A grant consultant can use AI to draft a funder progress report by giving it a tidy outcomes table and anonymised case notes, asking it to write only from those, then checking every claim against records. AI handles structure and wording; the consultant verifies outcomes, quotes and figures before the report reaches the funder.
Most reporting templates online are empty boxes. The difficult part is turning a messy spreadsheet and a folder of case notes into something that fits them. That is where drafting helps, as long as the data and the checking stay yours.
The report skeleton that works for most funders
If a funder supplies a form, use it. If not, or when the form's boxes are open-ended, this skeleton covers what most funders want to read:
- Summary: three or four sentences with the headline result and the main issue.
- Outcomes against targets: a table, one row per agreed outcome, with target, result and a status.
- Activity narrative: what was delivered in the period, in the order of the work plan.
- Beneficiary evidence: anonymised stories or quotes, with consent noted.
- Lessons learned: what did not go as planned, why, and what changes.
- Finance note: spend against budget, with explanations of variances.
- Next period: planned activity and any request, such as a timeline change.
Match order and headings to the funder's guidance whenever it exists, because reviewers tick against their own list. The skeleton is the fallback.
Prepare the data first: the part AI cannot do
Drafting quality depends on input quality. Before any prompt, build three things.
An outcomes table. Columns: outcome as written in the funding agreement, target, result this period, cumulative result, evidence source. If the evidence source cell is empty, you do not yet have a claim you can make.
Anonymised case notes. Convert each note to a short summary with no name, address, date of birth, health detail or distinctive combination of facts. Use a code such as "Participant 14". If the grantee wants quotes in the report, confirm consent first and record it.
A variance list. For each target missed or exceeded by more than 10 per cent, a one-line reason in the grantee's own words. The model should not invent reasons, so you must supply them.
If you track grants in a spreadsheet already, much of this exists. Grant tracking software: spreadsheet, paid tool or AI-built sheet compares ways to hold it, and tracking grant expenses against budget lines covers the finance side.
The Complete Excel Course with AI Integration
From formulas and pivot tables to Power Query, DAX, VBA and AI with Copilot, ChatGPT and Claude — hands-on from the first lesson.
By Sagnik Bhattacharya and Virginia Thorn, Primetime Emmy Award winner
Buy nowA worked example: a literacy project's six-month report
Say a small charity runs a reading project in primary schools and has a six-month progress report due. The following figures are illustrative. The funding agreement lists four outcomes:
| Outcome | Target (12 months) | Result at 6 months | Evidence | Status |
|---|---|---|---|---|
| Children enrolled | 120 | 104 | Register export, 30 June | On track |
| Children attending 80% of sessions | 90 | 61 | Attendance sheet | Behind |
| Children improving reading age by 6+ months | 70 | Not due until month 12 | Baseline test results | Baseline complete |
| Parents attending a workshop | 40 | 47 | Sign-in sheets | Exceeded |
The variance notes from the grantee: attendance is behind because two schools moved session times in March, and a third school lost its room for three weeks. Parent workshops exceeded the target because the charity ran an extra evening session.
The prompt that drafts from the data
You are helping a grant consultant draft a six-month progress report to a funder.
Use ONLY the data below. Do not add facts, numbers, quotes or outcomes.
Do not round, calculate or compare figures other than those given.
Where something needed is not in the data, write [NOT IN DATA: what is missing].
Tone: plain, honest, specific. Match the funder's form and spelling.
Sections: Summary (60-80 words); Outcomes against targets (a sentence per outcome);
Delivery narrative (150-200 words); Lessons learned (3 bullet points, each with a reason
and an action taken, using the variance notes only).
