What is a GEO audit?
The featured-snippet answer, in plain language. The rest of the page expands every claim.
A GEO audit, short for Generative Engine Optimisation audit, asks a different question from an SEO audit. An SEO audit asks whether a page can rank. A GEO audit asks whether a page can be retrieved, extracted, attributed and cited. It diagnoses 52 points across seven sections: AI crawler and retrieval access, the classic SEO foundations that gate AI visibility, passage and chunk structure, entity and authority, structured data, prompt visibility testing across seven AI surfaces, and measurement and monitoring. The output is a written report with every finding tiered by impact and effort, delivered in 5 to 8 business days, telling your team what to fix this week, this quarter, and not at all.
Who this is built for
The shortlist of situations where a GEO audit pays back. If you don't see yourself here, the closing note covers the edge cases.
-
01
Brands missing from category answers
You rank on Google, but ask an engine for the best option in your category and three competitors come back while you don't.
-
02
Teams with AI traffic they can't explain
Referrals from ChatGPT or Perplexity are showing up, or the new GA4 AI Assistant channel has started reporting, and nobody knows which pages are earning it or how to get more.
-
03
Sites after a rebuild, migration, or rebrand
Where crawler rules, entity signals, and third-party mentions have all drifted at once, and the answers have gone quiet since.
This audit is not built for brand new sites with no content, no reviews, and no third-party mentions; there is nothing yet for an engine to cite, and the report would mostly tell you to go and make some. If that is you, build the foundations first. The audit is worth far more once there is something to measure. If you are unsure which side of that line you fall on, send a message with your URL and we'll tell you honestly, at no charge.
How much does a GEO audit cost in 2026
One audit, one price, published on the page. Here is what you pay and exactly what you get for it.
| GEO Audit | |
|---|---|
| Price | |
| Scope | All seven sections, 52 checks, in full |
| Check count | 52 points, single tier |
| Audit time | 10 to 14 hours of manual work before the report is written |
| Engines checked | ChatGPT, Perplexity, Gemini, Claude, Copilot, Google AI Overviews, Google AI Mode |
| Competitor depth | Top 3 competitors, share of answer benchmarked prompt by prompt |
| Best for | Brands absent from category answers, teams with unexplained AI referral traffic, sites after a migration or rebrand |
| What's included |
|
| Delivery | 5 to 8 business days |
| Output | Written report, tiered GEO action plan, reusable prompt set |
* Discount offered to the first 10 clients in exchange for a review on clutch.co. 9 offers left
For the full breakdown of what is and isn't included, see the pricing page.
The seven sections of a GEO audit
Every check is manual. No API automation and no auto-scoring, because the prompt-testing work in particular requires someone to run the queries personally and record what comes back. The audit runs 10 to 14 hours before the report is written, and the weight sits deliberately on sections 3, 4 and 6, which is where the findings that change outcomes come from.
-
AI crawler and retrieval access (9 points)
The gate. Every AI vendor runs three crawler classes doing completely different jobs: retrieval bots that build the index an assistant searches at answer time, user-triggered fetchers that fire when a live prospect pastes your URL into a chat, and training crawlers that feed future models. Blocking training is a licensing decision. Blocking retrieval removes you from the candidate set entirely, and one blunt rule does both. We check all three classes separately, what your CDN actually returns as opposed to what robots.txt claims, server logs for real crawler behaviour, your Google AI feature opt-out and snippet directives, Bing indexation for Copilot grounding, raw HTML parity, and llms.txt.
-
Classic SEO foundations (8 points)
Only the SEO checks with a demonstrable effect on retrieval, framed in those terms. Indexation health, canonicalisation, sitemaps and discovery paths, speed and bot fetch reliability, status codes and redirect hygiene, title and H1 coherence, internal linking and topical architecture, and your organic ranking baseline. The relationship is asymmetric and we say so plainly: good SEO does not earn citations, but bad SEO reliably prevents them.
