Write an impact report
Turn activity data and beneficiary feedback into a report that separates outputs from outcomes and is honest about the gaps.
Yours to copy, change, and make your own.
Click any [BRACKETED PLACEHOLDER] and type your own material straight into the prompt.
Help me write the impact section of a report for a UK charity. Audience: [AUDIENCE] Period covered: [PERIOD] What we did, with numbers: [OUTPUTS] What changed as a result, with any evidence: [OUTCOMES AND EVIDENCE] Beneficiary feedback we have consent to use: [FEEDBACK] What did not go well: [WHAT DID NOT WORK] Length: [WORD COUNT] Structure: 1. Headline: the single most important change this period, in one sentence. 2. What we did: outputs, as figures, briefly. This is not the impact. 3. What changed: outcomes, each tied to the evidence supporting it. 4. What we learned, including what did not work and what we changed because of it. 5. What is next. Rules: - Never present an output as an outcome. "We ran 40 sessions" is an output; "31 of 40 participants reported sleeping better" is an outcome. Flag anywhere I have given you an output where an outcome is needed. - Use only my figures. Do not round, extrapolate, or annualise. - Where a claimed outcome has no evidence behind it, say so plainly in the text rather than dropping it. - Use British English. Finish with the claims a sceptical funder would most want evidence for.
- [AUDIENCE]
- Who reads this: trustees, a funder, supporters, the public.
- [PERIOD]
- The period the report covers.
- [OUTPUTS]
- What you delivered, with the figures exactly as you hold them.
- [OUTCOMES AND EVIDENCE]
- What changed for people, and what you have that shows it.
- [FEEDBACK]
- Quotes or survey responses, anonymised, that you may publish.
- [WHAT DID NOT WORK]
- What underperformed or was dropped, and why.
- [WORD COUNT]
- Roughly how long the section should be.
Where it shines, and where it falls over.
- Annual reports and trustees' annual reports
- End-of-grant reports back to a funder
- Turning a year of monitoring data into something a supporter will read
- Give it the bad news too. Reports that admit what did not work read as more credible, and funders notice.
- Ask it to rewrite your outcomes as a short theory of change. The gaps in your evidence become obvious.
The characteristic failure is quiet inflation. Asked for impact, a model will phrase an output as though it were an outcome, attach a cause to a correlation, and smooth over the point where your evidence runs out. Read section 3 line by line and ask of every sentence: what would I show a funder who asked me to prove this?
Numbers are the second risk. Models merge figures, attach them to the wrong period, and helpfully annualise a quarter. Check each one against your source. Nothing identifiable about a beneficiary should go into a public AI tool either, quotes included, unless it is anonymised and you hold consent to publish it.
AI output is a first draft, not a finished product. You are responsible for whatever you send, publish, or decide with it.
The Institute of AI sets the standard of AI practice, works to that standard itself, and puts it within reach of everyone else. The AI for All Prompt Library is one of the ways it does that.
More prompts worth exploring.
Draft a difficult email
Get a firm, fair first draft of an email you have been putting off: chasing, declining, apologising, or pushing back.
Write a grant application
Turn your project notes into a funder-ready answer: need, response, outcomes, and evidence, inside the word limit.
Draft a fundraising appeal
Get a first draft of an appeal letter or email built around one beneficiary story and one clear ask, in your charity's voice.