Scans and prompts

A scan is a single execution of one query (prompt) across the four LLM platforms. It produces four responses, and that bundle is the smallest unit of data we store.

Lifecycle of a scan

pending → running → completed (or failed)

A scan typically completes in 20–40 seconds. The slowest LLM (usually Perplexity, which does live web search) determines total scan time.

What's inside a scan

When you open a scan detail page you see:

• The prompt text that went out
• Four platform cards, one per LLM
• For each card:
  • The verbatim response
  • Mention pills for your brand and each competitor, color-coded by sentiment when extracted
  • Citations (URLs the LLM referenced) with the publisher
  • Cost in USD (BYOK only)
  • Latency in milliseconds
  • Generation ID (provider-side request reference for support escalation)

Single scans vs bulk scans

Single scan — you click "Run scan" on one specific query. You'd do this when iterating on a new prompt or chasing a specific question.

Bulk scan — you click "Bulk scan" on a topic and all its queries fire in one HTTP roundtrip. You'd do this for a snapshot of your whole topic, or to seed a new workspace.

Bulk scans run sequentially per query (so per-query free-tier enforcement still works), but each query's four platform calls run in parallel. Total time for a 25-query bulk is typically 8–15 minutes.

What prompts get sent

The text you type into the query is sent verbatim — no rewrites, no system prompts that bias the answer, no "hidden context" about your brand. The LLM sees only the customer-style question you wrote.

The exception: a generic helpful-assistant system prompt is prepended ("You are a helpful assistant that provides accurate, well-researched information about brands and companies"). This is a neutral instruction — the LLM doesn't know which brand belongs to you and which are competitors.

We're testing what a real LLM would say to a real customer. Anything else would game the measurement.

Quotas and rate limits

• Free tier — 5 scans per month, max 2 per hour, shared-key gated (cheap models, regex-only mention extraction, 600 max output tokens)
• Pro tier — 500 scans per month on the shared key OR unlimited on BYOK
• Team tier — unlimited scans on BYOK; not applicable on shared key

A scan counts as one against the cap. A bulk run of 25 queries counts as 25 scans.

See Free tier limits for the full breakdown.

Failure handling

If one platform call fails (timeout, 5xx, content policy block), the scan still completes — you just get three responses instead of four, and the missing platform shows a friendly "couldn't reach this platform" panel.

If all four platforms fail, the scan marks failed and the dispatcher logs it. The free-tier counter doesn't increment on full failure (you don't get charged for our infrastructure being broken).

See also

• Bulk scans — how the bulk path works in detail
• Scheduled scans — daily / weekly / monthly automation
• Mentions & sentiment — what we do with the responses after they land