How to Write Reports with AI: Prompts + Worked Example

To write a report with AI, provide your source material, name the audience and decision, request a clear structure, and check the resulting claims against the originals. Then refine the recommendations and format the report for its reader.

This guide walks through a monthly marketing report from two text files to a six-page document. You can download the inputs, copy the prompt, and compare your output with the calculations below. The example answers a specific business question: should next month's advertising budget stay where it is, or should the team test a different allocation?

The screenshots show an actual generation in ML Clever, the product we build. ExampleCo and its results are fictional sample data. The writing and review steps can also be used in another AI tool that accepts your source files; the interface will differ.

The report-writing workflow at a glance

1. Define the audience and decision.
2. Attach the data and business context.
3. Use the report prompt and generate a draft.
4. Check the calculations and sources.
5. Review the executive summary and recommendation.
6. Edit the design and prepare the report for sharing.

You can also jump to the reusable prompt template, final review checklist, or FAQ.

Step 1: Define the audience and decision

Before choosing a chart or writing an introduction, decide what the report needs to help someone do. “Write a marketing report” leaves the AI to guess the scope. A brief with a reader, reporting period, and decision gives it a useful boundary.

Here is the brief for this example:

The budget constraint matters. The report can propose moving money between channels, but it cannot solve the problem by quietly increasing total spend. Similarly, naming Finance as a reader tells the writer to explain costs, uncertainty, and approval requirements.

For your own report, replace these fields with a real decision: whether to extend a client project, investigate a missed target, or approve a pilot.

Step 2: Attach the data and business context

This example uses two small files. Download both to follow along:

• Download marketing-metrics.txt — monthly spend, customer counts, and metric definitions.
• Download marketing-context.txt — the brief, campaign changes, budget limit, and evidence gaps.

The first file contains these six data rows. All amounts are USD, and the results are fictional.

The definitions explain that customers are attributed by last paid click and deduplicated across channels. Costs include advertising spend only; salaries, agency fees, creative production, and sales expenses are excluded. These details determine which calculations and labels are appropriate.

The context file adds information a spreadsheet cannot convey on its own:

• A new paid-search landing page launched on September 15, but no randomized test isolated its effect.
• Paid-social targeting changed on September 20; its effect is unknown.
• Revenue, margin, retention, and customer quality are unavailable.
• Any budget change needs approval and should begin as a limited, reversible test.

In ML Clever, open Docs mode, attach both files, and select a visual style. The screenshot uses Studio Blue with Auto pages selected; the prompt requests six pages.

If your material is in an existing PDF, see the separate guide to creating an AI report from a PDF. The same principle applies: provide the evidence and the context needed to interpret it.

Step 3: Paste the AI report-writing prompt

The prompt below is the one used for this example. It specifies the six-page structure and asks the AI to distinguish results, explanations, and proposals. Attach the two files before submitting it.

Create a polished six-page document titled “ExampleCo — September 2026 Marketing Performance Report” using the attached marketing-metrics.txt and marketing-context.txt files.

Write for the Marketing Director and Finance Lead. Help them decide whether to keep the current advertising allocation or test a reallocation within October’s fixed $20,000 budget.

Structure:
1. Executive summary: three key findings, headline metrics, and the recommended decision. Include “Fictional example — sample data.”
2. Performance overview: compare August and September spend, new customers, and blended acquisition cost, with a concise table and clear chart.
3. Channel performance: compare paid search, paid social, and retargeting. Show customer counts, acquisition costs, and percentage changes.
4. Interpretation: separate observed results from possible explanations and explain what the available evidence cannot establish.
5. Budget recommendation: present a practical, limited, reversible test or explain why the allocation should remain unchanged. Any proposed allocation must total $20,000. Explain tradeoffs and a proposed review checkpoint.
6. Next steps: include proposed responsible roles, decisions requiring approval, additional data needed, and source notes.

Calculate and check all figures against the attachments before writing. Calculate blended acquisition cost as total advertising spend divided by total new customers. Label acquisition costs “ad spend only.”

Do not claim the landing-page or targeting changes caused the results. Do not invent revenue, ROI, benchmarks, or future customer forecasts. Clearly label proposed actions and unconfirmed details.

Use only the attached files, with source references to their filenames and relevant row or note IDs. Include a source note on every page containing numerical findings.

Use the selected visual theme, generous spacing, readable charts, concise commentary, and clear headings. Create the complete report, not just an outline.

Because the structure is already defined, this run goes straight into document generation. For a report whose scope is still unclear, you can first ask for an outline and settle the questions it should answer before requesting a full draft.

The completion screen gives you a report preview to open and review. In this run, its first page highlights improved overall acquisition efficiency, stronger paid-search results, and weaker paid-social results.

Step 4: Check the calculations and source references

Treat the generated report as a draft until its main claims agree with the inputs. For this example, you can check the arithmetic with a calculator or spreadsheet using rows A1–A6 of marketing-metrics.txt.

Check the headline metrics

The blended calculation is $20,000 ÷ 110 = $181.82 per new customer for September. Calculate the percentage change before rounding the currency value.

Do not take a simple average of the three channel costs. Their September values are $150, $300, and $200; an unweighted average would give $216.67, which is the wrong blended figure because the channels acquired different numbers of customers.

Check the channel results

All costs in this table are ad spend only. Paid-search cost per customer fell 25%; paid-social cost rose 50%. Those are different measures from the changes in customer counts, which were +33.3% and −33.3% respectively.

The executive summary and channel chart visible in the screenshots agree with these calculations. When reviewing your own output, check the chart bars and labels as well as the prose. These inputs cover whole calendar months of different lengths, so the findings describe monthly totals, not daily acquisition rates.

