SiteRank AI is a WordPress plugin that does three separable jobs: it audits how machines read your site, it manages the crawler-facing resources those machines use, and it observes what AI answer engines say about you for a prompt set you approve. The audit, clustering, crawler and structured-data work runs entirely inside your install with no outbound request and no content leaving the server. Only visibility monitoring needs a provider API key, and that key is yours.
The design rule behind every screen is that a number you cannot decompose is not shipped. There is no 0–100 grade, no "AI score", and no "ChatGPT ranking". Each area reports a status band and the counts it came from, and every finding points at a URL and an observed value.
Site audit
The technical SEO audit reads your published content and applies twelve rules covering titles, meta descriptions, canonical conflicts, heading structure, thin content, duplicate titles, broken and insecure internal links, orphan pages, weak internal linking and image alt text. Each finding names the page, the observed value and the reason it was raised, and each rule carries an evidence tier and a severity.
Results roll up into one of four bands — good, needs attention, critical, or not measured — derived from severity counts per document, so a large site is not penalised for being large. The full rule list, the thresholds and the arithmetic are on the technical SEO audit page.
AI readiness
The GEO audit asks a different question from the SEO audit: not "can this be indexed" but "can a machine tell what this page is about, and can a passage of it stand on its own". Six checks cover entity clarity, machine-readable structure, passage retrievability, authorship transparency, site transparency and content freshness.
Every one of them is labelled with its evidence tier, and items at EXPERIMENTAL or HYPOTHESIS never affect a status band — an exclusion enforced in the query, not left to a developer to remember. Nothing here is presented as a ranking factor, because no provider documents one. See GEO and AI readiness.
LLM visibility
Visibility monitoring sends a fixed set of prompts to a provider API on a schedule you set, records every response with full provenance, and derives six metrics from those records: mention rate, citation rate, domain citation rate, share of voice, citation share and prompt coverage. Volatility is reported alongside them as a first-class number rather than as an error bar.
No metric is returned as a bare percentage. Each one carries its numerator, denominator, sample size, a 95% Wilson score interval and a plain-language definition. Bring your own key; only the prompt text leaves your site. The measurement model is set out on AI visibility monitoring, and the reason none of it is called a rank is on LLM rank tracking.
Prompt monitoring
The prompt set is the instrument, so it is treated like one. Prompts are written or reviewed by you, categorised by intent — informational, comparison, navigational, recommendation, brand and competitor — and versioned, because editing the wording creates a different question and breaks comparability with everything measured before.
Each prompt has its own schedule (manual, daily, weekly or monthly), its own provider and model, and its own number of samples per run, up to ten. A hard budget guard checks your monthly limit and daily run cap before each call, not after. Details on prompt monitoring.
Citation intelligence
A citation is only useful if you know where it actually points. Providers routinely return wrapper and redirect URLs — grounding-API redirects, search-engine click trackers, ?url= style forwarders — and comparing hosts before unwrapping those counts them as citations of the wrapper, which usually reports zero when your site was in fact cited.
SiteRank AI normalises every citation offline, extracting the destination where it is embedded in the wrapper. Where a wrapper is an opaque server-side redirect with no recoverable destination, it is recorded as unresolved rather than guessed. Each citation is labelled as pointing at your domain, a tracked competitor, or neither, and the domain citation rate is built only from what resolved. More on citation monitoring.
Topical authority
Topic analysis runs locally with a lexical TF-IDF engine — no embeddings, no provider call, no content leaving the install. It groups documents into clusters by cosine similarity, labels each cluster from a shared taxonomy term where one exists (a human-authored label beats a derived one), and reports coverage and connectivity bands, pillar candidates, orphan content and internal linking opportunities.
Generated labels can be wrong and the interface says so. The engine stops at 600 documents and reports that it stopped, rather than returning a partial result that reads as complete. See topical authority.
Crawler intelligence
The crawler module reads your robots.txt, parses it, and evaluates it against a registry of named crawlers grouped into five categories: search engines, AI search and answers, AI model training, user-triggered fetches, and advertising. It shows what each crawler is currently allowed to do and where the file is silent or contradictory. On the live site it flags a robots.txt that returns an unexpected status code, which is the sort of server-level fault a plugin dashboard would otherwise hide.
