Report Writing and Data Storytelling with AI

The best SQL query in the world is worth nothing if the report that wraps it is boring, buried, or misread. Writing is the part of analyst work that gets skimmed or skipped entirely, and stakeholders often judge an analysis by the email or deck they see — not the query behind it.

AI is exceptionally good at this part. It can transform your bullet points into a clean narrative, restructure a shaggy 2,000-word draft into a crisp exec summary, and produce the three-sentence Slack version alongside the long doc.

What You'll Learn

• A report structure that consistently gets read
• Prompts for generating exec summaries, narratives, and Slack-length updates
• How to turn raw findings into a story with a lead, a driver, and a recommendation
• Tailoring tone and length for different audiences without rewriting from scratch

The Report Structure That Works

Analysts who always write in this order produce reports that always get read:

1. Headline. One sentence. The one thing the reader should walk away with.
2. TL;DR. Three bullets: the finding, the driver, the recommendation.
3. Key numbers. A small table of the metrics that matter.
4. Context. Why this analysis, what question it answered, who asked.
5. Findings. 2–4 main findings, each with a chart.
6. Method. One paragraph on data source, date range, caveats.
7. Appendix. Supporting charts, alternative cuts, details on request.

The headline and TL;DR go first because 80% of readers will stop there. If they only read those four sentences, they should still walk away with the right conclusion.

Generating the Headline and TL;DR

Start every report by asking AI to write the headline from your findings, not from the data:

I just completed an analysis. My key findings are:

• March 2026 revenue was $4.2M, down 8% vs February
• The decline was driven entirely by the UK region, which dropped 31%
• UK decline tracks the May 2026 price increase rollout that happened there first
• All other regions grew in March

Write:

1. One-sentence headline — lead with the most important point, include the number
2. Three-bullet TL;DR with the finding, the driver, and a recommendation
3. A 30-word exec summary paragraph

The headline you get back will be sharper than anything you would have written from scratch.

Turning Bullets Into a Narrative

Analysts often draft reports as bullet lists. Stakeholders read narratives. Convert with:

Turn these analyst bullets into a 250-word narrative suitable for a monthly business review. Requirements:

• Start with the headline finding and the number
• Connect findings with transitions — do not just restate the bullets
• Include one concrete quote or example per finding
• End with the recommended next step
• Do not use phrases like "deep dive," "at the end of the day," or "circling back"
• Active voice only

Bullets: {paste}

Read the output critically. If a number changed or a nuance was lost, edit it back in.

The Three-Version Rule

Write every report once, then ask AI to produce three versions:

Here is my full analysis {paste}. Generate three versions:

1. Slack version (30–50 words) — for a quick note in a team channel. Lead with the number, skip methodology.

2. Email version (150 words) — for a VP who will read it on their phone. Include headline, key numbers, one driver, and a recommended next step. No charts.

3. Long version (800 words) — the full report with context, method, findings, caveats, and appendix pointers.

Keep the numbers and claims identical across all three. Only the length and emphasis change.

You will use all three throughout the week. Writing all three at once is much faster than writing each separately.

Tailoring by Audience

Different stakeholders read in different ways. Adjust with:

Rewrite this analysis for a CFO. The CFO cares about:

• Dollar impact (not percentages)
• Trend vs budget
• Cash implications (billing vs revenue vs collections)
• Risk and downside scenarios

Do not use technical terms like "cohort," "funnel," or "regression." Use "customer group," "conversion path," and "correlation."

Repeat with different audience profiles: CTO, CMO, head of support, board member.

Chart Captions That Do Work

A chart without a caption forces the reader to interpret it. Good captions tell the story.

For each chart below, write a 15-word caption that:

• Names the most important pattern
• Names one implication
• Avoids describing what the chart "is"

Example — instead of "Monthly revenue trend 2025–2026," write "UK decline pulls total revenue down despite growth everywhere else."

Charts: {list or upload}

Use the AI-generated captions as headings above each chart, not as boilerplate below.

Avoiding Common Writing Mistakes

AI will catch these if you ask:

Review this draft report and flag any of the following mistakes:

1. Burying the lead (the most important finding is not in the first two sentences)
2. Vague claims ("significantly," "notable," "material") without specific numbers
3. Passive voice that hides who did what
4. Unexplained jargon
5. Phrases that add no information ("it is worth noting that...")
6. Missing units (revenue without currency, time without timezone)
7. Claims beyond what the data actually shows (causation from correlation)

For each issue, quote the offending sentence and suggest a fix.

Do this on every report before you send it. You will see patterns in your own writing that you can learn to avoid.

The Recommendation Section

A report without a recommendation is just data. A good recommendation:

• Names the decision to be made
• Identifies the person who should make it
• Offers two options with pros and cons
• States what you would do

Generate with:

Based on the findings in this analysis, write a recommendation section. It should:

1. State the decision that needs to be made, in one sentence
2. Identify the decision-maker
3. Present two or three options with pros, cons, and estimated impact
4. State your recommended option and why
5. List the next three concrete actions and who owns each

Generating Appendices

Appendix material is often the last thing analysts write. AI can generate it from the main analysis:

Create an appendix section for this report with:

• Data sources and date range
• Methodology (1 paragraph per major calculation)
• Known caveats and limitations
• Alternative cuts of the data that did not make the main report but are worth preserving
• Queries used (I will paste the final SQL)

Version Control for Reports

Keep the narrative evolving. Use AI to produce a diff:

I have two versions of my report, one from last month and one from this month. Highlight what has changed: which numbers moved, which findings shifted, which recommendations are new. Organize as a changelog.

This makes updates easy — stakeholders see a concise "what's new" and can decide whether to re-read.

Key Takeaways

• Structure: headline, TL;DR, key numbers, context, findings, method, appendix
• Use AI to write the headline from your findings, not the data
• Produce Slack / email / long versions of every report simultaneously
• Tailor tone and jargon by audience (CFO, CTO, PM)
• AI-generated chart captions tell the story — use them as headings
• Always include a recommendation section with an owner and next steps