Funder's instructions: [paste headings and word limits]
Outcomes table: [paste]
Variance notes: [paste]
Anonymised case summaries: [paste]The line telling the model to write [NOT IN DATA] gives you a clean list of gaps, instead of confident invention. A fair illustrative output for the summary reads: "At the six-month point, 104 of a 12-month target of 120 children are enrolled and 47 parents have attended a workshop, against a target of 40. Attendance at 80 per cent of sessions is behind plan, with 61 children against a target of 90, because of timetable changes in three schools. Baseline reading assessments are complete." Every number in it appears in the table. What to fix: the phrase "behind plan" is accurate, but the funder's form may ask for a rating such as amber, so you adjust to the form.
Evidence checks before anything is sent
This is the step that makes the draft safe. Do each of these, and record that you did.
- Figure tie-out. Ask the tool to list every number in the draft, then tick each against the outcomes table and the evidence source. 104, 120, 61, 90, 47 and 40 must all appear, unchanged.
- Source existence. For each outcome, confirm the evidence source named actually exists and says what the table says. A claim that rests on "the register export" needs the export on file.
- Cause check. Each reason offered for a variance is the grantee's own, not the model's. If the draft says "owing to staff illness" and nobody said that, strike it.
- Quote check. Any quote from a participant must match the original note word for word, with consent recorded, and no identifying detail.
- Claim strength. Replace "improved" or "transformed" with what the data supports. The reading-age outcome is not due until month 12, so a sentence implying progress on it would be unsupported.
- Grantee read-through. The person responsible for the project reads the whole report and confirms it matches what happened.
A realistic failure: the draft states that "all schools reported increased confidence among learners". Nobody collected that. The model had picked up a phrase from one case summary and generalised it. The tie-out finds no matching source, so it comes out. This is why the prompt says use only the data and why the evidence column exists.
Lessons learned that a funder will believe
Funders read the lessons section to judge whether you run the project well. Good entries have a cause, an action and, where possible, an early result. Weak ones are generic ("communication is important"). Ask the model to draft each lesson in that three-part shape from your variance notes, then improve them yourself.
Illustrative before and after. First draft: "Attendance was affected by scheduling issues." After: "Two schools moved session times in March, which clashed with after-school clubs for 19 children, and a third lost its room for three weeks. We now agree a fixed timetable with each school in September and have a back-up room arrangement with one school; we will report attendance for the affected groups at month nine." The second version is only usable if the 19 children and the back-up room are real, so the consultant checks those against the attendance sheet and with the grantee.
If the funder wants frank assessment of what did not work, say so plainly. Do not let the model soften a miss into positive language. Ask for a neutral tone, then read the draft for euphemism.
Handling client data and consent
Your client is the data controller in most arrangements, and beneficiaries trust them with sensitive information. Three precautions keep you on the right side.
- Minimise. The model gets outcomes figures and anonymised summaries, not raw case files. If a task seems to need the raw notes, rethink the task.
- Pick the plan with care. Use a business or team plan whose terms say content is not used for training by default, confirm that on the vendor's current page, and switch off any history or memory features you do not need. Is it safe to put customer data into ChatGPT explains the differences between plan types.
- Agree it with the client. Put a line in your engagement terms: you may use AI tools to draft, using anonymised data, and you verify all content. Some funders and clients require disclosure of AI use, so ask which applies.
Keep the anonymisation key, if you use one, with the client and not in the tool. Delete working copies when the report is submitted, in line with your agreement.
Adapting the draft to a funder's form
When the funder's portal has set questions, such as "Describe progress against your outputs (max 300 words)", draft answer by answer. Paste the question, its limit and the relevant rows of the outcomes table, and ask for an answer that stays under the limit. Check the word count yourself, because models miscount, and some portals truncate silently. For scored or narrative-heavy reports, paste any scoring guidance and ask the model to list which criteria the draft has not addressed. Take that list as a prompt for your own review, not a verdict.
The same data can feed other documents. A one-page summary for trustees, a short donor update and the funder's report all come from the same table, with different levels of detail. Keep the table as the single source of truth, so the figures in each document agree. That habit also makes next period's report faster, because last period's table is the starting point. If you want prompts for the application stage that precedes all this, see ChatGPT prompts for grant writing: needs to outcomes, fact-checked, and for the logic model that defines the outcomes your report measures, building a logic model with AI.
