Mention
The brand or product name appears in the answer.
AiVisible methodology
How we measure brand presence in AI-generated answers.
We publish definitions, sample-design rules and limitations so results can be interpreted and cited responsibly.
The method is versioned. Every published change has a date and description.
Method flow
From buying situation to finding
1Buyer research
2Question set
3Model answers
4Coding
5Aggregation
6Interpretation and limitations
A result always relates to a defined sample, models, market and date.
Definitions
A metric only makes sense with its denominator and sample.
Recommendation
The answer presents the brand as an appropriate option for the question.
Citation
The answer names a source connected to the information.
Recommendation rate
The share of answers meeting the recommendation definition within a sample.
Question set
Questions come from buying situations, not a random keyword list.
option discovery
shortlisting
comparison
use case
budget
trust
Limitations
Every result is published with its limitations.
answers are variable
models and sources change
conversation context may affect the result
presence does not prove sales impact
comparison requires a stable sample core
Citing results
When citing a result, include the category, sample, models and date.
Important:We do not treat one answer as evidence of a brand's position across an entire category.
Sample design
The question set represents decision stages, not keyword volume.
Intent map
Coverage across the buying decision
Active system structure
CohortQuestionCoded criteria
1Option discovery
Which solutions are worth considering?
category · alternatives
2Shortlist building
Which provider or product fits this use case?
segment · specialisation
3Comparison
Option A or B against these criteria?
differences · limitations
4Fit
What works at this scale and budget?
use case · budget
5Trust
Which option has the right experience?
evidence · sources
Discovery
Questions about available options, categories and ways to solve a problem.
Shortlist
Questions narrowing selection by segment, industry, use case and risk.
Comparison
Questions about differences, alternatives, limits, budget and fit.
Trust
Questions about experience, credibility, sources and selection conditions.
Coding
We separate mentions, recommendations, citations and reasons for selection.
whether the brand appears
whether it is presented as an appropriate option
which need it is matched to
which arguments support the selection
whether sources are cited
how results differ by cohort and model
Comparing waves
Change only matters when comparison conditions are preserved.
The control core remains stable and exploratory questions are marked separately. Changes to models, scope or coding must be documented before interpreting differences.
Wave comparison
Baseline, publication and follow-up measurement
Active system structure
T0
Baseline
- control question core
- models and date
- full answers
Log
Implementation log
- asset and owner
- publication date
- scope of change
T1
Follow-up wave
- the same core
- the same definitions
- description of the difference
Exploratory questions are labelled separately
Model changes are recorded as limitations
No difference is also a result
stable sample core
question-set version
model list and dates
consistent metric definitions
new questions documented separately
full answers available for audit
Next step
See how the method works in your category.
We first determine whether a useful, comparable sample can be built.
FAQ
Questions before you decide
Is the number of prompts the most important factor?
No. Coverage of buying situations, a clear denominator, consistent definitions and repeatability matter more.
Why can results change?
Models, sources and answer generation evolve. The question, date, context and sample scope also affect results.
Does recommendation rate represent sales share?
No. It measures observed recommendations in a defined sample. Connecting it to business outcomes requires client data and separate analysis.
Can AiVisible results be cited?
Yes, when the category, sample, models, date and relevant limitations accompany the result.
Related resources