How we measure AI visibility — and what our scores do and do not mean

1. Principle: AI Visibility is a point-in-time assessment, not a fixed ranking

AI systems do not generate one permanent, universal answer. Outcomes can change because of query wording, context, location, freshness, available sources, model updates and the AI or search system being used. Our analysis is therefore a structured point-in-time assessment of the visibility, clarity and recommendation readiness of an organisation, brand, website and — where relevant — its products.

2. The four assessment pillars

3. How the AI Visibility Score is produced

The headline score combines the four assessment pillars using a fixed server-side calculation method. The model performing the analysis does not freely choose the final headline score. This reduces unnecessary variation when measurement data is translated into the final number.

We do not publish the exact internal weighting or anti-gaming rules. This reduces the risk of websites being optimised only to manipulate the score or deliberately game the measurement system. The assessment areas, meaning and limitations are public on this page.

All scores are proprietary standardised AVR indicators from 0–100. They are not official scores from OpenAI, Google, Microsoft, Anthropic, Perplexity or any other external provider, and they are not a universal industry standard.

4. Product scope and measurement depth

The analytical principles remain consistent while research depth increases by report tier. The scope below reflects the public product specification on the publication date; the current product pages remain authoritative if a package is later updated.

5. Free Check versus paid analysis

The Free AI Visibility Check is a low-friction first indication. It checks a fixed set of technical and content signals and displays improvement points directly, but it is not a substitute for a paid analysis.

Paid reports add direct scenario measurements, deeper interpretation, package-dependent competitor analysis and prioritised actions. The customer therefore receives not only a score, but also the context behind that score.

6. AI Shopping: what we do and do not assess

For ecommerce and product businesses we examine whether public product information gives AI systems sufficient signals to recognise, compare and match products to buying questions. Examples include product names, attributes, variants, price, availability, retailers, structured data, product relationships and other publicly observable commerce signals.

7. From observation to recommendation

We deliberately distinguish three levels:

• Observation: a technical, content or scenario signal detected during the analysis.
• Finding: an interpretation of multiple relevant signals in context.
• Recommendation: a concrete improvement action that logically follows from a finding.

A recommendation is not presented as a guarantee of a ranking, AI mention, revenue increase or other commercial outcome. External platforms determine their own inclusion and selection processes.

8. Reproducibility and variation

We aim to apply a consistent and repeatable analytical method. Generative AI answers themselves are not perfectly deterministic. The same or similar query may produce a different answer at another time or in another system.

We therefore use multiple scenarios and fixed scoring logic. Repeated measurement is most useful for assessing direction, patterns and change, rather than implying that every individual AI answer is permanent.

9. Sources, verification and uncertainty

An analysis uses information and technical signals that are publicly available and technically assessable at the time of execution. We cannot prove what a closed AI model internally “thinks”, which source it will select in the future or why a non-transparent algorithm makes every individual choice.

When available signals do not sufficiently support a conclusion, that conclusion should not be presented as certainty. Where relevant, a report can therefore identify a signal that still requires verification, missing evidence or a limitation of the measurement.

10. Explicit limitations

• AI Visibility is a point-in-time assessment, not a permanent ranking.
• Outcomes can differ by system, model, location, time, source set and query wording.
• Scores do not predict revenue, conversion, market share or future AI mentions.
• We do not control inclusion, ranking or recommendations by external AI and search platforms.
• Non-public information cannot automatically be included as independent evidence.
• Competitor comparisons apply to the examined context and scenarios; they are not a complete assessment of a competitor.
• A technically strong website is not automatically highly visible in actual AI answers.

11. Independence

AI Visibility Report Group is an independent analytical service. Our reports, scores and methodology are not issued, certified or endorsed by OpenAI, Google, Microsoft, Anthropic, Perplexity or other referenced AI or search platforms unless explicitly stated otherwise.

12. Methodology versioning

Current public version: 1.0 — 15 August 2026.

AI search and AI Shopping evolve quickly. When we materially change the public assessment areas or interpretation of the methodology, this page will be updated. Internal implementation details, source code, prompts, security rules and anti-gaming logic remain non-public.

For current pricing and package contents, see Reports. Any organisation can start with the Free AI Visibility Check.