Generative engine optimization is a field with more advice than evidence. Plenty of it is sensible; very little of it is confirmed by the companies whose systems it claims to influence. Google states there are no additional requirements or special optimisations for its AI features beyond normal SEO, and no provider publishes a list of things that make a page more likely to be cited.
SiteRank AI's GEO audit is built around that gap. It runs six checks on how machine-readable your site is, it labels each one with the evidence tier that supports it, and it refuses to let a weakly-supported idea affect a status band. The whole audit runs locally: no content leaves the install, and no API key is involved.
What "AI readiness" actually means here
The SEO audit asks whether your pages can be crawled and indexed. The GEO audit asks a narrower question with a longer answer: if a retrieval system pulls a fragment of this site into an answer, can it tell what the fragment is about, who published it, when, and whether the passage stands on its own?
That framing keeps the checks honest, because each one is about a property of the document that you can observe directly — a name, a date, a heading, a schema type — rather than about an outcome nobody can observe. The distinction that has to survive every claim on this page is the one in crawled is not cited: being reachable is a precondition for being retrieved, retrieval is a precondition for citation, and none of the three implies the next.
The six checks
| Check | Rule | Evidence tier | Affects status |
|---|---|---|---|
| Entity clarity | geo.entity.organization_identity |
Emerging practice | Yes, at reduced weight |
| Machine-readable structure | geo.machine_readable.no_schema |
Established standard | Yes |
| Passage retrievability | geo.structure.long_passage |
Strong evidence | Yes |
| Authorship transparency | geo.trust.authorship_missing |
Emerging practice | Yes, at reduced weight |
| Site transparency | geo.trust.site_transparency |
Emerging practice | Yes, at reduced weight |
| Content freshness | geo.freshness.superseded_date |
Emerging practice | Yes, at reduced weight |
Entity clarity
A site should say who publishes it, in one consistent form. The check compares the organisation name as it appears across pages and in structured data, and reports two conditions: the identity is absent from structured data entirely, or it appears in inconsistent forms across the site. Either way, anything reading the site has to guess which name is the real one.
This is an inference from how entity resolution works, not a documented provider behaviour, so it sits at EMERGING PRACTICE. The reasoning is developed in entity clarity.
Machine-readable structure
Pages with no Schema.org JSON-LD at all are flagged. Structured data states, in a standard vocabulary, what a page is and what it describes, which removes guesswork for any machine reading it. Schema.org is a published standard, so the check sits at ESTABLISHED STANDARD — but the tier applies to the vocabulary, not to an outcome. Structured data does not guarantee a rich result, and no AI provider documents it as a reason to cite a page. The finding is raised at low severity for exactly that reason. See structured data for how conflicting and duplicate graphs are handled.
Passage retrievability
Retrieval systems generally index passages rather than whole documents, so the unit that gets returned is usually a section, not a page. The check measures the text between headings and flags any passage running past roughly 450 words without a subheading.
Breaking such a section up does not change your content, only its addressability. This sits at STRONG EVIDENCE because passage-level retrieval is a well-documented property of retrieval systems in general — but it is not a documented behaviour of any particular AI product, and the rule's own description says so. Passage-level content structure covers the practice in detail.
Authorship transparency
Editorial content — posts, not pages — is checked for machine-readable authorship. Stating who wrote something, and marking it up so a machine can read it, helps both readers and retrieval systems attribute information.
One rule matters more than the check: SiteRank AI will never invent an author for you. If authorship is missing, supply a real one or leave the field empty. Fabricated bylines, credentials, dates or reviews are prohibited across the product, and the absence of a signal is always preferable to a manufactured one.
Site transparency
The check looks for pages that say who runs the site, how to reach them, and how data is handled, and reports whether those pages exist and are indexable. Missing more than two raises the severity from low to medium.
That is the whole measurement. It is not a trust score, no AI system publishes one, and the presence of an about page is not evidence that anything will cite you. It is a check that a reader — human or machine — can find out who is behind the source.
