Frequently asked questions

No — and it never pretends to be. Every result is a model-based estimate on a synthetic population and carries a SYNTHETIC SIMULATION watermark on screen and in print. Samplio is for fast, directional reads before you spend real budget; it complements fieldwork, it does not replace it.

We do not publish an accuracy figure yet, on purpose. A number appears only once it is measured against a sourced corpus of real-world cases (the tooling for that backtest exists and ships with the product). Publishing an invented benchmark would break the product's core rule: no fabricated statistics, anywhere.

Country packs for all 27 EU members plus Türkiye and Georgia, built from live Eurostat data (NUTS-2 population shares, EU-SILC income quintiles where published), five macro regions (Western/Northern/Southern/Central/Eastern Europe) and an EU-27 average. Country packs simulate in the audience's own language where a prompt pack exists (8 languages today, English otherwise). Macro regions and the EU average combine EU member packs only.

A pilot package is a one-time purchase (no subscription) that we deliver hands-on: your messages, variants or a conditional survey run against the audience scopes you pick, with a printable, watermarked report per test. Studio covers a single message round, Team a variant-comparison round, Agency a multi-market round — write to us and we scope it to your case before you pay.

No. Election and political-campaign use is prohibited by the terms of service and enforced mechanically: an eight-language keyword screen plus an LLM classifier reject political content at the engine boundary, failing closed on any doubt. This is not configurable.

No real person's data is processed to build audiences. The synthetic population is derived from aggregate public statistics (Eurostat/EU-SILC); your account data (name, email, company) is used only to operate the service. Every LLM call is audit-logged (model, prompt hash, duration) without storing prompt text; no customer identifier ever enters a prompt.

Honesty stamps. The value-orientation axis is not yet calibrated against real survey data, so the engine mechanically marks every output that depends on it. Custom surveys are always marked uncalibrated because free-form questions have no backtest anchor. Treat those splits as directional — the report says so because the engine cannot be configured to hide it.

Because it would be fake precision. Sampling error describes random samples of real people; synthetic agents have no sampling error, so showing one would be misleading. What Samplio can honestly measure is re-run stability: run the same test with consecutive seeds and report how much the distribution moves.