-
Passage and chunk structure (10 points)
The heaviest section. Retrieval systems do not read pages; they split them into passages, score each passage independently against the query, and cite the strongest ones. A page with strong average quality and weak section structure loses to a plainer page with tight, self-contained answers. We check answer-first openings, whether each section survives being read in isolation, paragraph chunking, heading literalness, explicit entity naming instead of pronouns, tables and structured lists, FAQ blocks phrased in prompt language rather than marketing language, fact density, semantic HTML and hidden content, and date hygiene.
-
Entity and authority (8 points)
Retrieval decides which passages are candidates. Entity and authority decide which candidates get trusted. Engines cross-reference across independent sources before treating a claim as verified, so on-site assertion establishes very little on its own. We audit brand and author entity definition, name consistency across every profile you own, Knowledge Graph and Wikidata presence, your footprint on the specific third-party sources the engines cite in your category, the accuracy and sentiment of what those sources say about you, and whether you publish anything original enough to be worth citing over the alternatives.
-
Structured data (5 points)
Schema does not make content citable; it removes the ambiguity that prevents citation. Article and author markup, FAQPage and QAPage, Organization and the correct Service or Product subtypes, BreadcrumbList, and a full validity and content-parity pass. One 2026 change we flag for every client: Google retired the FAQ rich result and removed FAQ support from its testing tool, so FAQ markup is now purely an AI extraction signal and has to be validated through Schema.org instead.
-
Prompt visibility testing (7 points)
Direct measurement, and the section clients read first. A set of 10 to 15 prompts built from your buyers' real language, run logged out with personalisation off, two to three times each, across ChatGPT, Perplexity, Gemini, Claude and Copilot plus Google AI Overviews and AI Mode as separate surfaces. We record whether the engines know you and describe you accurately, whether you surface unprompted when the brand is not named, citation presence per surface, your citation share against the competitors that do appear, and the full list of domains the engines actually cite for your category. That last list is your target list.
-
Measurement and monitoring (5 points)
How you keep score after we leave. Google's Search Console Generative AI report, launched June 2026, and an honest account of its limits: impressions only, no clicks or queries, three surfaces blended into one. The native GA4 AI Assistant channel added in May 2026, plus the custom channel group needed because it does not recognise Perplexity. AI crawler log monitoring, which is the leading indicator, since retrieval visits drop weeks before citations do. Source attribution carried through to your CRM. And the fixed prompt set, which only works as a baseline if it stays constant between cycles.
What typically comes back in a GEO audit
The single most common pattern across GEO audits: the site is technically accessible, the content is written well for humans and badly for passages, the entity layer is thin, and the brand is absent from the sources the engines actually cite. These are the specific findings that pattern breaks down into.
-
01
A training block that caught retrieval
One blanket rule added to opt out of AI training, scoped broadly enough that it also excluded OAI-SearchBot and PerplexityBot. The intent was to opt out of training; the effect was to opt out of visibility.
-
02
The edge overruling robots.txt
A CDN bot-management preset returning 403 to AI crawlers while robots.txt permits them perfectly. Everything looks correctly configured and the site is still absent from every answer.
-
03
The answer four paragraphs down
Priority pages opening with brand narrative and scene-setting, so a retrieval system extracting from the opening passage finds nothing but preamble. Usually fixable by adding a 40 to 75 word answer block above the existing introduction.
-
04
Sections that die in isolation
Well-written prose built on connective flow, where every section opens with "this means" or refers to its subject as "it". Retrieval strips the flow away, and the passage arrives at the engine unattributable.
-
05
Absent from every cited source
The engines draw on the same four or five domains when answering the category's questions, and the client appears on none of them. This single finding usually explains most of the visibility gap, and no amount of on-site work compensates for it.
-
06
An author nobody can verify
Content attributed to a name with no Person schema, no sameAs links, and a bio stating enthusiasm rather than experience. Typically an hour or two to fix, and the credibility layer everything else rests on.
The report is specific to the site, and the prompt log and cited source list are attached. No generic findings. No filler.
How an audit takes shape, in five steps
The same five steps for every engagement. The prompt set changes; the method does not.
-
Intake
A short brief: the site, the category, your competitors, and the question you want answered. WhatsApp or email; no calls required.
-
Prompt set
We research your category and build 10 to 15 prompts from how buyers actually phrase the problem, then agree the set with you before anything runs. It becomes your permanent baseline.
