Methodology
Evidence first. Every result shows what was checked, when, and what it does not prove.
This page is the honesty contract behind every AI Visibility report.
Evidence first.
What we measured
- Publicly available pages
- Technical accessibility and indexability signals
- Structured identity and service clarity
- A locked, dated paid AI measurement cycle, when an approved order exists
- Public links and citations returned by named providers during that paid cycle
What we did not measure
- Every possible prompt
- Every user location or personal context
- Private analytics
- CRM conversions
- Revenue impact
- Guaranteed future recommendations
- Causality between one website change and one AI answer
How to compare AI visibility across markets and languages
Agree the business location, search location and question language separately. A Russian-language question about a Bali hotel is not automatically a search from Russia. If a system cannot verify the search location, record that limitation.
Keep the same questions, systems, channel and repeat count for a recheck. Save the date, full answers and source links. Report missing or failed runs separately; one answer is not evidence of a stable trend or an effect caused by a website edit.
For restaurant, hotel, spa and beach club audits, guest questions test the venue’s discovery. Questions from business owners looking for an audit test the service provider’s discovery. Keep these two prompt sets separate.
Metric definitions
What each number means.
Public Readiness
A deterministic 0-100 score built from access/indexability signals, machine-readable identity, offer and page clarity, and conversion path — every dimension shows what was actually measured.
Entity Clarity
How consistently your brand, offer, audience, geography and proof are described across your homepage, metadata and structured data.
Paid AI measurement
What an approved paid measurement cycle observed in its locked prompt set and named systems — reported with the exact denominator, date and channel, never as a universal visibility score.
Paid citation evidence
Which owned pages were cited in the same approved paid measurement cycle, with the source URL, system and run context preserved.
Conversion Path
Whether a visitor (human or machine-readable signal) can find a clear next step: contact, booking or a direct answer to a commercial question.
What is measured
Four separate layers, reported separately.
Each layer answers a different question and is reported on its own evidence. They are never collapsed into one universal score — a business can be perfectly discoverable and still be impossible to act on.
01
Discoverability
Can public search and AI systems access and find the business?
Reachability, crawler access, sitemap, canonical and indexability signals — read from public pages only.
02
Understanding
Can they understand who the business is, what it offers, for whom and where?
Brand consistency, service clarity, market and location clarity, structured-data evidence and direct-answer coverage.
03
Recommendation Evidence
Is the brand mentioned or cited in an approved, disclosed paid measurement cycle?
Reported as ratios — mentioned in X of Y valid answers, cited in X of Y — with the prompt set, environments, channel and date disclosed.
04
Action Readiness
Can a customer, or an agent acting for them, complete the next action?
Measured as three independent states: human-ready, machine-readable and agent-executable.
Action Readiness
Three independent states, not one checkbox.
Most sites are built for a human with a mouse. Increasingly, the visitor is an assistant acting on someone's behalf. These are different capabilities and we measure them separately.
Human-ready
A person can find and complete the intended action unaided.
Example: A visible, reachable booking button that works on mobile.
Does not prove: Does not prove a machine can parse the action, or that an agent could execute it.
Machine-readable
The action and its parameters are exposed in structured, parseable form.
Example: Structured action markup that names the required inputs.
Does not prove: Markup being present does not prove the action can actually be completed programmatically.
Agent-executable
An autonomous agent can complete the action end to end and receive a confirmation it can verify.
Example: A documented endpoint plus a machine-verifiable confirmation.
Does not prove: This is the strictest state and is the one most often assumed without evidence.
A visible button, a WhatsApp link or a block of JSON-LD is not, by itself, proof of agent execution. We report what each state actually demonstrates and nothing beyond it.
Local Business Mode
A profile of the same engine, not a separate scanner.
When the business model is local or hospitality, the same four layers are applied with location-aware expectations. It is a configuration, not a different product.
May include
- Website evidence, as in every other mode
- Location and service-area clarity
- Opening hours as published on the site
- Contact, booking and ordering paths
- Later, and only with verified provider capability: Google Business Profile and Maps facts
No Google Business Profile or Maps integration exists in this version. Nothing in a report is sourced from those services today, and none is implied.
Methodology
What we measured, when, and what it does not prove.
Every result in an AI Visibility report carries its evidence: what was checked, when it was checked, the sample size behind it, and the limits of what it can honestly claim.
Observed
Directly fetched from a public page: HTML, headers, structured data.
Provider
Returned by a named AI-tracking or performance provider, with the provider and date recorded.
User-supplied
Entered by you: brand name, target market, category, optional competitor.
Derived
Calculated from other evidence, always referencing the source evidence IDs.
Generated explanation
A plain-language summary of evidence — never the sole basis for a finding.
Unavailable
Not measured. Shown as "not measured," never treated as zero or as a failing score.
Supported now
- Live public page crawl (up to 5 pages)
- Indexability and structured-data checks
- PageSpeed performance evidence (when available)
- A locked eight-system measurement contract in the authenticated client workspace, with Perplexity measured through its API
- Evidence Ledger and grounded Recommendation Engine outputs
Not yet supported
- Public self-service checkout and production provider execution
- Connected Google Search Console, Google Business Profile or Instagram analytics
- Google Places (excluded from the MVP)
- Automatic recurring cycles before production activation
- Perplexity as a visitor sees it: Perplexity is measured through its API only
Every report records a methodology version, check version, prompt-set version and scoring version. If the methodology changes, older reports are not silently recalculated with new rules.
The check only reads publicly available pages. It does not log in, does not read private dashboards, and does not collect data from forms on the site being checked.
What this product does not claim
- It does not claim to know what any AI system "thinks" about a business.
- It does not guarantee rankings, citations or recommendations.
- It does not estimate lost revenue — that needs your analytics, conversion data and an explicit model.
- It does not treat llms.txt as a ranking factor; its weight is zero.
- It does not present a generic model API response as a consumer ChatGPT or Search result.
- It does not claim complete AI visibility — every sample is limited, dated and disclosed.