Anonymising a case note before it goes near a tool
Case notes are the richest material and the riskiest. Here is the conversion in practice, with an invented note. Original, as held by the charity: "[first name], 8, in the third year at [school name] on [street name], started in January reading at the level of a five-year-old. Her mother told us she now asks to read at bedtime. Her teacher says her hand goes up in class." The converted version for the model: "Participant 14, primary age, started in January well below age-expected reading level. Parent reports the child now asks to read at home. Teacher reports increased participation in class."
What was removed: the name, the exact age, the year group, the school and its street. What was kept: the change, the sources of the evidence (parent, teacher) and the timing. Check for indirect identification too. A small village school with one reading project might still point to one child, so for very small groups, generalise further ("a younger participant") or ask the charity whether to use the story at all. Where the funder wants a quote with a first name, get written consent from the parent first, and keep the consent with the grantee's records. The model never needs to see the unconverted note.
The delivery narrative, drafted and then corrected
The delivery section is where drafts drift into marketing language. Use the activity log, which should list sessions delivered, partners and notable events. An illustrative log extract: 118 sessions delivered across five schools between January and June; two schools changed session times in March; one extra parent evening held in May; baseline reading assessments completed for 104 children by 30 April.
Ask for 150 to 200 words from that log only. A plausible output: "Between January and June the project delivered 118 sessions across five partner schools. Baseline reading assessments were completed for all 104 enrolled children by 30 April. Two schools changed their timetables in March, and a third lost access to its room for three weeks. An additional parent evening in May brought the total number of parents attending a workshop to 47." Compare with the log: 118, five, 104 and 30 April are present, 47 comes from the outcomes table. Good. Now notice what to fix. The output says "all 104 enrolled children" while the log says assessments were completed for 104 children; if enrolment is 104 as well, that is consistent, but if enrolment had been 110 the phrase "all" would have been wrong. This kind of small overreach is why the tie-out reads the draft against the source, not only the numbers.
A finance note with a variance explained
If your report includes spend, supply the numbers from the grantee's accounts and do not let the model calculate. Illustrative six-month figures: budget to date $37,800, spend to date $33,950, a variance of $3,850 under. Two reasons from the grantee: the coordinator started three weeks late, and the venue hire for the extra parent evening was donated. The funder wants to know what happens to the underspend.
Write a finance note of 80-100 words for a funder report.
Use only these figures: budget to date 37,800; spend to date 33,950; variance 3,850 under.
Reasons (from the grantee, use these only): coordinator started three weeks late;
venue hire for the May parent evening was donated.
Do not suggest what the underspend will be used for. Add [GRANTEE TO CONFIRM: plan for underspend].The placeholder matters. Whether to carry the underspend forward or request a variation is the grantee's decision, often governed by the funding agreement, so the draft leaves a visible gap instead of guessing. The consultant checks 37,800 minus 33,950 equals 3,850 in the sheet, and that the spend figure matches the grantee's ledger for the same period.
Keeping the data tidy month by month
The easiest way to speed up reporting is to stop collecting data at the end. Set up one shared sheet with the outcomes table and update it monthly, on a fixed day, by the person who runs delivery. Add three columns that make later drafting easier: the date of the last update, the evidence file or location, and a free-text note for anything unusual. When reporting time comes, the table is already current, the evidence is already filed, and the variance notes were written when the memory was fresh.
A small habit with a big return: when a target looks at risk, write the one-line reason in the notes column that month. By report time you have a log of reasons in the grantee's own words, which is what the prompt needs. Without it, you reconstruct causes from memory, and that is where invented reasons creep in. For the budget side of this routine, AI bookkeeping for nonprofits and restricted funds covers how to keep spend tagged to the grant.
