Turn approved evidence into a clear narrative. Use a practical prompt, check each claim, and preserve the report version your funder receives.
AI can turn an approved evidence pack into a first draft of a funder report. Give it the reporting brief, verified results, source references and known limitations; then check every claim before approval. The assistant helps with structure and wording. Your team remains responsible for the figures, interpretation, privacy and final submission.
Goal: Draft and review a report from a bounded evidence pack.
Start with: Use your approved report brief. If you write without AI, the same claim-by-claim review still applies.
Carry forward: Leave with a dated draft, review notes and an approval owner. This completes the core reporting journey. Continue to valuation only when a monetary estimate serves a real decision.
This lesson completes the core Impact Measurement and Reporting path. Bring the brief and evidence table from the donor-report planning lesson. You will finish this chapter with a reviewed narrative, a record of corrections and an approved report version. The optional valuation readings cover monetary estimates where appropriate; a donor report does not have to include an SROI ratio.
A useful evidence pack tells the assistant what it may say and gives the reviewer a way to check it. It can be a set of approved tables and documents; it does not need to be elaborate. Keep one index that identifies each source, its period, version and owner.
Do not upload an unrestricted collection of personal records when an approved aggregate table will answer the question. If the source includes sensitive material, use your organization’s approved access and data-handling arrangements. Confirm that the assistant can actually read the intended files; an inaccessible link is not a source it has verified.
A narrative report explains what the findings mean in the context of the work. It should connect delivery, observed outcomes, challenges and next actions. The number of findings depends on what matters in the period; there is no rule that three numbers are always enough.
For each main finding, ask: What happened? What supports that statement? What does the evidence leave unresolved? What should happen next? Keep those answers together. A paragraph that begins with a strong result should not hide its missing-data limitation in a distant appendix.
A narrative can be reused carefully. In Candid’s account of simplifying grant reporting, the team describes maintaining a substantial narrative document and adapting it to funder forms, with financial reports prepared alongside it. This is an example of a reporting practice, not a Sopact customer claim. Reuse approved facts, while checking the period and instructions for each submission.
Paste this prompt with your brief and evidence index. If the output is too long, shorten the prose after checking the claims; do not remove essential qualifications simply to fit a preferred word count.
Draft a funder progress report using only the approved evidence supplied below. Audience, purpose, period and required format: [insert brief]. Evidence release: [version and date]. Sources: [index and accessible content]. Organize the draft into delivery, outcomes, spending, limitations, learning and next actions, adapting to the required format. Attach the source ID to each factual claim. Preserve denominators, time periods and uncertainty. Include material shortfalls. Do not invent a quotation, calculation, explanation, financial value or causal conclusion. Mark missing information as “review needed” and name the question to resolve. Put any future funding request in a separate section and use only approved plans. Return the draft followed by a claim-and-source review table.
A source ID is a checkable reference, not proof by itself. The reviewer still needs to open the relevant source and see whether it supports the wording. If the assistant returns a citation that does not exist or cites a source that says something different, correct the claim before proceeding.
The following is a fictional training-program example using a self-contained fictional release A: 80 starters, 60 with a known employment status at 90-day follow-up and 36 employed. It is an excerpt for review, not a complete financial or donor report.
The cohort included 80 starters. At 90 days after exit, employment status was known for 60 participants, of whom 36 were employed. That is 60% of participants with a known status. Follow-up coverage was 75%, and 20 participants had an unknown status. Confirmed employment represents 45% of all starters. [R1: approved cohort and follow-up table, release A]
These findings describe employment at one follow-up point. They do not establish that the program caused employment or that the jobs were sustained. Twelve of 18 purposively selected interviewees mentioned transport difficulties. Those interviews identify a concern to investigate; they do not estimate its frequency across the full cohort. [R2: reviewed interview summary]
The team should examine gaps in follow-up and investigate the transport theme before deciding whether to change delivery. The final report also needs the approved financial statement and an agreed action owner. Those items are not supplied in this exercise and remain open for review.
The references R1 and R2 are fictional labels for this exercise. Replace them with your real source IDs. The draft deliberately does not invent an expenditure total, a renewal amount, a participant quotation or an explanation for missing follow-up.
| Check | Reviewer question | Example defect |
|---|---|---|
| Scope and period | Does the draft use the agreed cohort and follow-up point? | A Q3 report uses an earlier quarter’s attachment. |
| Counts and percentages | Do they match the approved calculation and denominator? | 60% of known statuses becomes 60% of all starters. |
| Citations | Does the source exist and support this exact claim? | A link opens, but contains no evidence for the sentence. |
| Interpretation | Does the conclusion stay within the study’s limits? | Employment at 90 days becomes sustained employment caused by the program. |
| Financial statements | Has finance approved the same reporting scope? | An unexplained underspend is called efficiency. |
| Privacy and quotations | Is this use approved, necessary and accurate? | A private interview is quoted in a public report without appropriate permission. |
Review arithmetic outside the narrative generator when possible: use the approved calculation or query, and compare the result with the draft. For the example, 36 ÷ 60 = 60%, 60 ÷ 80 = 75% and 36 ÷ 80 = 45%. These percentages answer different questions and should not be substituted for each other.