Content freshness
Rather than treating page age as a defect, the check flags pages whose title or headings state a year that is now in the past. A "2024 guide" sitting unedited in 2026 is a concrete, observable signal that the information may be superseded. Page age on its own says nothing about whether content is still correct, and is not flagged.
How the evidence tiers work
Every finding in SiteRank AI carries one of six tiers, and the tier controls whether it can affect a status band.
| Tier | Meaning | Affects status | Weight |
|---|---|---|---|
| ESTABLISHED STANDARD | A published specification or settled web standard | Yes | 1.0 |
| OFFICIAL PROVIDER GUIDANCE | Stated in current first-party documentation | Yes | 1.0 |
| STRONG EVIDENCE | Reproducible, publicly documented measurement | Yes | 0.8 |
| EMERGING PRACTICE | Widely practised, plausible mechanism, unconfirmed | Yes | 0.4 |
| EXPERIMENTAL | Little or conflicting evidence | No | 0.0 |
| HYPOTHESIS | First-principles reasoning only | No | 0.0 |
The exclusion of the bottom two tiers is enforced in the database query that builds the severity matrix, not left to each rule's author to remember. A check at EXPERIMENTAL or HYPOTHESIS can appear in the interface, visually separated, as something to consider — it cannot be scored as a deficiency and cannot be phrased as a fix. No GEO check currently shipping sits at those tiers; the mechanism exists so that adding one later cannot quietly inflate anybody's status band.
What this does not claim
- No GEO check is a ranking factor. None of these is documented by any provider as affecting retrieval or citation, and none is described that way in the product.
- Structured data is not a citation mechanism. It makes a page's subject explicit. That is the claim, and the whole claim.
llms.txtpresence is not ingestion. SiteRank AI drafts and publishes an llms.txt file on your explicit action, and states plainly that no major provider documents consuming one.- Nothing here measures what an AI system actually did with your site. These checks read your pages. Observing what a model says about you is a separate exercise with a separate evidence base — see AI visibility monitoring.
- Status bands are counts, not grades. The GEO domain reports good, needs attention, critical or not measured, derived from the same per-document density arithmetic as the technical SEO audit.
Who it is for
Publishers and site owners who want to do the defensible parts of GEO — clear entity identity, real structured data, addressable passages, honest authorship and transparency pages — without buying into claims nobody can support. If a consultant has told you that a technique is a confirmed AI ranking signal, this audit is the counterweight: it will tell you what tier the evidence sits at, and it will not score you against a rumour.
Official sources & further reading
- AI features and your website — Google Search Central
- Schema.org — Schema.org
- Page structure: headings — W3C Web Accessibility Initiative
Related reading
Frequently asked questions
Is GEO different from SEO?
Partly. The technical foundations overlap almost entirely; the differences are in emphasis — entity clarity, passage structure, source transparency. SEO vs GEO works through what genuinely differs and what is the same work under a new name.
Will fixing these get me cited by ChatGPT?
Nobody can promise that, and this product will not. These checks make the site easier to read correctly. Whether an answer engine cites you is a separate thing, observable only by measuring it.
Do I need an API key for the GEO audit?
No. It runs entirely inside your WordPress install, contacts no provider, and sends no content anywhere. The audit is complete without a key configured at all.
Should I publish an llms.txt file as part of this?
You can, but do not count it as a check you have passed. No major provider documents consuming one, so publishing it is a cheap bet rather than a fix, and the product treats it that way — it drafts the file, validates it, previews it and publishes it only when you act. The defensible reason to keep one is that it makes you maintain an accurate index of your own canonical pages, which is worth doing whether or not anything ingests it.
One of these checks does not apply to my site. Can I dismiss it?
Yes. Every finding becomes a recommendation with a state, and marking one ignored keeps it out of your working list without pretending the underlying condition changed. That matters most for the authorship and transparency checks, where a small brochure site may legitimately have no bylined editorial content. What the product will not do is fill the gap for you: if authorship is missing, the only honest options are a real author or an empty field.