About SiteRank AI

SiteRank AI is a WordPress plugin for technical SEO, AI-search readiness, topical structure, crawler policy and AI visibility monitoring. It was built because the tools in this category mostly report numbers their users cannot check, and because the arrival of AI answer engines has produced a great deal of confident advice with very little evidence behind it.

The product's position is simple: tell the site owner what was actually observed, say how confident anyone should be in it, and never present a guess as a measurement.

Why it exists

Two problems converged.

The first is old. SEO tooling has long reported composite scores — a number out of a hundred, a coloured dial — whose weightings are unpublished and whose movements are unactionable. The number feels like information, and it cannot be traced to a page, an observed value, or a rule you can read.

The second is new. AI answer engines changed how people find things, and an industry of advice grew around them faster than any evidence did. Techniques with no confirmation from any provider are routinely described as ranking factors. Tools report a "ChatGPT ranking" for a system that does not rank. Google states there are no additional requirements or special optimisations for its AI features beyond ordinary SEO, and a large amount of what is sold as AI optimisation quietly contradicts that without citing anything.

A site owner facing both problems needs a tool that separates what is standardised, from what is documented by providers, from what is genuinely just a hypothesis worth trying. That separation is the product.

Evidence-first GEO

Every finding and every recommendation carries an evidence tier: ESTABLISHED STANDARD, OFFICIAL PROVIDER GUIDANCE, STRONG EVIDENCE, EMERGING PRACTICE, EXPERIMENTAL or HYPOTHESIS. The tier is visible to you and it is load-bearing in the code: findings in the bottom two tiers are excluded from status calculations at the database layer, so no rule author can turn an untested idea into a reported deficiency.

Experimental ideas are still shown. They are shown as experiments — separated from standards compliance, phrased as something you might test, never as something you have got wrong.

The distinctions that hold this together are stated everywhere in the product and are never collapsed for a cleaner headline: crawlability is not retrieval, retrieval is not citation, and citation is not recommendation. An llms.txt file existing is not evidence anything ingested it. Schema markup is not a citation. One query is not a ranking.

The full specification is on the methodology page, including every formula.

Engineering philosophy

The analysis engine is deterministic and local. Content scanning, SEO auditing, GEO auditing, topic clustering, crawler inspection and structured-data inspection run inside your WordPress install with no outbound request and no content leaving the site. The core product works with no API key at all, and that is a design constraint rather than a free tier: an analysis you cannot reproduce offline is an analysis you cannot verify.

Business logic is separated from WordPress. Parsing, scoring and normalisation take plain values and return plain values, which means they are unit-tested without a WordPress bootstrap. Each audit rule is one class with one responsibility. Provider behaviour sits behind an adapter interface, so no model identifier appears in domain logic and adding a provider never means editing analysis code.

Work that touches many pages runs as a batched, resumable background job with a cursor, because a tool that behaves well on a demo site and falls over on a real one has not been built for its users. Where a limit exists, it is enforced and stated rather than producing a partial result that reads as complete.

Open standards, not a proprietary layer

The crawler-facing surface of a website belongs to published standards — the Robots Exclusion Protocol, sitemaps, Schema.org vocabulary, HTTP semantics — and SiteRank AI works in those terms rather than inventing its own. robots.txt is inspected against a registry of search, AI-search, AI-training and user-triggered crawlers and explained directive by directive. Structured data is read as JSON-LD in Schema.org types, with conflicting or duplicated graphs reported rather than silently overwritten. llms.txt is treated as what it is — a proposal with real adoption questions — and described that way, never as a requirement.

Nothing it produces is locked to it. Remove the plugin and your site keeps whatever you published.

User control

The plugin proposes; you decide. It does not rewrite content, canonicals, metadata, schema, sitemaps, robots.txt or llms.txt on its own. Anything that changes site-visible output requires an explicit action from you, shows you the difference first, is reversible, and is recorded.

It also assumes it is not alone. Most WordPress sites worth auditing already run Yoast, Rank Math, All in One SEO, SEOPress, The SEO Framework or Slim SEO, and a second plugin emitting canonicals or schema damages a site rather than improving it. SiteRank AI detects what is installed, reports where a fix actually belongs, and takes ownership of an output only when nothing else owns it and you have explicitly granted it.

Data follows the same principle. Your data stays in your database, credentials are your own, and uninstalling preserves your data by default — destructive cleanup is opt-in.

Transparency

Every reported figure decomposes. A status band opens into severity counts, counts open into findings, and a finding names the URL and the observed value that produced it. Every visibility metric shows its numerator, denominator, sample size and a 95% confidence interval, never a bare percentage. Failed API calls are excluded from rates and the excluded count is reported. Cost figures are labelled as estimates and are floors.

The formulas are published rather than described. Definitions are versioned, and stored results carry the version that produced them, so a change to a definition cannot silently rewrite what you saw last month.

What it deliberately refuses to do

• No composite 0–100 score. Status bands with published counts instead.
• No "ChatGPT ranking", no "position" in a generative answer. Those things do not exist and naming them would be the first lie.
• No scraping of consumer chat products to simulate an API. It breaches provider terms and the resulting number is no more faithful.
• No fabricated measurement. Nothing modelled, estimated or interpolated is ever presented as something observed.
• No generated trust signals. No invented authors, credentials, reviews, testimonials, dates or organisation details. An unknown field stays empty and asks you.
• No content generation at scale. The product does not mass-produce pages, and it will not make scaled low-value content easy to create.
• No silent site changes, and no keys of ours: you bring your own provider key and nothing is proxied through our infrastructure.

Honest status

SiteRank AI is early software, and its limitations are documented rather than discovered: one LLM provider is implemented, the topic engine has a document ceiling, and interoperability with other SEO plugins is implemented by detection but not yet exercised against every live install.

This page makes no claims about customers, users, funding or awards, because inventing them would contradict the entire product. To judge it, read the methodology and the editorial policy, and check whether the numbers do what those pages say.