Transparency

How Prospekteer uses AI

Built for an industry where trust matters. Here’s exactly what our AI does, what data it sees, and where it stops.

What’s AI vs. what’s deterministic logic

Most of Prospekteer is rules-based logic that behaves the same way every time. Generative AI is reserved for the writing tasks where it actually helps.

Deterministic logic (no AI)
Coverage gap detection rules
Risk score calculation
Fit score / win probability
Renewal date estimation
Pipeline funnel math and analytics
Vertical-specific signal rules
Generative AI
Cold opener writing
Conversation brief drafting
Outreach sequence copy generation
Natural language prospect search interpretation
Risk Copilot summaries

When you see a Sparkles icon in the app, that’s AI. Everything else is rules-based logic you can trust to behave the same way every time.

Which AI models we use

Short, factual. We use the right model for each job and disclose it here.

Claude Haiku by Anthropic

Used for cold openers and sequence copy on Free and Starter tiers. Fast, lower-cost, well-suited for short generative tasks.

Claude Sonnet by Anthropic

Used for Risk Copilot summaries and natural language search on Pro and Agency+ tiers. More capable for longer, context-heavy outputs.

Perplexity Sonar by Perplexity

Used on Pro and Agency+ tiers for real-time web research (recent news, public business activity). Cached for 21 days to control cost and reduce API load.

What data our AI sees

We send AI the minimum needed to do the job. Anything sensitive stays inside Prospekteer.

What we send to AI
Public business name and address
Industry classification (NAICS code)
Public risk signals (weather, regulatory data, business activity from public sources)
Aggregated category scores (not raw underlying data)
The user’s prompt or search query
What we never send to AI
Your AMS credentials, API keys, or any login data
Your client list, policy details, or premium amounts
Personally identifiable contact info beyond what’s already publicly available
Internal notes you write in Prospekteer
Other agencies’ prospect lists or activity

How AI outputs are reviewed

AI-generated cold openers and briefs are drafts, not final messages. The user reviews and edits before sending. We label every AI output with a Sparkles icon.

Confidence scoring is exposed at the field level. If the AI’s underlying data is uncertain, the field shows a low-confidence indicator so the user knows to verify.

User feedback (thumbs up/down, edited cold openers) is logged to improve future generations. Feedback is anonymized — we don’t tie individual feedback to individual prospects in training data.

What AI does not do

Just as important as what AI does. These are hard limits.

  • We do not auto-send any outreach without explicit user action
  • We do not make underwriting decisions or recommend coverage levels
  • We do not provide legal, regulatory, or compliance advice
  • We do not predict the future — risk scores reflect current observed signals, not guarantees
  • We do not share your activity, prospect list, or outcomes with other Prospekteer users

Questions?

Have a specific question about how our AI handles your data? We’d rather answer it than have you guess.

Contact us

Last updated: May 3, 2026