Impact measurement and evaluation
Turn outcome numbers into a plain-language impact story
Helps you explain results so board members and funders actually remember them.
Take care first — Share totals and percentages only, and be careful with very small groups that could identify someone.
Please note: This site is only a guide to help people understand. Consult your IT, security and/or privacy department for clarification.
Before you start, gather
Copy [OUTCOME NUMBERS AND TIME PERIOD] from your final reported data for that period. Note the group size behind each figure. Set [AUDIENCE] to board, funder or public.
Starter prompt
Write a short impact story for [AUDIENCE] using only these figures from [PROGRAM NAME]: [OUTCOME NUMBERS AND TIME PERIOD]. Explain what the numbers mean in plain English, avoid overstating cause and effect, and include one honest limitation. Do not add percentages, comparisons or causes I did not give; write MISSING INFO instead of guessing.
Paste the prompt into your AI tool first, then replace the placeholders with your own details. Don't use real names of clients or donors.
Open an AI tool
You are leaving this guide. Copy the prompt first, then open a tool. The link goes to the website of the company named on each link, in a new tab. That company's terms and privacy policy apply there. Paste the prompt in, then replace the placeholders with your own details. Don't enter real names or confidential details.
- Open Claude (Anthropic) (opens Anthropic's website in a new tab)
Anthropic documents pre-filled links only for the Claude Desktop app, not for the website, so copy the prompt and paste it in.
- Open ChatGPT (OpenAI) (opens OpenAI's website in a new tab)
OpenAI does not document a link that pre-fills a prompt, so copy the prompt and paste it in.
- Open Gemini (Google) (opens Google's website in a new tab)
Google does not document a link that pre-fills a prompt, so copy the prompt and paste it in.
- Open Grok (SpaceXAI) (opens SpaceXAI's website in a new tab)
SpaceXAI (the company behind Grok) does not document a link that pre-fills a prompt, so copy the prompt and paste it in.
- Open Amazon Bedrock (hosts Llama; needs an AWS account) (opens Amazon Web Services' website in a new tab)
Llama has no chat website of its own, and Meta's own assistant (Meta AI) now runs on a different Meta model. One host is Amazon Bedrock: sign in to AWS, open a playground, and paste the prompt.
- Open Microsoft Copilot (Microsoft) (opens Microsoft's website in a new tab)
Microsoft documents no link that pre-fills a prompt, so copy the prompt and paste it in. The Copilot page shows a Sign in button, and you may need to sign in.
- Open Perplexity (Perplexity AI) (opens Perplexity AI's website in a new tab)
Perplexity documents no link that pre-fills a prompt, so copy the prompt and paste it in.
- Open Mistral Le Chat (Mistral AI) (opens Mistral AI's website in a new tab)
Mistral renamed Le Chat to Vibe, at the same web address. Mistral documents no link that pre-fills a prompt, so copy the prompt and paste it in.
- Givebutter (Givebutter): no website to open from here. Givebutter Magic works inside a Givebutter account; there is no public chat page.
- Grantable (Grantable): no website to open from here. Grantable's AI works inside a Grantable account; there is no public chat page.
Check the result
Check every number, and make sure claims do not say your program caused something your data can't show.
What it might get wrong
New percentages or year-over-year comparisons it works out itself. Cause words like resulted in, when the data only shows two things side by side. A limitation that is generic rather than yours.