Policy changes are proposed as a diff and applied only when you act. The same is true of /llms.txt: SiteRank AI drafts it, validates it, previews it and publishes it on an explicit action, never automatically. See AI crawlers and llms.txt.
Structured data
Every scanned page is inspected for Schema.org JSON-LD: which types are present, which plugin or theme emitted them, and where two sources conflict — two Organization nodes with different names, or a second Article graph fighting the first.
SiteRank AI detects Yoast SEO, Rank Math, All in One SEO, SEOPress, The SEO Framework and Slim SEO, and reports where a fix belongs rather than taking ownership of output another plugin already owns. It will never generate markup that misrepresents a page, and it will never invent an author, a review, a rating or an organisation detail to fill a required field. See structured data.
Recommendations
Findings become recommendations, and recommendations have a lifecycle: open, reviewed, ignored, resolved, or auto-resolved when a re-scan finds the underlying condition gone. Each one records how it would be fixed — manually, with guidance, or assisted — and links to the affected pages.
Nothing is applied on your behalf. The plugin proposes; you decide. Anything that would change site-visible output is shown as a diff first and recorded in an activity log afterwards, and that log is redacted before storage so no credential or payload fragment survives in it.
Reporting and history
The overview screen summarises the current state of each domain with its band and counts. Underneath it, the audit, recommendations, topics, crawler, structured-data and visibility screens each expose the rows the summary was built from, filterable and paginated.
Visibility history is stored per observation, so any reported rate can be expanded into the individual responses that produced it, each with its prompt id and version, provider, model returned, web-search flag, locale, timestamp and raw payload. When the returned model or the prompt version changes mid-series, the chart marks a break instead of drawing a continuous line through a changed instrument. A diagnostics export is available for support, redacted by the same component that scrubs the logs.
What SiteRank AI will not do
- It will not give you a single number for your site's health, because no honest one exists.
- It will not tell you an experimental technique is a ranking factor.
- It will not report a position, a rank, or a place in an AI answer.
- It will not change your content, robots.txt, canonicals, schema or sitemaps without an explicit action from you.
- It will not send your content to a model. Only prompt text goes to a provider, and only when you run monitoring.
Who it is for
Site owners, in-house SEO teams and agencies running WordPress who need conventional technical hygiene and a defensible answer to "are we showing up in AI answers". It suits people who will be asked to justify a number in a meeting, because every number here can be traced to the observation that produced it.
Related reading
Frequently asked questions
Do I need an API key to use SiteRank AI?
Only for visibility monitoring. The SEO audit, the GEO audit, topic clustering, crawler inspection and structured-data inspection all run inside your install with no outbound request, so the plugin is fully useful before you configure a provider at all. The key stays yours, you supply it, and it is used only when monitoring runs.
Will it conflict with Yoast SEO or Rank Math?
No, because it does not try to own the same output. SiteRank AI detects Yoast SEO, Rank Math, All in One SEO, SEOPress, The SEO Framework and Slim SEO, and where one of them already produces a title, description, canonical or schema graph, the finding tells you which tool to make the change in rather than emitting a competing tag. The only crawler-facing files it will ever write are robots.txt and /llms.txt, and only on an explicit action from you.
What happens on a very large site?
The audit and crawler work are batched into resumable background jobs, so a large corpus is a matter of time rather than a timeout. Topic clustering is the exception: the local engine stops at 600 documents and says it stopped, instead of returning a partial map that reads as complete. That ceiling is a property of the current lexical implementation, and it is stated in the interface rather than discovered later.
Which AI answer engines can it monitor?
One provider adapter is implemented today — OpenAI — and the rest appear in the interface as not yet available, rather than as options that look configurable and quietly do nothing. Adding a provider is additive by design: model ids, endpoints and capabilities are configuration, so no analysis code changes when the next adapter lands. Whichever provider you use, only the prompt text leaves your site.
Can I point it at a site that is not WordPress?
Partly. The plugin runs on WordPress, but it can also run a bounded external scan of another site: robots.txt, sitemap, llms.txt, homepage metadata and structured data, plus a small crawl limited by page count. That is a much thinner evidence base than the full corpus audit it performs on its own install, and the two are not presented as equivalent.