How it works

How Sivlo measures AI visibility

We turn scattered data into clear results, so you can see where your company appears and what to do next.

PProjektioOverviewAI searchesCompetitors

Fictional example

Mention rate

34%

136 / 400 answers

Mentions by service

ChatGPT52
Gemini38
Claude27
Perplexity19

Reading the results · all figures and companies below are fictional

01

We map your visibility

See where your company appears in the measured AI services and which topics it is mentioned for.

ChatGPT

52 mentions / 100 answers

Gemini

38 mentions / 100 answers

Claude

27 mentions / 100 answers

Perplexity

19 mentions / 100 answers

02

We explain what it means

We group mentions by themes, products and buyer needs to give you a clear overall picture.

Key topics

Share of mentions · one mention may cover several topics

Task management
52%
Team collaboration
28%
Ease of use
18%
Integrations
12%
Price
8%
Example mention
Projectio is a good choice for a small team that needs a clear task overview and an easy way to share work.

Fictional answer

03

Compare with competitors

See how your visibility compares with competitors and where the biggest differences are.

Mention rates compared

Same questions · 400 eligible answers

34%
P

Projektio

62%
T

Tehtävä

58%
T

Työpolku

40%
T

Tiimio

28%
A

Askel

04

Find opportunities

We identify question areas where competitors appear more often and suggest practical improvements.

Opportunities

Fictional prioritisation example

Buyer situation importance ↑
Team task trackingIntegrationsProjectio vs. Tehtävä
Competitor visibility →
Explore the measurement principles Metrics, sampling, uncertainty and evidence limits

01What do we measure?

Key metrics

Four core metrics describe visibility. We distinguish a company being mentioned, recommended, or cited through its website.

Mention rate

Answers mentioning the companyRelevant, successful answers to questions that name neither the company nor a competitor

How often your company appears in AI answers.

Recommen­dation presence

Answers recommending the companyEligible recommendation-oriented answers

How often your company is recommended for a buyer’s needs.

Share of recommen­dation

The company’s recommendation eventsRecommendation events for all identified brands

Your company’s share of recommendations in the comparison.

Citation presence

Answers citing the company’s own websiteSuccessful answers exposing citation or source data

How often your own website appears as a source.

One answer, different signals

Fictional example · not a measured AI answer

Which project management tool suits a small team?

Options include Projectio, Tehtävä and Työpolku.Mention

Projectio is a good choice if your team needs a clear task overview.Recommendation

Its website describes team views and task tracking. [1]Citation

Sources

1. projektio.example/features

Which answers count?

Recommendation presence includes successful, relevant, recommendation-oriented answers. Questions naming the company or a competitor, and brand-check questions, are excluded. A failed service call is a missing observation, not an absent company.

Citation presence includes only answers exposing source data. Each share retains its numerator and denominator; without eligible observations, the result is unavailable, not zero. Filters cover services, markets, languages and buyer situations actually measured.

We also examine third-party support and company understanding. Co-occurrence with independent sources is not a model’s internal trust score. Profile comparison is labelled a Sivlo model evaluation; dimensions lacking evidence remain unevaluated. We do not use a composite 0–100 visibility score.

02How do we measure?

A repeatable measurement process

A consistent methodology, documented settings and clear steps make measurements auditable. Results are compared only under comparable conditions.

  1. 1. Market model

    We identify products, buyers, markets and competitors.

  2. 2. Buyer situations

    We describe buyer needs, use cases and selection criteria.

  3. 3. Questions

    We create question variations and identical repeats for each situation.

  4. 4. Answers

    We collect answers from the selected services with consistent settings.

  5. 5. Results

    We calculate shares, preserve evidence and assess changes.

Why several services?

Services can produce different answers. Measuring several reveals differences and coverage limits.

ChatGPT

Mentions and recommendations

Sources, when available

Gemini

Mentions and recommendations

Sources, when available

Perplexity

Mentions and recommendations

Sources, when available

Claude

Mentions and recommendations

Sources, when available

Settings, sampling and service differences

The market model uses the company’s website and customer-confirmed information. We record whether a fact was supplied by the customer, inferred from the site or drawn from fixtures. Sampling prioritises strategic buyer situations and realistic question variations within the observation budget. Identical repeats help estimate service variation. Competitors named in the same answers remain in the analysis.

Official model APIs and consumer search surfaces are different measurement types. Gemini API is neither Google AI Overview nor AI Mode. A search surface appears in results only if it was actually measured. AI used internally for classification or recommendations is not a visibility observation.

Methodology version: sivlo-measurement-v2.22

03Sample size and uncertainty

A number needs a sample size

An individual change may reflect statistical variation. Eligible rates include sample size and uncertainty so you know what the result supports.

Uncertainty in an example metric

Fictional data · recommendation presence

Example recommendation rates and Wilson 95% intervalsCompany A: 35/124, 28%, 21%–37%. Company B: 52/124, 42%, 34%–51%0%20%40%60%28%Company A35 / 12442%Company B52 / 124

Wilson 95% confidence interval

For eligible rates, the interval describes statistical sample uncertainty. It does not cover every source of measurement error. k is the success count and n is the number of eligible answers.

Rate = k / n

Sample size

n = 124

Fictional example: 124 eligible answers for each company.

When is a change comparable?

Comparison requires the same market-model version, buyer situations, sampling strategy and service settings. If these change, runs are marked non-comparable. Monitoring cadence follows the subscribed plan.

With at least 3 identical repeats, the noise threshold comes from measured variation and is at least 3 percentage points. Without calibration, non-overlapping 95% intervals are used; this is a weaker basis and is labelled accordingly.

If either run has fewer than 5 eligible observations, we report a measurement gap. Other outcomes are meaningful growth, meaningful decline, movement within expected variation, insufficient evidence or a non-comparable configuration. Competitors follow the same rules.

04Evidence classes

Observation, association or hypothesis?

We distinguish measured observations from interpretation. Each diagnostic conclusion has one evidence class and a reason.

Observed

Measured directly in this run or verified unambiguously on a page.

Directly measured

Associated

A recurring relationship in the measured market. It does not establish causation.

Recurring market relationship

Hypothesis

A plausible explanation that does not yet meet the evidence requirements.

Plausible, not yet proven
What qualifies as an association?

An association requires at least 3 brands, 2 buyer situations, 2 measured services or surfaces, and 12 eligible observations. The difference must be at least 10 percentage points. An explanation below these thresholds remains a hypothesis, with missing evidence stated. Association does not establish causation.

05What we do not claim

Transparent and realistic

Interpretations and decisions need sound information. That includes explaining the limits of our methodology.

No guaranteed rankings

We do not guarantee a position or visibility in AI answers.

No claim of direct causation

A result after a change does not prove that the change caused it.

Results vary

Answers depend on questions, services, settings and measurement time.

Answers are a sample

Every figure comes from a sample of answers and is shown with its sample size and uncertainty.

Sivlo suggests improvements and prepares drafts. Your team can implement changes manually. Optional publishing supports selected WordPress sections when enabled. An owner or admin approves each exact change. Connecting or saving a draft does not publish it. We do not predict a visibility uplift. Ask about the methodology

The result

A clearer picture. Better decisions.

Know where your company appears, what the observations mean and which next steps matter most.

Current positionMeasured visibility
OpportunitiesEvidence-led next steps
Follow-upCheck the change