AI training for charity teams: practical skills for the work they do

AI training for charity teams is most useful when it starts with the work people already do: preparing a report, planning an appeal, organising a meeting or responding to a routine administrative question. A session built around those tasks gives colleagues a way to judge whether a tool helps, where its output needs checking and when it should not be used.

Charities have varied roles, budgets and information risks. A fundraising officer, volunteer coordinator and finance manager may all use the same product but need different skills. Buying access or attending a general demonstration does not establish that each person can apply it responsibly in their role.

This guide describes a practical training plan, with illustrative exercises and follow-up. It does not assume that AI will reduce headcount or that sensitive service information belongs in a training tool. The aim is useful capability the organisation can support after the trainer leaves.

Ask teams where work becomes difficult

Collect a small number of recurring tasks from each participating role. Ask what the person is trying to produce, what information they use and where time is lost. “Writing reports” is broad; “turning approved project notes into a first draft of the monthly narrative” is a teachable task with a clearer review boundary.

Look for tasks where the output can be checked against a known source. This gives learners a concrete way to evaluate a result. Avoid selecting a high-stakes decision merely because it is time-consuming. Advice to beneficiaries, safeguarding judgements and financial approvals need appropriate professional processes and accountable people.

Ask about access and confidence. Some colleagues may have approved business accounts, others may only have seen consumer tools at home. Volunteers may have different permissions from employees. The training should not assume everyone can create an account, install software or upload files on the day.

Use the findings to define a short set of learning outcomes. For example, participants might learn to write a task brief, check a draft against approved evidence and recognise information that must stay out of the tool. Those outcomes are more useful than a promise to cover every popular product.

Agree the information rules before the session

Confirm the tools and account configurations approved by the charity. Ask the relevant information and technology leads what can be entered, what must be excluded and where outputs may be stored. Training is easier when staff have a clear rule they can apply rather than a vague instruction to be careful.

Prepare synthetic examples for practice. A fictional programme summary can teach report drafting without exposing beneficiary records. A made-up event budget can support an exercise in checking assumptions without sharing confidential accounts. Label the materials clearly so nobody later treats the examples as real organisational evidence.

Explain that removing a name does not necessarily make information safe to share. A combination of circumstances can still identify a person. Staff should use the charity's approved review route if they are unsure, rather than deciding independently that a sensitive case can be entered after a few details are changed.

Keep credentials out of training conversations and shared worksheets. Account setup, access problems and confidential information should use the organisation's normal support channels. A trainer needs enough access to deliver the agreed session, not unrestricted access to the charity's systems or supporter database.

Build a programme around role-based practice

A useful session can combine a short shared introduction with separate exercises for different teams. The common part explains task briefing, source checking and information boundaries. The role-based work then shows how those principles apply to the material each group actually handles.

• Administration: turn synthetic meeting notes into draft actions, checking owners, dates and decisions against the source.
• Fundraising: prepare an outline from an approved fictional evidence pack, identifying claims that still need verification.
• Reporting: compare a draft narrative with a small synthetic table and find statements the figures do not support.
• Volunteer coordination: draft a clear event instruction using a supplied schedule, without adding unapproved commitments.

These are illustrative exercises, not a complete curriculum or a recommendation to automate every task. Choose examples that match the participants' responsibilities and existing systems. A team with limited digital confidence may need more time on one task rather than several rapid demonstrations.

Leave room for questions about situations the tool cannot resolve. Staff should finish knowing where to seek help and who authorises a new use. Otherwise an engaging session can inadvertently encourage experiments outside the organisation's agreed information and decision boundaries.

Teach a reusable task-briefing method

Ask learners to state the task, audience, approved source, required format and constraints. For instance: “Prepare a short internal action list from these fictional meeting notes. Keep only agreed decisions. Mark missing owners as questions. Do not infer a deadline.” This makes the desired output easier to inspect.

Have learners compare a broad request with a more specific one using the same source material. Ask which result is easier to review and why. The lesson is not that a particular wording guarantees quality; it is that a clear task helps expose whether the output satisfies the business requirement.

Include a stopping rule. If the output introduces a fact absent from the source, the learner should identify and resolve it rather than repeatedly asking the tool to sound more confident. If the underlying task remains unclear, return to the brief or ask the responsible colleague.

Provide a simple editable worksheet instead of a collection of supposedly perfect prompts. Tools and interfaces change, while the need to define a task and check evidence remains. Staff should be able to adapt the method to a new approved product without relearning the whole process.

Practise checking an output against its source

Give participants a deliberately flawed draft based on a synthetic source. Include a wrong date, an invented action owner and a sentence that turns an aspiration into an achieved result. Ask the group to identify the errors and explain what evidence would be needed before publication.

Use a second draft that is accurate but awkward. This helps distinguish factual review from editing for clarity. A polished paragraph can be wrong, while a clumsy paragraph may preserve the facts. Staff need to evaluate both dimensions instead of treating fluent wording as evidence of reliability.

For numerical tasks, require a check against the actual data and calculation method. A generated explanation of a spreadsheet is not a substitute for understanding which cells, filters and definitions produced the result. Keep the financial review with the person authorised to interpret the figures.

The NIST generative AI profile offers a broader voluntary risk-management reference. Within training, keep the application concrete: show the source, identify the claim and decide who can verify it before the output is used in real work.

