commission an audit
Before signing off on any proposal, write down the specific AI platforms your customers are most likely to use when searching for a business like yours, and ask the provider to confirm their testing covers exactly those platforms rather than a generic default list. Request a written scope that names the crawlers to be checked, the schema types to be validated, and the format of the final deliverable, and keep your own baseline prompt results so you can independently judge the accuracy of what comes back. Treating the audit as a defined, checkable deliverable rather than a vague service will make it far easier to tell whether you’ve bought something genuinely useful.