⭐ Our core principles

These hold whichever tool you use and whatever the task.

• Give it context:
  AI doesn’t know your organisation, your audience or your purpose unless you tell it. The quality of what comes back reflects the quality of what you put in.
  
• Be clear on audience and purpose:
  Say who the output is for and what you want them to think, feel or do. The same content tells a very different story depending on who you’re talking to.
  
• Ask it to ask you questions:
  Consider ending a prompt with "ask me any questions before you start". It turns a one-way request into a proper back-and-forth.
  
• Review, push back, refine:
  The first response is the start of a conversation, not the finished product. Say what’s not right, be specific, and go again - exactly as you’d brief a colleague.
  
• Check before you trust:
  AI gets things wrong confidently and agrees with you too readily. Verify facts against original sources, and ask for counterarguments when it matters.
  
• Protect people’s data:
  Describe roles and situations, not names. Anonymise before uploading. If someone could be identified or harmed, it doesn’t go in.

🪜 Building your first prompt: REACH

Throughout our Use Cases you’ll notice a prompting structure called REACH. There are many prompt structures out there; REACH is designed for the nonprofit sector - it’s accessible, covers the key components of a good prompt, and reminds us (and the AI) to check for potential harms.

REACH stands for:

• R - Role: who should the AI act as?
• E - Expectation: what do you need it to produce, and in what format?
• A - Audience: who is the output for?
• C - Context: what background information does the AI need?
• H - Harm check: what risks should it watch for and flag?

One thing to be clear about: REACH is how you build your opening prompt. It gets the conversation off to a strong start, but it’s not the whole method. The next section covers what comes after.

REACH in action

Here’s the prompt from our "Draft a presentation outline" use case, with each part labelled. Every use case in the library follows this same pattern - you’ll recognise it everywhere.

‍R — Role: Act as a communications professional with experience in the UK charity sector.‍

E — Expectation: Help me build a clear, structured outline for a presentation. The outline should include a title, a suggested opening, 4–6 main sections with brief descriptions, a closing and call to action, and any suggested slides or visuals where these would help.‍

A — Audience: Start by asking me one or two questions at a time about my audience, the purpose of the presentation, and the key messages I want the audience to take away. 

C — Context: Then ask me questions about the length and format it needs to be; my organisation and the context for this presentation. Then ask if I have any drafts or notes to upload. Then draft the outline. 

H — Harm check: As you draft, watch out for and avoid: jargon or insider language that a non-specialist audience might not follow; assumptions about what the audience already knows; and any framing that talks about the people we support in a stereotyping or disempowering way.

Something to consider: to help open the conversation, invite clarifying questions. For example, for the above prompt, you might add: “Before producing the outline, ask me any clarifying questions that would help you get the structure right.”

🔁 Working with what comes back

REACH gets the conversation started. Getting a genuinely useful result usually takes two or three exchanges - think of the AI as a collaborator you’re working through ideas with, not a machine that produces a finished product on the first go.

Push back like you’d challenge a colleague

If the first response isn’t right, say so and be specific - for example: "This is too formal", "the second section should come first", "the opening doesn’t grab attention, suggest three alternatives". Vague feedback gets vague revisions; specific feedback works as it does with people.

Threads and topics

Stay in the same conversation while you’re refining one piece of work as the AI keeps the full context. Start a new conversation when you change topic, because a long thread about one subject will colour what it says about the next.

Add what only you know

Once you have a solid draft, fill in what the AI can’t know: real stories and examples only your team could tell, your actual data, recent developments, and your organisation’s personality and voice. That last layer is what turns a competent draft into something that’s genuinely yours.

✅ Keeping people’s data safe

The rule of thumb: how much you share depends on your existing privacy and data policies, and how much control you have over the tools you use.

• Free consumer tools (ChatGPT, Claude, Gemini free plans): assume the least control. Nothing sensitive - no personal data, case details or anything that could identify the people you support. Check the settings: most train on your data by default unless you switch it off.
• Paid and organisational accounts: these offer more control over data handling, but the arrangements vary - if your organisation is adopting a tool formally, the data processing terms need a proper review.
• Worthwhile Chat: this tool is privacy-first by design: your data stays in the UK and EU and is never used to train models. Uploading documents is fine but good data practice still starts with you.

And whatever the tool:

• Describe roles, not names: "A trustee who consistently pushes back on technology spending" gives the AI everything it needs without creating a potential data protection issue.
  
• Anonymise before uploading: Review documents for personal or sensitive information about beneficiaries, and remove names and identifying details first.
  
• If you’re unsure, ask yourself: if this were seen by someone outside our organisation, could anyone be identified or harmed? If yes, anonymise first.
  
• Work within your organisation’s AI policy: if you have one, and if a new tool comes into your workflow, check its data settings before anything of yours goes in. Many train on your data by default.

📋 Reviewing the output

Nothing AI-generated should go anywhere without your judgement applied to it first. Use this checklist before you share, publish or present. Individual use cases add questions specific to their task - this is the foundation they all build on.

Accuracy

• Check the facts: Verify statistics, names, dates and specific claims against your original sources. AI can misread a figure or fill gaps with plausible-sounding but inaccurate information. Click any links it gives you; AI-generated URLs can sometimes lead nowhere.

Voice and audience

• Does it sound like us? Our voice, values and priorities - or a generic charity document?
• Would this audience follow it? No jargon or insider language, no assumptions about what they already know, and a structure that fits the format.

People and representation

• Are the people we support described in a way they’d be comfortable with? Focus on what people have achieved, not only what they’ve been through.
• Check consent for quotes: Anyone quoted - colleagues, beneficiaries, partners - has agreed, is happy with how they’re represented, and can’t be identified if they haven’t consented.

Judgement

• Get a human review: Ask a colleague to read it fresh: do they recognise the organisation in it? Does anything feel inaccurate, out of character, or likely to land badly? Would you stand behind it if challenged?
• Be transparent about AI use: Consider noting where AI tools helped prepare the material. More organisations are being open about this, and it’s in line with Worthwhile AI values.

🗄 Uploading files to Worthwhile Chat

If you have documents ready to share - and you’ve done the data check above - you can upload them directly into Worthwhile Chat. It currently accepts PDF, DOCX, XLSX and CSV files. Rough is fine: previous reports, old presentations, a page of bullet points all give useful context.

1.      Click the + button in the bottom left of the chat box.

2.      Select Upload Files from the menu.

3.      Choose the files you want to share from your device.

4.      Press the arrow button to send.

🖼️ Worthwhile Chat master prompt 

Worthwhile Chat uses a master prompt. It includes operating context, intended audience, differentiators, purpose and some general behaviours.

To see the full master prompt, go to the FAQ page and look under Features & Interface.