Visible in the answer, or verifiable around it.
A name, a rank, a displayed source, a recommendation, a page’s indexing, an identified referral. Recorded, archived and counted.
PUBLIC READING FRAMEWORK · KAIS
KAIS AI Visibility Chain™
Reading visibility in AI answers.
Four readings and nine dimensions, from being found to being contacted. They separate what is observed in AI answers, what is inferred from them and what is only a signal. A reading framework: neither a promise of results nor a description of how models work internally.
Discovery
Can your business be found?
Credibility
Is your expertise understood?
Consideration
What place in the answer?
Outcome
Does it bring enquiries?
ObservedInferred against a referenceIndirect signal, no causality* Observed only when the interface displays its sources, or when analytics identify the referrer.
01 / THE PRINCIPLE
Observe before interpreting.
And say what cannot be observed.
Every dimension carries an observability level. A business owner must be able to tell what we saw, what we infer from it and what remains a signal.
Compared against a reference.
Accuracy is measured against the company’s fact sheet; credibility, against the evidence and sources available; enquiries, against CRM records and declared origin.
A signal, never proof.
The reuse of wording or a criterion specific to the company may point to a contribution. It establishes no causality.
The nine dimensions are read independently, and KAIS claims no access to the internal reasoning of models.
02 / THE NINE DIMENSIONS
Four readings.
Nine dimensions.
For each dimension: the public definition, what KAIS actually collects, the indicator reported and the limit of what it allows us to claim.
READING 01 / 04
Discovery
Can your business be found?
Elements drawn from search and from visible sources. Retrieval that is not visible is not deemed observable.
Observed
Outside the answer, through technical checksDiscovered
The brand, its pages or its entities can be found or appear in an observed search context.
- What KAIS collects
- Indexing of priority pages in Google and Bing, access for answer-engine bots, their visits in server logs, and whether the company appears on a search for its name.
- Indicator
- Share of priority pages indexed; bots allowed or blocked.
- Limit
- Being findable says nothing about whether a page is used in an answer.
Observed, conditional
Only if the interface displays its sourcesSourced
A page from the company is displayed as a source for the answer, whether or not the company is named in the text.
- What KAIS collects
- The sources the interface displays for each question, and whether a page from the company is among them. We also record whether the answer shows any web search at all.
- Indicator
- Source rate, calculated only on answers whose displayed sources were fully captured.
- Limit
- A displayed source was not necessarily used. Without displayed sources, sourcing is declared not observable, never zero.
READING 02 / 04
Credibility
Is your expertise understood?
An assessment inferred from the descriptions and evidence available, with no access to the internal reasoning of models.
Inferred
Against the company’s fact sheetUnderstood
The accuracy of how the brand and its services are described is assessed from observable answers.
- What KAIS collects
- Claims about the company (name, address, services, specialisms, prices, experts), compared with a dated fact sheet validated by the company.
- Indicator
- Accuracy: share of answers that describe the company without a critical error.
- Limit
- Without a validated reference, accuracy is not quantified. A vague answer is not an inaccurate one.
Inferred
Against the evidence and sources availableTrusted
The presence of verifiable evidence and independent sources is examined without assuming the model’s internal criteria.
- What KAIS collects
- The evidence the company publishes and the third-party sources engines cite on its questions: specialist media, professional directories, comparisons, reviews.
- Indicator
- Third-party coverage: how many of the most-cited sources on the panel mention the company.
- Limit
- We measure what is available, not how much weight the model gives it.
READING 03 / 04
Consideration
What place do you hold in the answer?
Mentions and recommendations are observed. Influence on the answer is assessed through an indirect indicator.
Observed
In the answer textMentioned
The company is named in the text of the observed answer, and its rank among the named players is recorded.
- What KAIS collects
- Mentions of the company in the answer text and its rank among the named players. When no web search is shown, the mention may rest on what the model already knows.
- Indicator
- Presence and position: this is the dimension the KVS score summarises.
- Limit
- Being named without being recommended is not a commercial success.
Indirect
A signal, never proofInfluence
The possible contribution of information associated with the brand is described as an indirect indicator, without a firm causal inference.
- What KAIS collects
- Reuse in answers of distinctive elements specific to the company: wording, a published figure, a selection criterion, with or without naming it.
- Indicator
- Uptake rate, on elements verified as specific to the company.
- Limit
- A signal consistent with a contribution, never proof of influence on a model.
Observed
Coded against a written gridRecommended
The brand is presented as a relevant option for the question asked, distinct from a mere mention.
- What KAIS collects
- How the answer presents the company (absent, mentioned, considered or recommended) and the outcome of comparison questions.
- Indicator
- Recommendation rate and preference in comparisons.
- Limit
- Recommended does not mean first: rank belongs to the Mentioned dimension.
READING 04 / 04
Outcome
Does it bring enquiries?
Visits and enquiries are reported by origin: observed, declared or unknown. They sit next to the visibility data and are never folded into a score.