Next, review words that imply causality or certainty: “caused,” “proved,” “all,” “sustained” and “guaranteed.” Some may be justified by an appropriate study; none should appear merely because they make the narrative stronger. Ask whether a reasonable reader would infer more than the method supports.
NIST’s Generative AI Profile describes the risk of confidently generated false content, including problematic reasoning or citations. Giving a model good sources helps define the task, but does not make review unnecessary.
Compare this draft with the supplied evidence pack. List each unsupported claim, changed denominator, mismatched period, invented quotation, omitted material limitation and inaccessible or incorrect source reference. Quote the draft sentence and the source passage relevant to your concern. Propose a narrower correction. Do not mark the report accurate merely because the language is consistent. Return unresolved items for human review.
This second pass is a useful aid, not an independent audit. The same system can miss an error twice. A reviewer should still inspect high-stakes claims and reconcile the report against the approved tables. Record the correction and its reason so the next reporting cycle can reuse the lesson.
A funder may need progress against commitments; a program team may need follow-up actions; a board may need material risks and resource decisions. The emphasis can change, but a different audience is not permission to change the denominator, reporting period or level of certainty.
Keep a common approved evidence release and create audience-specific summaries from it. If a required indicator uses a different definition, calculate and label it separately. Do not relabel the same number to imply it satisfies a different standard.
In a configured Sopact workflow, connected collection records, documents and reporting context can support that work. Test the workflow with one real reporting requirement: can the team retrieve the source, inspect a result, identify missing evidence and control who sees it? Treat demonstrated behavior as the basis for your process, rather than assuming every requested export or approval feature is available in every setup.
A live dashboard and an approved report serve different purposes. The dashboard can show current information; the report preserves what was submitted for a defined period. One public link should not silently erase that distinction.
Watch the video · 11 minutes 58 seconds. The walkthrough shows impact reporting with Theory of Change, IRIS+, SDG and IMP context. Use it to examine how evidence and framework context inform a report; verify the current framework edition and your own configuration separately. Browse the video library.
Start with the agreed scope and reporting requirements. Explain delivery, outcomes, spending, challenges and next actions using approved evidence. Keep each important finding with its source and limitations, then have the responsible reviewers approve the report.
Provide the reporting brief, approved result tables, definitions, source register, financial statement and known limitations. Use only information permitted for the task and confirm that the assistant can access the intended sources.
It can help identify inconsistencies and unsupported claims, but a second AI pass is not independent assurance. A human reviewer should reconcile important figures, open source references and check whether conclusions fit the evidence.
Use approved calculations or queries for report figures and compare them with the draft. If an assistant proposes a calculation, check its inputs, filters, denominator and formula before approving the result. Do not rely on arithmetic embedded only in generated prose.
Yes, where the scope and evidence are appropriate. Adapt the framing and required format while preserving definitions, dates and limitations. Calculate a separate result when another funder requires a genuinely different measure.
Preserve the submitted report, assess whether the new evidence changes a material claim, and issue a clearly labeled correction or revision when needed. Record the affected figure, reason and approval rather than silently changing the original.
Use the export available in your workflow, then check the rendered document. Recheck figures, headings, tables, source links and permissions after formatting. Treat the final version as a separate release that needs review.
No. Include financial valuation only when it answers a relevant question and the evidence and assumptions support it. A clear account of outcomes and limitations can be more useful than an unsupported ratio.
Complete the draft, resolve its claim review table and save the approved version. If you still need a structure, return to the impact report template guide. For writing guidance, get the How to Write an Impact Report ebook. To compare presentation approaches, explore report examples.
Optional valuation reference: Decide when a financial value on impact is credible. Continue if valuation is relevant to your decision; do not add a monetary claim solely to make the report look stronger.
Need help with the reporting workflow? Bring one report and its evidence to a working session. Review what your team collects, how the records connect and where checking still takes manual effort.
Lesson revised September 12, 2026. The numerical narrative and source labels are fictional course examples.
Bring a report and its supporting evidence. Identify where collection, definitions and review can reduce repeated work.
Explore Impact Measurement →