Check what the sources actually support

A source note should let the reader find the relevant evidence. For the search-cost calculation, a useful reference is marketing-metrics.txt, A1 and A4; definition D4. A filename on its own gives less help.

For external research, also open the original source and check its date, scope, and support for the claim. Our guide to AI reports with citations covers that workflow in more detail.

Step 5: Review the finished report's summary and recommendation

Open the document in the editor and read its executive summary alongside the detailed findings. The page rail in this example shows six pages covering the summary, overall performance, channels, interpretation, budget proposal, and next steps.

Here is the executive summary shown on that page:

Fictional example — sample data. September spend remained flat at USD 20,000 while new customers rose from 100 to 110. Blended acquisition cost fell from USD 200.00 to USD 181.82, ad spend only.

The proposed decision is a limited, reversible October test: move USD 1,000 from paid social to paid search while keeping total advertising spend at USD 20,000. This is a proposal requiring approval, not an approved decision.

That proposed shift translates into the following budget. The October column represents the recommendation, not observed results.

The arithmetic fits the constraint. The business judgment still needs review: two months of results do not establish that $13,000 is the best search budget, or that search will retain its September efficiency after an increase. Keeping the original allocation while gathering more evidence is also an option.

Before approving a test, the team would need to agree on a review date, decision criteria, and responsibilities. Marketing Director and Finance Lead are suggested roles in the brief; named owners, thresholds, and approvals are not supplied.

A follow-up prompt for a clearer recommendation

Use this if your recommendation page leaves those decisions vague. It is a suggested revision prompt, separate from the generation shown above.

Revise the budget recommendation and next steps for the Finance Lead.

Keep total proposed October advertising spend at $20,000. Show the current and proposed allocation by channel, explain the tradeoff, and state that the change requires approval.

Separate what the data shows from what the test would investigate. Do not forecast new customers or claim improved ROI.

Add proposed responsible roles and a review checkpoint. Label any suggested timing or decision criteria as proposals to confirm. Preserve the source notes and the “ad spend only” metric definition.

Step 6: Edit the design and prepare the report for sharing

Once the findings and recommendation read clearly, refine the presentation. In the editor, use Themes to change the visual style and AI Edit when a section needs revision.

The screenshot below shows the channel-cost page with Arctic Shelf selected. Its chart makes the comparison visible: paid search moves from $200 to $150 per customer, paid social from $200 to $300, and retargeting stays at $200.

Choose a theme that keeps small labels and source notes readable for your audience. Review each page after changing the design, especially tables, chart legends, and footers.

Use Share or Export for delivery. PDF export access depends on your plan; see current pricing. If you export, open the resulting file and check every page for clipped text and unreadable labels before sending it.

Reusable AI report prompt for your own business

Replace the bracketed fields below to adapt the workflow to a client report, project update, research summary, or management report. For a more specialized research structure, see our market research report example.

Create a [length] report titled “[title]” using the attached sources.

Audience: [reader and role]
Reporting period: [dates]
Decision or question: [what this report should help resolve]
Constraints: [budget, scope, deadlines, or approval requirements]

Structure:
1. Executive summary answering the main question.
2. Key findings with supporting evidence and useful tables or charts.
3. Interpretation, alternative explanations, and limitations.
4. Recommendations with rationale and tradeoffs.
5. Proposed next steps and source notes.

Use only the supplied evidence unless I explicitly request external research. Calculate metrics using the definitions in the sources. Identify missing information rather than filling gaps with invented facts.

Separate observations, hypotheses, and recommendations. Reference the source file and relevant page, row, or note for important claims. Mark unconfirmed owners, dates, and decisions as proposed.

Use concise language and a readable layout. Produce the complete report, then identify any unresolved questions that need my review.

Final AI report review checklist

Before sending the report, confirm:

• The executive summary answers the stated question and agrees with the detailed analysis.
• Totals, percentages, and chart values match the original inputs.
• Dates, currency, attribution, and metric definitions are consistent.
• Source notes point to evidence that supports the specific claim.
• Proposed actions are distinguished from approved decisions.
• Missing information and alternative explanations remain visible.
• The report and any exported file are readable on every page.

For this example, the mistakes to catch are concrete: averaging channel costs incorrectly, treating lower acquisition cost as proof of higher profitability, or presenting the $1,000 shift as an approved budget. These checks are useful even when the document looks finished.

AI report-writing FAQ

What should I include in a prompt to write a report with AI?

Include the audience, reporting period, question or decision, source files, required sections, and constraints. Ask for source references and specify how to handle missing information. The example prompt above includes each of these fields.

Can I write a report with AI from notes or documents?

Yes. This walkthrough starts with two text files: one containing a small data table and another containing business context. You can use research notes, meeting notes, or existing documents as inputs, provided your chosen tool can read them. Ask it to flag gaps instead of inventing missing facts.

How do I check an AI-generated report for accuracy?

Recalculate its main metrics independently, compare charts with the underlying data, and trace important claims to their sources. Also check interpretation: a correct number can still be used to support a conclusion the evidence does not establish.

Can I use ChatGPT or another AI report generator for this workflow?

You can adapt the brief, prompts, and review checklist to another tool that accepts the source material. File handling, page design, editing, and export options will vary. See our comparison of AI report generators for business if you are choosing a tool.

What if I do not have enough data to write the report?

Ask for a structure and a list of missing evidence first. A qualitative report can use documented observations or interviews, but it should not acquire invented percentages or benchmarks to look more complete. In this example, revenue is missing, so the report cannot establish ROI or profitability.

Create a report from your own source files

Start with the two sample files above to try the workflow, or bring the sources for a report you already need to deliver. ML Clever AI Documents is the workspace shown in the screenshots.