Observed, conditional
Only when analytics identify the referrerReferred
Visits reaching the company’s site from AI answer surfaces are recorded when the referrer can be identified.
- What KAIS collects
- Sessions whose referrer is an AI assistant, or whose link carries an identifiable AI source, in the analytics tools the company authorises.
- Indicator
- AI-referred sessions and their landing pages, reported as a minimum.
- Limit
- Many visits prompted by an AI answer arrive as direct or search traffic. Growth may also reflect the growth of the platforms themselves: an effect is attributed only against a comparison group.
Inferred
Against CRM records and declared originContacted
Qualified enquiries are linked to AI answers through an observed or declared origin.
- What KAIS collects
- Qualified enquiries, each with its origin: observed (identifiable link or referrer), declared (the buyer says an AI assistant pointed them to you) or unknown.
- Indicator
- Qualified enquiries by origin, never added up across the three categories.
- Limit
- The dimension with the least evidence behind it. One buyer may see an AI answer, open a map and then call: the journey is kept, without counting three acquisitions.
03 / CROSS-READINGS
Being named, being the source,
being chosen.
Three distinctions that change the diagnosis: being named is not being the source, being named from memory is not being found on the web, and being mentioned is not being recommended.
Mentioned and Sourced do not say the same thing.
Named in the answer
Absent from the text
Company page among the sources
Full case
The site serves as evidence and the company is visible. Protect it: accuracy, freshness, a link to a useful page.
Evidence without credit
The content is used, the name does not come through. Reused passages should name the company.
Company page absent from the sources
Borrowed visibility
The company is carried by third parties: directories, media, platforms. What they say becomes its description.
Absence
Read the Discovered and Trusted dimensions first.
Named from memory, or found on the web.
When an answer names a company without showing any search, the mention may rest on what the model already knows. A company can be described accurately when its name is typed in, yet absent when buyers ask a generic question. That gap between recognition and discovery is often the first finding for smaller brands.
From mention to recommendation.
The Recommended dimension is coded on four levels, following a written grid. When in doubt between two levels, the lower one is kept.
0
Absent
The company appears neither in the text nor in the displayed sources.
1
Mentioned
Named in a list, without qualification or justification.
2
Considered
Presented with positive arguments, without the answer steering the choice.
3
Recommended
Explicitly advised for the question asked, or preferred in a comparison.
04 / FROM FRAMEWORK TO ACTION
A finding, a cause,
a verifiable action.
In the KAIS method, each measured gap is attached to the dimension that best explains it, then to a first action and to the indicator that should move. If the indicator does not move, the hypothesis is revised rather than the fix repeated.
The KVS score summarises the presence and position measured: it mainly reads the Mentioned dimension. The other dimensions form the Chain profile, reported next to the score and never added up into a single mark. Outcome figures are reported separately, with their origin.
This profile explains why a visible company is not always recommended.
05 / REFERENCE
Citing the framework.
And knowing what it is not.
How to cite
KAIS (KaiZen AI Strategy). KAIS AI Visibility Chain™: a public framework for reading visibility in AI answers, version 2.0, September 2026. https://kaizen-ai-strategy.com/en/ai-visibility-chain/. Public definitions licensed under CC BY 4.0.
What the Chain is not
- Not a score. The KVS score remains the headline indicator; the Chain explains what it does not say.
- Not a sequence. No dimension is a mandatory step towards the next.
- Not a description of models. It describes answers, displayed sources and verifiable states.
- Not a promise. It guarantees neither inclusion, nor citation, nor recommendation, nor enquiries.
What changed in version 2.0
- Retrieved becomes Sourced. The dimension records a page displayed as a source; the new name says exactly that.
- Cited becomes Mentioned. Being named in the text and being displayed as a source are now two clearly separate dimensions.
- A fourth reading, Outcome. Referred visits and enquiries by origin, reported next to visibility, never inside the score.
- Prior knowledge made explicit. A mention without any visible search is flagged, to separate recognition from discovery.
Frequently asked questions
Is the Chain a score?
No. The KVS score summarises the presence and position measured on an agreed question panel. The Chain is a reading framework: it shows the nine dimensions side by side, each with its denominator and observability level, without adding them up into a single mark.
Do the dimensions have to be worked through in order?
No. The dimensions are read independently. A company can be named without its page being displayed as a source, or be displayed as a source without being named. The diagnosis looks for the weakest dimension relative to the commercial stakes.
Why are some dimensions not quantified?
Because they are not always observable. When an interface does not display its sources, sourcing is declared not observable. Without a validated fact sheet, accuracy is not quantified. Influence remains a signal, never proof. Referred visits are reported as a minimum.
Does the Chain measure business results?
Partly. The Outcome reading records AI-referred visits and enquiries by origin: observed, declared or unknown. These are reported next to the visibility data, never folded into the score, and an effect is attributed to our work only with a comparison group.
Can the definitions change?
The public definitions are fixed for a given version. Any change creates a new, dated version of this framework; the version in force is shown at the top of the page.