Method
Forty criteria, eight pillars, and the ones we could not reach.
The free check reaches ten of them from one page. This says which, and what the other thirty need.
What the free check covers
One page, your robots.txt and your sitemap. From those it derives business identity, the profiles your page declares, four foundations, three readiness readings and up to three verified problems, each citing evidence.
The eight pillars
Five criteria each, forty in all. Each pillar asks one question of the site.
- T · Technical discovery
- Can a crawler reach, fetch and index the pages at all: status codes, redirects, canonicals, sitemaps, caching.
- C · Content and search relevance
- Does each page answer a real intent, and is it the page that should.
- X · Experience
- What a visitor's device measures: loading, layout stability, response to input, taken from field data at the 75th percentile where it exists.
- A · Answer usefulness
- Can a clear, correct answer be lifted from the page and attributed to it.
- E · Business identity and trust
- Does the site state, in a form a machine can read, who the business is, and do outside sources agree.
- G · AI retrieval readiness
- Which providers the site admits, refuses or leaves unstated, provider by provider, from the published controls.
- R · Reputation and external evidence
- What the rest of the web says about the business, sampled and read rather than scored.
- M · Measurement readiness
- Whether the owner can see any of this happening: analytics, search data, and the definitions behind each figure.
The six results
Every criterion returns one of six words. There is no seventh, and nothing is added up.
- Pass
- Every tested unit meets the condition, with the sampled scope stated.
- Mixed
- At least one unit passes and one fails. The exact counts are published.
- Fail
- Every tested unit fails, with the condition and the evidence identified.
- Not assessed
- No adequate evidence, or the test could not be completed. Never converted to zero, never to partial credit.
- Not applicable
- Genuinely irrelevant to this site, with the reason written down. Not a substitute for missing evidence.
- Advisory
- An opportunity or context worth knowing. Not a requirement, not a proven defect.
Three readings on every criterion
The result word above is only the first reading. Each criterion also records an evidence state (complete, limited, unavailable, collection error, or owner evidence required) and a confidence (confirmed, probable, possible, or unknown), each with its reason. Not assessed is the display of unavailable evidence, never a score of zero. Every repair then carries a priority from P0 to P3, a named owner, an acceptance test and a later outcome check.
Which profiles the review opens, and why
A profile is a unit of evidence, not a channel to be scored. The review opens every profile a site declares, in its structured data or as a link on the page, and compares each one with the site on three facts: the business name, the telephone and the postcode. A profile that names a different business is a finding. One that names the right business with a different telephone is a finding. One that cannot be read without signing in is left unassessed, with the platform named, and never counted against you.
Which platforms matter depends on who is asking. Two independent measurements of what AI assistants cite, one covering ten engines to December 2025 and one covering six months to January 2026, agree on the order and disagree on the numbers: YouTube first, Reddit second, LinkedIn behind them, with X cited heavily by Grok and little elsewhere, and TikTok, Instagram and Facebook a small share on every engine. The numbers differ because the two studies count different things, and both move month to month. So this page gives you the order and its reason, and no percentage.
- YouTube
- The most-cited social source across Google's AI features and Gemini. A transcript is text an assistant can lift; a short feed post is not.
- What people say when nobody is selling to them. Heavily used for comparison and recommendation questions. Rarely something a business builds; often something it should know is being said.
- Company facts and who runs it. Matters when the customer is another business. Articles carry further than posts.
- Google Business Profile
- Not social, and more important than any of them for a local business. Maps, the local pack, Gemini and AI Overviews all draw on it. Opened in the full review and compared on the same three facts.
- Facebook, Instagram
- Real community channels for a local business and a small share of assistant citations. Reviewed for the same three facts. Meta rarely answers an anonymous read, so the full review reads them as a person would.
- X, TikTok
- Reviewed where present. X matters if Grok's users are your buyers; TikTok if they are young or visiting. Neither is a gap for a plumber in Penrith.
- Review and directory sites
- Trustpilot, TripAdvisor, Checkatrade, TrustATrader, Yell, Nextdoor. For a local business these feed Google and the assistants more reliably than a new social account would, and they are where the name, telephone and postcode most often disagree with the website.
The rule for a local business. A Penrith firm that uses Facebook and nothing else is not told to open five accounts. The report audits the Facebook Page properly, treats Google Business Profile as the public profile to fix first, notes the directories that already list the business whether it asked them to or not, and records YouTube, Reddit, LinkedIn, X and TikTok as gaps that are not defects. The absence of a profile the business's customers do not use is not a finding. The biggest leak for a Facebook-only business is almost never the missing TikTok; it is an unclaimed Google Business Profile and a telephone number that reads differently in three places.
Citation-order sources: Goodie, Social citations in AI search, 6.1M citations across ten engines, August to December 2025; Bluefish data reported by Adweek, 26 January 2026, six months to publication. Both are vendor measurements, quoted here for the order only, and never used to set a result. Read on 20 September 2026; review by 20 December 2026, after which the order is treated as unconfirmed until re-read.
What it cannot reach
- Consistency across the rest of the site, since it reads one page.
- Whether your pages answer the questions customers actually ask.
- How your identity corroborates across maps, profiles and directories.
- Whether assistants cite you today, which needs repeated dated observation, not a scan.
- Anything held in your own accounts: Search Console, analytics, listings you have claimed.
Those are reported as not assessed, with the evidence named. Never as zero, and never folded into a score that hides them.
Where the criteria come from
Each traces to published provider documentation, with the source register re-read on a schedule. Where providers disagree or a behaviour is undocumented, the criterion says so rather than guessing.