Skip to main content

​
Direct answer

Use Analyze > Sentiment Analysis to filter results by topic and inspect the stored answers classified as negative. Confirm that each answer concerns your brand and the intended topic before treating it as a pattern.

​
Data required

Use a stable prompt set, primary topics, optional tags, provider and model, period, stored answers, sentiment classifications, SELF/DIRECT mentions, and citation URLs. Keep topic assignment separate from reusable tags.

​
Workflow

1

Choose the topic cohort

Select the relevant topic, period, providers, and models. Confirm that the configured prompts actually test the concern.
2

Review the distribution

Compare sentiment categories and answer counts within the selected cohort. Retain the denominator and filters in any report.
3

Read the underlying answers

Inspect representative negative and non-negative answers. Separate factual criticism, trade-offs, unsupported claims, and irrelevant classifications.
4

Compare related slices

Check whether the pattern persists across prompts, providers, models, and later runs without changing several filters at once.

​
Interpretation

Sentiment is a classification of observed answer language, not a customer-satisfaction or brand-health score. A higher negative share in one topic can reflect prompt wording, a small answer count, provider mix, or a real recurring narrative. The data supports investigation, not a universal threshold.

​
Possible next actions

  • If a factual negative claim recurs, verify it and test a correction workflow with authoritative evidence.
  • If criticism reflects a genuine trade-off, test clearer positioning about the intended audience and limitations.
  • If the result is driven by one prompt, revise the monitoring design only after confirming that the prompt is unrepresentative.

​
Limitations

Automated sentiment can misclassify nuance, comparisons, or quoted text. Results cover the selected stored answers only. The REST answers endpoint exposes its own fields; do not assume product sentiment aggregations are reproduced without verification.

Answer Details

Report on the underlying answer records.

Sentiment Analysis

Review the product sentiment workflow.

Answers API

Use the documented REST answers contract.

Sentiment after an event

Compare stable cohorts across periods.

Monitoring prompts

Audit whether prompts cover the intended topic.

Correct misinformation

Address verified factual errors.

What AI says

Read stored answers in context.