A sample, with its workings shown.

Retimos measures independent responses from configured API providers. This does not represent all ChatGPT or Perplexity consumer-app traffic, does not train models and does not guarantee recommendations.

Discovery and knowledge are separate

Discovery questions exclude your brand and domain. Each probe sees only its question and a neutral instruction, with no profile or previous answers. Knowledge questions name your product and ask about confirmed fact IDs. Only the later evaluator receives your confirmed profile.

Visibility and recommendations

Brand visibility = unambiguous text mentions / successfully evaluated discovery answers × 100. Recommendation rate uses explicit positive recommendations instead. Sources alone do not count as mentions. Ambiguous names remain visible as uncertain observations. API errors and evaluation failures are excluded and counted separately.

Knowledge Score, coverage and accuracy

For each expected fact in each successful knowledge answer we count correct (C), incorrect (I), missing (M) or uncertain (U). N = C + I + M + U. Knowledge Score = C/N, coverage = (C+I)/N, accuracy = C/(C+I), each × 100. Zero denominators display N/A. Technical evaluation failures are excluded, not reclassified as uncertainty.

Example: C=3, I=1, M=2, U=2 gives Score 37.5%, coverage 50%, accuracy 75%. Accuracy is relative to an owner-confirmed profile, not independent verification of the whole market.

Citations and competitors

Citation rate is the share of successful discovery answers citing a confirmed product domain. We retain provider citations separately from uncited search results. Competitors are evaluated on the same discovery sample; frequency is not market share. Rank is N/A unless explicitly ranked.

Changes over time

We compare the intersection of successful probes with identical question revisions, language, model/settings and methodology; knowledge also requires an identical fact version. Changes are percentage points. A single response is noisy, and a before/after difference does not prove your edits caused it.

Evidence and drafts

Classification quotes must occur in the saved answer and refer to valid snapshot facts. That check verifies provenance, not the truth of the classification. Always inspect the original answer. Drafts are suggestions you review and publish yourself. “Implemented” is an owner-reported status.