How we use AI

MarketImpact builds and teaches responsible AI practice, and we hold ourselves to the standard we sell. AI tools are used throughout our work: research synthesis, data collection across languages, drafting, data analysis, and software development. Judgement is not delegated. Every published piece and client deliverable is reviewed, verified, and edited by a named person with direct domain expertise, and the arguments, conclusions, and recommendations are theirs.

Our internal rules are written down and enforced. Work is risk-tiered before AI touches it. We use only enterprise AI tools with contractual no-training guarantees. Highly sensitive material is processed locally on company hardware, and beneficiary-level data never enters any AI tool, ours or anyone else's. Higher-risk outputs carry a documented sign-off recording who verified what against which sources. These rules live in our AI Use Policy (MI-POL-001, v1.0, March 2026), available to clients and partners on request.

Each published analysis carries a specific disclosure stating what AI did for that piece and what the author did. We cite sources. We name our limitations. When we get something wrong, we correct it publicly and quickly.

This discipline is also what we teach. Our AidGPT training exists because we believe the sector's choice is not whether to use these tools but whether to use them with verification, transparency, and human accountability. This page is that belief applied to ourselves.