AI can now draft a report in minutes. It can also invent a number or a source, and say it with complete confidence. Evaluation AI gives you a simple, no-code way to check every claim before it reaches a donor. Written for evaluators, not engineers.
No code. No jargon. Works with the AI you already have.
You stay in chargeWhen a funder asks “how do you know?”, a confident-sounding paragraph is not enough. It has never been easier to produce a polished report, and it has never been easier to ship an error with it.
A consulting firm had to repay part of a government report after AI-generated errors were found in it.
fake citations from AI tools entered the scientific record in a single year, according to researchers.
evaluation teams now use AI, but almost none can prove which sentences came from a source and which were invented.
You do not need to understand how AI works. You need a routine that refuses to let an unsupported claim onto the page. We call it a harness. It is plain English and it takes an afternoon to set up.

Everything is built for busy people who care about rigour, with no assumption that you are technical.
The full method in plain language, plus the checklist. Yours to keep and share.
Get the free bookBuild your first harness with us, step by step on video. Early-bird $199.
See the courseWe set the whole system up for your organisation, inside your own account.
Speak with us
You keep using the AI tools and documents you already use. The harness runs inside your own account, so a client's confidential material never leaves your control. If you can use Word and a chatbot, you can use this.
The drafting is slow and repetitive. Our method lets AI do the first pass, while you keep the judgement and the credit.
You need the pattern, not another spreadsheet. AI helps you see it, and our checks make sure you can stand behind what you find.
Let the machine do the typing. You spend your time on analysis and meaning, not on listening back at half speed.
So were we. That is exactly why this whole system exists: to make AI safe enough to actually use.
Everything is written for people who did not grow up with technology, and who cannot afford a mistake in front of a funder.
The rules, the test set, the checklist and the proof table, explained from the beginning.
Copy-and-paste instructions for reports, logframes, indicators, interviews and data.
One page to review any output before it reaches a donor.
Every technical word defined, so nothing is assumed.
See a real paragraph repaired, line by line, with the sources shown.
How to keep client data safe, and the exact wording to disclose your use of AI.
Understand the problem and the four-file method. About forty minutes.
Paste the rules into your own AI. One afternoon, once.
Run the checklist before anything reaches a donor. Minutes each time.
A curated, plain-English directory of AI tools for monitoring and evaluation, sorted by the work you actually do: data collection, analysis, reporting, design, visualisation and learning.
Surveys · Transcription · Analysis · Reporting · Visuals · Learning
No. The whole method is plain English, and every term is explained from scratch.
No, it protects it. AI drafts; you decide. The checks make the machine's guesses visible.
The method runs inside your own account, and we show you exactly how to keep client information private.
With the free handbook. It costs nothing, and it may be all you ever need.
No spam. Unsubscribe any time.
Tell us about your programmes and your donors. We'll show you exactly what your team's setup would look like, no obligation.