Relocation Lab — Europe

Which immigration lawyers, tax advisors and company-formation agents in Portugal, Spain and Cyprus do AI assistants name when someone relocating asks? Measured weekly on a frozen panel of 40 questions, four engines, five locations, seven runs each.

What is asked

A frozen panel of 40 English questions in three published classes — choosing a provider (16 questions, e.g. “best immigration lawyer for a Portugal D7 visa”), a relocation problem (12, e.g. “how do I get residency in Portugal as a remote worker”) and comparing options (8, e.g. “Portugal vs Spain vs Cyprus for a founder relocating from the US”). Provider questions are asked from five locations — Lisbon, Madrid, Limassol, London and New York; the other classes from Lisbon. The full panel text and its SHA-256 are in the open repository; the panel was frozen on 2026-09-28, re-frozen on 2026-09-29 when the start date was moved forward before any data (prompts unchanged), and is not edited.

What is counted

For each firm: the share of answers that name it, per question class, engine and location, with a 95 % Wilson interval; the domains the engines cite; the week-to-week stability of the set of named firms. Firms enter the list when an engine names them; every firm in the list gets a row every week, including zeros. Position inside an answer is not a metric. By Machines itself is reported in a separate row, never inside the table.

What is not claimed

Nothing here says which firm is better. The tables say which firms the engines name, how often, and where the engines look. Four API calls are not the consumer apps: the method explains the difference and the weekly manual check that keeps it honest.

Latest table

First reading of the Relocation Lab: Monday 2026-10-19. The panel has been running on the lab server since 2026-09-29; the first weeks of data are published together with the first reading.

Weeks

The first week of data appears here after the first reading (2026-10-19).