Prompt quality classification
Every active prompt is automatically classified based on the last 15 LLM responses it has collected. The classification helps you spot prompts that aren't generating useful, commercially relevant answers — before they distort your metrics.
The classifier runs daily in the background.
The four labels
Healthy
The prompt consistently returns commercial answers with links — i.e. AI responses that name brands and cite sources. These are the prompts that drive meaningful VS / SOV / TMR signals.
Action: keep as-is.
Watch
The prompt is borderline. Some recent runs returned commercial answers, others returned info-only or no-links responses. Quality may be drifting.
Action: monitor over the next few cycles. If it slips to Low, consider rewording.
Low
The majority of recent runs returned informational answers without commercial intent or citations (e.g. generic encyclopedia-style explanations). The prompt is unlikely to surface brands at all.
Action: edit the wording to introduce commercial intent (e.g. add "best", "near me", "recommended"), or soft-delete and replace.
Excluded
The prompt repeatedly returns answers with no links and no entity mentions, or matches the informational-domain exclusion list. Excluded prompts are kept visible for transparency but are dropped from core KPI calculations (VS / SOV / TMR averages) so they don't distort your dashboard.
Action: soft-delete and replace, or rework the prompt to elicit commercial intent.
Where to see classifications
- Stages / Journey Analytics — each prompt is tagged with its current label.
- Prompt list in site settings — filter by Healthy / Watch / Low / Excluded.
How the classifier works
Each of the last 15 responses for a prompt is bucketed into one of:
- commercial — names brands and includes citations/links.
- info_only — informational answer, no commercial intent.
- no_links — no citations and no recognised entity mentions.
The mix across the 15-response window determines the final label. 🟡 [verify: exact thresholds — these are admin-tunable in the Prompt Quality config panel and may differ from the defaults].
Why this matters
Without quality classification, low-signal prompts drag your averages down and make competitor comparison noisier. By isolating Excluded prompts from KPI math, your dashboard reflects performance on the prompts that actually matter for commercial visibility.