-
Engine runs
Every prompt run two to three times across five assistants and both Google AI surfaces, logged out, with each answer and cited source recorded.
-
Diagnosis
Crawler access across all three classes, passage structure on a priority page sample, entity, schema and third-party footprint. Every visibility gap traced to a cause. All manual.
-
Report
A written PDF with all 52 findings scored, the full prompt log, the cited source target list, and an action plan tiered from fix-now to de-prioritise.
Full process detail at How an Illucrum GEO audit takes shape.
Why this audit, not another
GEO is new enough that a lot of what is sold under the name is a screenshot of one ChatGPT answer and an invoice. Here is what makes this different, and what it does not promise.
-
01
Recorded, not remembered
Every prompt, engine, date and answer is logged and shipped with the report. You can re-run any claim in it yourself. An auditor's recollection of what an engine said last week is not evidence.
-
02
Causes, not screenshots
Anyone can show you that a competitor got named and you didn't. The work is tracing that back to a blocked crawler, an unrenderable page, a fragmented entity, or a missing listicle, and telling you which one to fix first.
-
03
Diagnosis, not promised citations
What this audit delivers is a diagnosis, a ranked remediation plan, and a measurement baseline. Not citations. Whether a business can realistically earn AI recommendations in its category depends heavily on an off-site footprint that takes months to build, and where that is the honest answer the report says so instead of implying a technical fix exists.
GEO audit FAQ
The questions that come up most often before a GEO audit. The full FAQ has more.
Is GEO a real discipline, or just SEO with a new name?
They overlap, and the overlap is smaller than most people assume. A classic SEO audit asks whether a page can rank. A GEO audit asks whether a page can be retrieved, extracted, attributed and cited, and only a minority of AI-cited URLs also sit in Google's top 10 for the same query. Eight of the 52 checks here are classic SEO, included because an unindexed, duplicated or timing-out page reliably fails retrieval. The other 44 are work no SEO audit does.
How can you audit something that gives a different answer every time?
By treating it as sampling rather than lookup. A fixed set of 10 to 15 prompts runs two to three times each across five assistants and both Google AI surfaces, always logged out with personalisation and memory disabled, because a session that already knows your brand produces results no prospect would ever see. Results are recorded as frequencies, not verdicts. Every run is logged with the prompt, engine, date and cited sources so you can re-run any claim yourself.
Which engines do you check?
ChatGPT as the primary platform, since it accounts for the large majority of AI referral traffic, then Perplexity, Gemini, Claude and Copilot. Google AI Overviews and Google AI Mode are tested as two separate surfaces, because they behave differently: AI Mode handles conversational follow-ups, and a site cited in one is often absent from the other. If your buyers concentrate somewhere else, tell us at intake and we add it to the set.
What if we block AI crawlers deliberately?
Then the audit records it as a deliberate position and moves on. The check is whether the block is intentional and correctly scoped, not whether it exists. Every major AI vendor runs three crawler classes: retrieval bots that build the index an assistant searches, user-triggered fetchers that fire when a live prospect pastes your URL into a chat, and training crawlers that feed future models. Blocking training is a legitimate licensing decision with little bearing on citation today. The common failure is a single blunt rule that takes the retrieval layer down with it.
Do we still need SEO as well?
Almost certainly, and eight of the 52 checks in this audit are classic SEO for that reason. The relationship is asymmetric: good SEO does not earn you citations, but bad SEO reliably prevents them. An unindexed page is in no retrieval index, content split across four URLs splits its own attribution signal, and a page that times out on a bot fetch is skipped. Where you need full coverage of those disciplines, that is a different and broader audit.
How long does the audit take to deliver?
Five to eight business days, counted from the moment we have your brief and the access we need. The audit itself runs 10 to 14 hours before the report is written: the prompt set has to be built, agreed and run repeatedly, the page sample reviewed by hand for passage structure, and the third-party source research done. A real audit cannot be compressed below that without cutting the part that produces the findings.
See the full FAQ for more questions, or read how the audit is run.
Ready to find out what the engines say?
Audits are scheduled in the order requests come in. The fastest route to a delivery slot is to message; describe the site and the question you want answered, and we will scope it from there.
Contact us