Skip to main content

​
Direct answer

Identify the highest observed sentiment and earliest recorded position within a matched provider dataset. Do not label a provider permanently best: provider output, model choice, prompt mix, and eligible sample size can change the result.

​
Data required

  • The same answered prompts across providers or models
  • Detected SELF mentions with sentiment values
  • Recorded SELF positions and their eligible counts
  • A fixed period, country, language, topic, tag, and funnel scope
  • Representative answer text from each provider

​
Workflow

  1. In Analyze > Sentiment, set a consistent analysis scope.
  2. Compare provider or model sentiment with the number of eligible SELF mentions.
  3. In Cockpit > Overview or Cockpit > Compare, record Avg Position with its eligible position count.
  4. Read answers in Monitor > Response Analysis to verify tone, claim accuracy, and whether a numeric rank is meaningful.
  5. Separate providers with incomplete or materially different answer coverage.
  6. Repeat the comparison on important topics rather than relying only on the aggregate.

​
Interpretation

Sentiment summarizes detected language around eligible SELF mentions; it is not a fact-check or a measure of user opinion. Avg Position is conditional on SELF mentions with a recorded rank and excludes brand absences. Always pair sentiment and position with Coverage or Mention Rate. A provider can show favorable sentiment among a small set of mentions while omitting the brand from many answered prompts.

​
Possible next actions

  • Test whether a negative pattern is concentrated in one topic, prompt, or outdated claim.
  • Review the sources cited in unfavorable answers before proposing a correction.
  • Compare models within the same provider when their results diverge.
  • Track a stable answer cohort after publishing verifiable corrective information.
These actions test explanations and do not guarantee a sentiment or position change.

​
Limitations

  • Sentiment classification can miss nuance, irony, or mixed evaluations.
  • Position is not meaningful for every narrative response.
  • Providers may return different answer volumes or formats.
  • Observed provider differences do not reveal training data or internal ranking logic.