How we handle data and where AI is used
Two statements in one place: how PMEye protects personal and research data, and the rules we apply when AI assists our analysis. Both describe the practice we already commit to on every assignment.
Privacy and data protection
We hold two very different kinds of data: the little that this website collects, and the research data entrusted to us on assignments. They are treated differently.
This website
The site is a set of static pages. It sets no advertising or tracking cookies and runs no third-party analytics or social pixels. Web fonts are requested from Google Fonts, which receives the request as part of serving them.
Our host keeps standard server logs (IP address, timestamp, page requested, browser) for security and troubleshooting. The enquiry form sends what you type to our office; the site does not build a profile of you from it.
When you contact us
We use your name, organisation and contact details only to answer your enquiry and to carry the conversation forward if it becomes an assignment. We do not sell, rent or share enquiry details with third parties, and we do not add you to marketing lists you did not ask for.
Research and assignment data
Data collected in the field belongs to the client and to the people who provided it. Our standing commitments on every assignment:
- Informed consent is documented before any interview or survey begins, in the language of the respondent
- Safeguarding and data-protection protocols apply to every enumerator and every device used in the field
- Personal identifiers are separated from analytical datasets, and reporting is at a level that does not identify individuals
- Ownership, hosting, retention and deletion are agreed in writing with the client at inception, before collection starts
- Case material from mental health and psychosocial work is confidential and is never used for illustration or marketing
Responsible data and AI
AI is part of how we work, and we are explicit about where it helps and where it must not be trusted. The principle is simple: the model accelerates, a person decides.
Where we use it
Transcription, translation support, thematic coding of qualitative material, summarising long documents, cross-checking datasets and drafting routine text. All of it is assistive work on material we already hold.
Human review is not optional
A named specialist reviews any AI-assisted output before it informs a conclusion. Accountability for every finding rests with that person, never with the tooling. No finding is published on the strength of a model alone.
What we will not do
We do not put identifiable participant data into public or consumer AI tools. We do not use client data to train models. And we do not let an automated system make a judgement about an individual's eligibility, risk or entitlement.
The rules we hold ourselves to
- Traceability. If a finding cannot be traced back to the evidence that produced it, it does not go in the report. AI assistance does not change that rule.
- Disclosure. Where AI-assisted analysis has materially shaped a deliverable, we say so in the methodology, alongside the sampling and the limitations.
- Data minimisation. Material is de-identified before any assisted processing wherever the analysis does not require identifiers.
- Bias is a methods problem. Assisted coding is checked against a human-coded sample, because a model trained elsewhere does not know the context we are working in.
- The client can say no. If you would rather no AI assistance was used on your assignment, tell us at inception and we will work without it.
Questions about either statement, or about what we hold on a specific assignment, should go to the office. We will answer them properly.
This page was last reviewed in August 2026. It describes practice, not aspiration; where a commitment here conflicts with a signed client agreement, the agreement governs.