Discover
Which solutions are suggested for a problem the product solves?
Understanding AI visibility
How to interpret AI visibility
Define questions, models and markets before interpreting a visibility measure.
Your brand · answers and sources
OverviewAnswersSources
The brand is missing: what should you check?
Example question
Which headphones do you recommend for working in a noisy office?
What you read
Example: a customer asks which headphones to choose for work. Open the answer to see whether your brand appears, which competitors are named and which pages are cited. Use this to identify information to clarify on your website.
The next step
Prepare a proposal to update the product page, then repeat the check using the same questions.
Mentions, recommendations and citations describe distinct signals: they are not added together. The denominator includes only valid answers, with the model and collection method stated.
What you get
Choose the questions that matter.
The question sample should represent the intentions relevant to your team: learning about a category, comparing alternatives or considering a purchase. Language and market remain explicit. Repeating observations at controlled intervals reveals differences, provided model changes are also recorded. Twenty prompts describe only those twenty observations.
| Intent | Question to represent | Signal to examine |
|---|---|---|
| Learn about the category | Which features matter when choosing this product or service? | Topics and criteria described, even when no brand is named. |
| Compare alternatives | Which differences emerge between solutions for the same need? | Presence, descriptions and sources, keeping comparison distinct from recommendation. |
| Evaluate a choice | Which offering is proposed for a defined need and context? | An explicit suggestion and reasoning within the answer, without extending them to every conversation. |
What you get
AI mentions, citations and recommendations
Build a first collection that can be read question by question. For each answer, note your brand’s presence, any recommendation and disclosed sources. The aggregate should let you return to a specific case worth checking, rather than stop at a percentage.
- Brand mentioned
- Recommended choice
- Cited source
What you get
Exclude incomplete tests from the denominator
If a collection is interrupted, list the cases to repeat before interpreting absences. Record the time of the new attempt and retain one outcome according to the established criterion; choosing the most favourable answer would distort the sample.
- Incomplete collection
- Recorded retry
- Retained valid outcome
Designing the observation
Questions matter. Choose them well.
AI visibility monitoring starts with customer needs. A useful panel includes discovery, comparison and verification questions: it represents a decision journey, not just a list of searches that already contain the brand name.
Gather recurring questions from support, on-site search and sales conversations, removing personal and confidential information. Assign each an intent, market and language. Preserve the exact wording to repeat comparisons. A team-built panel is a designed sample: how often a question is run does not measure how many people actually ask it. Before starting the panel, check that the questions represent different needs rather than simple variations of the same sentence. Keep prompts that already name your brand separate: the expected presence of the name does not demonstrate spontaneous discovery.
Compare
Which differences emerge when the customer evaluates relevant alternatives?
Verify
What is said about compatibility, support, conditions and limitations?
Understanding the sample
One answer. More information.
An answer may name your brand without recommending it, recommend it without citing one of its pages or cite a source that discusses it critically. These observations have different meanings and should be interpreted separately.
Record the question, usable answer and collection conditions. Engine, model, method, language and observation time accompany the result. Direct API answers, web-search-enabled APIs and consumer experiences remain separate groups. Failure to complete a request does not demonstrate your brand’s absence from the answer.
| Observation | What it describes | What it does not prove |
|---|---|---|
| Mention | Your brand appears in the observed text. | A recommendation, positive judgement or universal exposure. |
| Recommendation | The answer proposes your brand for the expressed need. | A purchase, conversion or stable model preference. |
| Citation | The answer provides a reference to a source. | Exclusive use of that source or its causal influence on the answer. |
From analysis to work
Repeat the method. Check the change.
A useful comparison keeps controllable conditions stable and makes changes visible. Altering the questions alongside website content makes subsequent results harder to interpret.
Before the action, define which questions to repeat and which answers to examine. If the model or collection method changes, start a separate comparison. Connect actions to evidence: missing information, outdated documentation or imprecisely described differences. Subsequent sampling measures what can be observed; it does not guarantee future citations or causal attribution.
- Panel version
- Documented intervention
- Comparable new observation
Related paths
Choose your next step.
Check your brand’s presence, compare results or choose content improvements: continue with the feature you need.
Guides to reputation, trust and AI
Prepare an AI check, verify a citation and compare results: guides with examples to improve your brand’s content.
ReadFrom question to action
From the question to checking the outcome: context, sources, hypotheses and action plans in the TrustInsider analytical workflow.
ReadA transparent method, not a black box
TrustInsider’s criteria for interpreting reputation and trust: provenance, samples, reviews and the limits of conclusions.
ReadFind the right answer
Find guides, FAQs and methodology to clarify indicators, coverage and access, with dedicated routes for data requests.
Read