AI transparency

What Scout is

Scout is the AI assistant built into SpotYourStartup. It answers questions about ideas, runs the fit quiz conversationally, and on paid tiers performs deep analyses of the library against your constraints. Scout is disclosed as an AI system at first contact on every surface where it appears (EU AI Act, Art. 50).

Models we use

Conversations are sent to Anthropic for processing under a data-processing agreement and are not used to train models. Chat conversations are not stored server-side by us beyond the request; deep-analysis briefs with their results and personal-angle reports are saved to your account until you delete them. Usage metering (token counts, cost) is retained for quota enforcement under the retention periods in the privacy policy, which also covers the data-protection side of all AI processing.

What is AI - and what is not

• Every published dossier passes mandatory human review - publication is always a human decision, never automatic.
• Dossiers originate in one of two ways: hand-researched by our team, or drafted by our AI-assisted research pipeline (below). The export of every idea carries a machine-readable ai_provenance block naming its origin and any AI-derived fields.
• The Opportunity Score is computed by a fixed, documented formula from measured sub-metrics; the numbers are evidence-based, the formula involves no AI at scoring time.
• Your fit score is deterministic arithmetic over your profile and an idea's demand vector - no AI involved at match time.
• Scout's chat answers and deep-analysis reports ARE AI-generated and are always labeled as such.

How the research pipeline works

Our research pipeline uses AI models (the same Anthropic models listed above) together with live web search and structured public sources to research startup-idea candidates in depth. Every claim in a resulting dossier must cite an entry in an evidence ledger of real, retrievable sources; candidates then face a mechanical quality gate (evidence density, source diversity, recency, citation integrity) plus an adversarial review pass. Most candidates are rejected - producing nothing is a normal outcome. What survives becomes a draft that a human reviews, edits and publishes, or discards. Spending on the pipeline is capped by hard monthly, per-run and per-candidate budgets.

Art. 50 measures in the product

• Disclosure line ("Scout is an AI assistant.") on every Scout surface, shown before first interaction.
• An AI badge on AI-generated messages and reports.
• Machine-readable provenance: exports carry an ai_provenance block naming content origin and any AI-derived fields.
• Grounding: Scout answers from the library through a fixed set of read-only tools; it cannot take actions on your account.
• Hard cost and usage caps per plan, so AI use is bounded and predictable.

Limits you should know

AI answers can be wrong, incomplete or outdated despite grounding. Scout is designed to point at evidence, not to replace your judgement - and where an idea does not fit you, it says so. How scores are built is public in the docs (How we score, Founder fit). Questions about an AI answer, or one that looks wrong or harmful: hello@spotyourstartup.com.