Reusing the same data for several funders
Many small organisations have three or four funders, each wanting the same facts in a different shape. The outcomes table is the single source; each funder's form is a view of it. A practical way to work: keep a folder per funder containing the form, the prompt you used last time and the report you sent. Next period, paste the new table and the old prompt, and compare the new draft with last period's to check consistency. If you reported 104 children enrolled to one funder and 102 to another at the same date, a funder who talks to both will notice. Where the definition of an outcome differs between funders (for example "attended 80% of sessions" against "attended at least ten sessions"), record both definitions in the table and report each as the funder defined it.
Mistakes that reach funders
- Cumulative figures reported as period figures. The table has columns for both. State which the funder asked for, and label the column in the draft.
- Targets restated. A model asked to "summarise progress" may paraphrase the target. Paste targets from the funding agreement and ban rewording.
- Attribution claims. "The project led to improved results" implies cause. If you have no comparison group, say "children who took part improved", and describe what you measured.
- Stale baselines. The baseline test used last year's group. Check the cohort in each row of the table before you compare.
Each of these is a data problem the model will faithfully repeat, which is another reason the table is the thing to perfect before drafting.
What it costs in time
As a planning figure for a small grant: one to two hours preparing the outcomes table and anonymised summaries, thirty minutes for the first draft and prompt adjustment, and an hour on the evidence checks and the grantee's read-through. Compared with half a day writing from scratch, the saving is real but modest, and it grows when you reuse the same prompt and table layout across several clients. The checks do not shrink, and nor should they. If you want to set up a reporting routine with templates for each of your funders, that is what my AI implementation consultation covers, at the contact page.
Grant reporting questions consultants ask
Can AI write the whole funder report?
It can draft the narrative from your data, but it should not decide what happened. Give it the outcomes table and anonymised case notes, restrict it to those facts, and have the grantee or you verify every claim against records before the report goes to the funder.
What should never go into the AI tool?
Names, contact details, addresses, health or other sensitive details of beneficiaries, and anything the grantee has not agreed to share. Convert case notes to anonymised summaries first, and use a business plan with no training on content by default.
How do I handle a target that was missed?
State the target, the result and the gap plainly, then give the reasons and the action taken. Funders usually respond better to an honest explanation than to a vague one. Ask the model to draft in a neutral tone, then check the reasons are true.
What if the funder has its own form?
Use it. Paste each question and its word limit into the prompt and draft answer by answer. A funder's form is the structure; the template here is for when you are free to choose or when the form has open boxes.
How long does a report take this way?
For a small grant with a one-page outcomes table, an hour or two of drafting and checking is realistic once the data is tidy. Most of the time goes into the data and the evidence checks, not the writing.
Further reads
- Grant Budget Narrative: Draft It With AI and Reconcile Every Figure — Drafting the budget narrative and reconciling every figure.
- Grant Writer Rates: What to Charge When AI Drafts Faster — Pricing reporting work when drafting is faster.
- Can You Use AI for Grant Writing Without Funders Rejecting It? — How funders treat AI-assisted grant documents.
- How to Stop AI Tools Training on Your Business Data — Setting up tools so client content is not used for training.
- How to Use AI to Write Grant Applications, Step by Step — The application stage that comes before reporting.
- Best AI App for Grant Writing: What Each Does Well on a Funding Bid — Choosing an AI app for grant work.
- Claude AI for Charities: Five Admin Jobs Set Up With Projects — Five Claude Projects for charity admin, each with files to upload, instructions to copy, a sample output and what to fix before anything is sent.
- Sports Club Treasurer's Report: A Plain Summary of the Books With AI — How a volunteer treasurer builds a one-page table, has AI draft a plain summary, three talking points and member answers, then checks every figure.
- Can AI Write a Sports Club Grant Application? What Funders Expect — A sports club grant application drafted with AI: what funders assess, how to read their AI rules, and how the committee checks every claim.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: general grant-reporting practice and vendor business-plan privacy pages checked in 2026. The project, figures and case notes in the example are invented for teaching.