The Fastest Way for Executive Assistants to Prepare Professional Reports Using AI

The fastest way to a finished report isn’t to feed AI all your sources and ask for a summary. That produces a generically competent paragraph that fits no specific reader. The fastest way is to build the structure first, decide what each section needs to answer, and then ask the model to fill the structure from your sources. Reverse the order most people use, and the time-to-final drops by half, because the editing pass disappears.

This is how it works in three prompts.

The briefing structure that fits almost any exec report

Before you touch the sources, sketch the structure. For most exec briefings (status, initiative update, decision memo, quarterly summary), this shape works:

1. Bottom line, three sentences max. What happened, what it means, what’s needed from the reader.
2. Decisions or actions requested. A list of two to four items, each with a deadline. If there’s nothing, say so.
3. Key numbers. Three to six metrics, each with the prior period for context. No metric without comparison.
4. What’s working. One paragraph, two to three specifics with names and dates.
5. What’s not working. One paragraph, same standard, no euphemism.
6. Risks and watchlist. Anything that isn’t a problem yet but might be by the next report.
7. Appendix or sources. For anyone who wants to dig.

Save this as a template. Every report you build from now on starts here.

Prompt 1: extract the structure-shaped facts

Drop your raw sources (meeting notes, emails, dashboards, quarterly numbers, vendor updates) into Claude (which handles long documents well), ChatGPT, or Microsoft Copilot and run:

“I’m preparing an executive briefing on [TOPIC] for [EXEC]. Audience context: [INDUSTRY, EXEC LEVEL, WHAT THEY ALREADY KNOW]. Time horizon covered: [E.G. PAST QUARTER, PAST WEEK].

Below are my source documents. Read them and extract the following, with every claim sourced to a specific document and date:

1. Three to six measurable metrics that changed this period. Include the number this period, the number last period, and the source.

2. Up to five concrete events or decisions that occurred (e.g. ‘X signed contract on [DATE]’, ‘Y missed deadline on [DATE]’). Source each.

3. Up to three forward-looking risks mentioned by name in the sources (not your inference, theirs).

4. Any direct asks from the sources that need exec attention.

Don’t summarise yet. Just extract, with sources. If something isn’t in the sources, write ‘[NOT IN SOURCES]’ rather than inferring.

[PASTE OR ATTACH SOURCES]”

The “[NOT IN SOURCES]” instruction is the load-bearing one. Without it, the model fills gaps with plausible-sounding inferences that you then have to verify line by line. With it, you get a clean inventory of what the sources actually say, and you can decide what additional information the report needs.

Prompt 2: fill the structure

Now drop your template and the extraction together:

“Here’s my report structure:

[PASTE 7-SECTION TEMPLATE]

Here’s the extracted material from the sources:

[PASTE EXTRACTION OUTPUT]

Fill the structure. Constraints:

• Bottom line is three sentences, no more.
• Every metric in ‘Key numbers’ shows this period, prior period, and direction (up/down/flat).
• ‘What’s not working’ uses the same level of specificity as ‘What’s working’. Don’t soften.
• ‘Decisions requested’ must include a deadline for each.
• Total length under 600 words excluding the appendix.
• Tone: confident, no hedging, no ‘we are pleased to report’, no ‘continued progress’.
• If a section has no material from the extraction, write ‘Nothing material this period’ rather than padding.”

What you get back is 80% of a finished briefing. The remaining 20% is your judgement: rewording the bottom line so it sounds like your exec, cutting one of the metrics that doesn’t actually matter, adding the political context the sources don’t contain.

Prompt 3: the critique loop

Before sending, run the draft through one more pass:

“Here’s the draft briefing. Critique it as if you were [EXEC]‘s harshest peer. Answer:

1. Does the bottom line actually say what happened, or does it use words like ‘progress’ and ‘engagement’ that don’t carry information?

2. Is there any metric in ‘Key numbers’ that’s there for show rather than because it changed something? Recommend cutting.

3. Are ‘What’s working’ and ‘What’s not working’ at the same level of detail? If ‘not working’ is vaguer, point it out.

4. Is there any decision request that’s missing a deadline or owner?

5. Anything in here that, if read by a sceptical board member, would prompt a follow-up question I should pre-empt?

Be specific. Don’t tell me ‘consider revising the introduction’; tell me which sentence and why.”

This prompt catches the issues your exec would catch in their second read. Fixing them at draft stage means the report goes out in one round instead of three.

A worked example, briefly

Take a quarterly update on a vendor onboarding programme. Default-prompt output: “The vendor onboarding programme made significant progress this quarter, with several key milestones reached and ongoing efforts to address implementation challenges.” Pure padding.

Run the three prompts and you get:

“Bottom line: 14 of 18 vendors live, two ahead of plan, two delayed by legal redline. Onboarding cost-per-vendor down 22% from Q2 ($8,400 to $6,560). Need exec sign-off on the revised legal SOP by Friday or both delayed vendors slip to Q1.

Decisions requested: Approve revised legal SOP (deadline Friday). Confirm Q1 cohort scope (deadline next Wed)…”

That’s the version your exec reads.

The failure mode that gets reports rewritten

Confidently-phrased fabrications in the metrics section. Models will sometimes confidently produce a number that doesn’t appear in the sources, especially comparison numbers (“up from $5.4m last quarter”) when the sources only stated this quarter. The “[NOT IN SOURCES]” instruction in prompt 1 catches most of this. The critique prompt catches the rest. But you should also do a 60-second pass yourself: open each metric, verify the number against the source. If you can’t find it, cut it. A briefing with three real numbers beats one with six numbers, two of which are wrong.

A counterintuitive observation

The cleanest exec reports use fewer metrics, not more. A briefing with twelve numbers is harder to act on than one with three numbers and a clear story. The temptation when AI makes generation cheap is to include everything you have. Resist it. The model can fill the structure; you decide what makes the cut. That’s the editorial judgement that makes the report worth the exec’s time.

What stays in your hands

Picking which decisions to lead with. Knowing which “not working” item is politically sensitive and needs softer phrasing in writing but a private follow-up in person. Reports built this way go from a 90-minute slog to a 25-minute task, and most of those 25 minutes are now spent on judgement calls, which is the part of report writing that always mattered.