Work through an illustrative reporting exercise

Imagine a charity preparing a monthly internal update from approved programme notes. The synthetic notes describe sessions delivered, a scheduling issue and a planned service expansion. The exercise asks learners to create a concise draft without including any personal information or making claims about outcomes not measured in the notes.

The first output says the planned expansion has already launched. Learners mark the sentence, compare it with the source and rewrite it as a future plan. The group discusses why this matters: a small tense change can misrepresent the charity's position even when the rest of the paragraph sounds reasonable.

Next, the trainer removes an important date from the source. The correct response is to flag the missing information and ask the owner, rather than generate a plausible date. This tests whether learners can recognise the limits of the material instead of only following a recipe for producing text.

Finally, a colleague reviews the completed draft using the normal organisational route. The exercise ends with a decision about what can be used, what needs clarification and where the approved version should be stored. That final handoff is part of the skill being taught.

Make fundraising practice accountable

Fundraising exercises should start with approved facts and a clear audience. Ask learners to identify which statements describe evidence, which express the charity's view and which remain ambitions. This reduces the chance that attractive draft language turns a proposed programme into an unsupported public promise.

The Fundraising Regulator's AI guidance provides a relevant UK reference for responsible fundraising use. Training should help staff apply the charity's review responsibilities, rather than presenting AI-generated appeals as automatically suitable for release.

Practise shortening a message without losing qualifications. A longer explanation may become misleading when reduced to a subject line or social post. Give learners a source statement and ask them to preserve its meaning across formats, then explain what they chose to leave out and why.

Keep promotional permissions and supporter preferences separate from writing practice. Being able to produce a campaign does not establish authority to send it to a list. The training plan should identify who approves recipients, scheduling and suppression checks in the charity's actual systems.

Include colleagues with different access needs

Ask about accessibility and practical arrangements before delivery. Participants may need captions, readable handouts, keyboard access, more processing time or an alternative to speaking in front of the group. Offer a way to raise needs privately rather than asking people to disclose them during the session.

Use a pace that allows learners to inspect outputs. Rapid demonstrations can look impressive while leaving participants unable to repeat the task. Build in pauses, written instructions and opportunities to compare answers with a colleague. Keep essential content available without requiring someone to copy text from a projected screen.

Provide an account-free observation route if appropriate. A learner should not be pressured to use a personal account because organisational access was not ready. They can still practise writing a brief and reviewing a supplied output, with the limitation recorded in the training evaluation.

Check that remote or hybrid participants can take part in the actual exercises. A video feed of an in-room presentation may be insufficient if the worksheets, instructions and discussion are not accessible online. Delivery format affects preparation and support, so scope it before agreeing the session length.

Evaluate whether learning carries into work

Before training, ask participants to complete a short task or explain how they would review an output. Afterward, use a comparable task with different material. Look for changes in their ability to define the request, identify unsupported statements and apply the information rules.

Avoid treating confidence alone as proof of competence. A participant may feel comfortable using a tool while missing a serious error. Combine self-reported confidence with observed practice and an explanation of the checks performed. Use the results to identify where more support is needed.

Agree a follow-up period with team managers. Ask which approved tasks people tried, what required correction and whether review time changed. Do not demand detailed logs containing sensitive source material. A concise task description and outcome can provide useful learning without copying the work itself.

Measure the complete task if you discuss efficiency. Drafting time, checking time and correction effort all matter. The automation ROI guide explains the distinction between released capacity and cash savings. Training should not promise a payroll reduction based on a quick classroom demonstration.

Commission a scoped training engagement

Tell the provider who will attend, what tasks matter, which tools are approved and how learning will be checked. Specify whether preparation, role-specific materials and follow-up are included. A generic day rate does not tell you whether the session will address the charity's real operating needs.

An on-site session may need separate groups, preparation interviews or several days where the organisation is large. A single day can introduce a shared method and practise selected tasks; it cannot establish competent use across every department and specialist responsibility. Agree realistic outcomes for the available time.

Our AI training service can be scoped around these practical requirements. For a charity with broader process problems, the first step may be to review the work and information flows before designing training. Teaching staff to compensate for a broken handoff is rarely a sustainable solution.

Ask for reusable materials, a handover to the internal owner and a clear route for questions. The charity should retain the method and approved examples after delivery. Training becomes more valuable when managers can support practice and new starters can understand the same information boundaries.

Keep the programme current without chasing every release

Assign someone to review approved tools and material when important conditions change. A new interface may require a small update; a change in data arrangements or available features may require a more substantial review. Staff need a clear notice when their permitted use changes.

Maintain a small library of approved examples with owners and review dates. Remove obsolete instructions from ordinary use while retaining necessary history. Encourage colleagues to submit improvement ideas for review, rather than letting unofficial prompts and copied guidance become the organisation's policy by repetition.

Use recurring questions to shape the next session. If people struggle to check numerical summaries, focus on that skill. If access or review responsibilities remain unclear, resolve those operational issues first. More product demonstrations will not fix an organisational decision that nobody has made.

Useful charity AI training leaves staff more able to explain their task, question an output and know when to involve another person. Those habits support fundraising, reporting and administration without pretending the tool understands the charity's responsibilities or the people it serves.

Checked sources

• NIST: generative AI risk-management profile
• Fundraising Regulator: using AI in fundraising