ContextLumenBook a conversation

OUR METHODOLOGY

Clear evidence.
Clear limits.

A website can be technically readable without appearing in the answers your customers see. We assess those questions separately and explain exactly what each result supports.

01 / THE FREE WEBSITE CHECK

What a page lets us observe.

The current automated checker inspects a public HTTPS homepage and its robots.txt rules. It reads the initial HTML without executing the website’s JavaScript, signing in, or changing the site.

  • Page access, limited same-site redirects, and crawl policies.
  • Page-level indexing controls, title, description, and headings.
  • Readable text and JSON-LD presence or syntax.

Each report identifies its scope, method version, timestamp, source, evidence, interpretation, and next step. A scan failure is not a zero score. A missing JSON-LD block is not an automatic failure. Word-count thresholds are screening heuristics, not platform ranking rules.

Observed A checked signal was present. This does not establish that the entire site is optimised.

Review A specific finding deserves attention. Context or owner confirmation may change the recommendation.

Note Useful context, a limitation, or an optional opportunity. Not a failed requirement.

A robots policy tells us what the published file permits. It does not prove that a real crawler reached the page. Search discovery and model-training permissions are assessed separately.

Google states that its AI search features do not require special AI files or new machine-readable markup. We do not sell llms.txt or schema as a guaranteed path to recommendations. Read Google’s guidance

The public check does not collect live ChatGPT, Gemini, Perplexity, or Google AI answers. It cannot establish brand visibility, search rankings, traffic, conversions, or the quality of every page on a website.

02 / A SEPARATELY SCOPED STUDY

What AI answers actually say.

For a visibility engagement, we agree the audience, buyer questions, competitors, platforms, location, and repeat count before collecting a baseline. We distinguish branded questions from discovery questions that do not name the business.

Each observation records the exact prompt, time, platform or model, collection method, relevant settings, answer text, and citation URLs where available. A provider-collected response is labelled as such. An API response is not silently presented as a consumer-app screenshot or a personalised user experience.

  • Mention rate: valid collected answers that explicitly name the brand, divided by valid answers in the stated sample.
  • Recommendation rate: valid answers that actually recommend the brand, rather than simply mentioning it.
  • Owned-source citation rate: valid answers with at least one citation to a verified owned domain, divided by valid answers in the sample.
  • Listed position: reported only when the answer explicitly orders recommendations. Unranked prose is not assigned an invented rank.

Reports show counts and denominators, with failed or unavailable observations listed separately. An answer without citations is different from a citation field we could not retrieve. We do not call a recommendation rate “market share” or claim it represents all AI users.

A third-party citation is an opportunity to investigate, not proof that the page should mention the client. We inspect the actual page before describing a content gap or suggesting relevant, disclosed outreach.

03 / IMPLEMENTATION AND RETESTING

Record the changes. Repeat the checks.

Your team approves the content, technical changes, and publication route. We keep a change log with affected URLs and launch dates, then repeat the agreed measurement protocol after publication.

Technical changes can be verified when deployed. Search indexing and AI answers may take longer and can vary between runs. We report the actual interval between baseline and retest, along with changes in platform behaviour, settings, or available data.

A before-and-after difference does not, by itself, prove that our changes caused the difference. We describe unchanged findings and uncertainty alongside improvements. Broader claims need stronger evidence.

04 / THE STANDARD WE WORK TO

No evidence, no outcome claim.

  • No invented testimonials, endorsements, client logos, or historical results.
  • No fabricated visibility scores or guaranteed recommendations.
  • No automatic website changes or advertising spend from a free audit.
  • No private agency records exposed to visitor accounts.
  • Demonstration data is labelled on screen and in exports.
  • Client examples are published only with permission.
Explore a sample report