AI & LLM Discovery
LLM Visibility
See which brands assistants recommend for your buyers’ questions, how they describe your business, and which sources they cite. Get a baseline you can inspect and a prioritized list of gaps to investigate.
How We Measure LLM Visibility
- 1Define the panel
Agree buyer questions, competitors, engines, collection surfaces, markets, and repetitions in the written scope.
- 2Capture the evidence
Save exact prompts, timestamps, answers, sources, and valid or excluded response status.
- 3Classify the outcomes
Count mentions, positive recommendations, owned-domain citations, and comparison-set share of voice separately.
- 4Review and retest
Prioritize factual, technical, and content gaps. Record deployed changes and compare later runs with panel and model changes disclosed.
Use our AI share-of-voice measurement guide to define the baseline, and our measurement definitions to distinguish mentions, recommendations, and source links.
Search Visibility vs. LLM Visibility
Search performance and assistant answers are related channels with different observation methods.
| Aspect | Search Visibility | LLM Visibility |
|---|---|---|
| Channel | Search results (Google, Bing) | LLM outputs (ChatGPT, Claude, Gemini) |
| Format | Links + snippets | Citations, summaries, recommendations |
| Optimization | SEO: rankings, keywords, links | LLM visibility: entity, structure, cited-worthiness |
| Goal | Traffic from search | Presence and attribution in AI answers |
What the measurement means
Each report describes a specific panel of buyer questions, engines, markets, and collection dates.
| Measure | Numerator / denominator |
|---|---|
| Mention rate | Valid answers that name your brand / all valid answers |
| Recommendation rate | Valid answers that positively recommend your brand / all valid answers |
| Owned-site citation rate | Valid answers linking to your domain as a source / all valid answers |
| Share of voice | Your counted brand appearances / appearances of all brands in the declared comparison set |
Illustrative example: a brand named in 20 of 100 valid answers has a 20% mention rate. If only 10 answers positively recommend it, its recommendation rate is 10%. A source link is counted separately.
The written scope sets the prompt panel, repetitions, collection surface, competitor set, and handling of failed requests or refusals. Reports show valid and excluded counts. Changed panels, model versions, and collection conditions are marked before comparing periods.
A sampled panel describes those questions under those conditions. It does not measure every buyer, the whole market, or revenue attributable to AI. Observed source patterns support hypotheses; they do not reveal a model’s internal decision process.
Frequently Asked Questions
- It describes how a brand appears in a declared sample of assistant answers. Mentions, recommendations, descriptions, and source citations are recorded separately.
- The report identifies the prompt panel, engines and versions where available, collection surface, markets, dates, repeated runs, competitor set, and valid-response counts. Its conclusions apply to that sample.
- We inspect cited sources, descriptions, product facts, and technical gaps to identify plausible explanations. Those are evidence-based hypotheses unless a suitable experiment supports a causal conclusion.
- No. Assistant outputs and retrieval systems change. We agree on the measurement method, implement reviewed changes, and retest under comparable conditions.
Ready to Get Cited by LLMs?
Discuss your buyer questions and competitors on a free call, then agree the scope of a paid baseline if it fits.