Check the answers. Keep the evidence.

Why isn't one check enough?

The same question can produce different answers. Our repeated-run archive records changes in which businesses appear and which holds first position. One check tells you what happened once; repeated observations show how consistent that result was within the measured panel.

Try this illustrative example: one buying-intent question with three possible answers. It shows why one result can be misleading.

Illustrative answers, not client measurements. Switch between runs to see how a single result can hide changes in the recommended businesses.

What does Nameworthy measure — the API or the app?

Every observation records its collection method. ChatGPT and Claude are measured in consumer-app sessions. Automated search surfaces retain their actual app or API label. API and app results can differ in model version, search behavior, and citations, so we report those limits and never use an API result as proof of what a consumer app displayed.

Why measure from a clean, logged-out vantage?

We aim to limit account-history effects and record the conditions of each observation: surface, session state, location settings and collection time. A logged-out or memory-off session does not prove that every source of personalization is absent. If an engine requires sign-in or a collection condition changes, that belongs in the report alongside the result.

How do you score an answer?

Scoring is pre-registered — defined before the baseline runs, and never reinterpreted mid-engagement. Every run is graded against four fixed events, for your business and for each competitor:

• Mention. Your business is named anywhere in the answer text.
• Recommendation. You appear in a recommended list — and we record your ordinal position.
• Citation. Your URL is linked as a source the engine drew from.
• Sentiment. How the answer treats you: positive, neutral, or negative.

Why report every engine separately?

Engine answer sets differ. Our reproducible archive analysis finds about 12% mean Jaccard overlap across eligible pairs, ranging from 6.7% to 21.4%. That set statistic is not a probability for one business. We report each surface separately so you can see where a result changed.

Why these six engines, and not ten?

Some tools advertise coverage of ten or more engines. We measure six on purpose: the surfaces where local, commercial-intent questions actually get asked and answered — ChatGPT, Gemini, Perplexity, Claude, Google AI Overviews, and Google AI Mode — each measured with its actual collection method disclosed and repeated high-intent runs. A wider engine list measured shallowly, through APIs that don't match what customers see, is a logo wall, not a measurement. When another surface earns real commercial-intent share, we add it the same way we measure the rest: declared surface, repeated runs, per-engine report. Until then we'd rather be right about six than decorative about twelve.

What's the statistics behind it?

We report successful observations and their denominators for each engine, with failed or unavailable runs shown separately. Two or three repeats can reveal inconsistency but do not produce a precise market-wide probability. The question panel stays frozen between comparisons, and we disclose changes in model, surface or collection conditions. A before-and-after difference is descriptive unless a stronger study design supports a causal claim.

Common questions