How to Write a Board Deck Prompt That Locks the Narrative Structure

Most board updates get rewritten from scratch every quarter. Someone opens last quarter's deck, deletes the old numbers, pastes new ones, and hopes the story still hangs together. By the third revision the structure has drifted, the lowlights are buried, and the board's one question ("what are you actually asking us for?") doesn't have a clear answer anywhere. A reusable board deck prompt fixes the drift by locking the narrative before anyone touches a slide.

The job here isn't the slides. It's the words. Turn a metrics table plus a few lines of management commentary into a board-meeting narrative with a fixed shape: what went well, what didn't, and the specific decisions you need from the room. Get that text right and the slides take ten minutes. Get it wrong and you spend the meeting explaining the deck instead of running it.

Search results for board decks are crowded with slide-template tools. The gap is the narrative itself, and that's the part a prompt can actually own.

Why deck-template galleries don't solve the real problem

Search "create a board deck with AI" and you land on tool comparisons and template galleries. Gamma turns a prompt into a layout. Beautiful.ai and Presentations.AI ship board-meeting templates with placeholder slides. These help with design. They don't write the story.

A few guides go further. The 2Slides walkthrough (2slides.com) does ship a copyable prompt with bracketed data fields and even a [MISSING DATA] flag, which is more than most. But it generates a generic 17-slide deck. It doesn't force a board-pack narrative with per-section commentary, and it doesn't separate highlights from lowlights as a hard rule. Section AI's piece (sectionai.com) is a thoughtful methodology comparison across four models, but it openly lacks a reusable template that converts metrics into a structured highlights/lowlights/asks narrative.

So the winnable gap is clear. No widely-ranking page gives you a prompt that enforces a locked board-pack structure and refuses to invent the parts your data doesn't cover. That's the post.

What you can do with this prompt

• Turn a quarterly metrics table plus three lines of commentary into a full board narrative
• Force a consistent section order so the deck reads the same every meeting
• Separate genuine highlights from lowlights instead of softening both into "updates"
• Surface the asks as an explicit list the board can vote on, not a buried paragraph
• Flag every metric the data doesn't support with a Not stated marker
• Draft the CEO summary last, after the sections exist, so it actually summarizes

Anatomy of the board deck prompt

The structure that survives contact with a real board has three moving parts: the inputs you paste, the locked output contract, and one hard rule about missing data.

Variables
  {{reporting_period}}   → e.g. "Q2 FY2026"
  {{metrics_table}}      → pasted table: metric, plan, actual, prior period
  {{management_commentary}} → 3-6 bullet lines on what drove the numbers
  {{board_asks}}         → decisions or approvals you need this meeting

Prompt
  Role: board reporting writer for the leadership team
  Task: write the board narrative in the LOCKED section order below
  Rule: if a driver or figure is not in the inputs, write "Not stated" —
        never infer, never estimate

Output contract (locked order)
  1. CEO Summary (write this LAST)
  2. Highlights — metrics that beat plan, with the stated driver
  3. Lowlights — metrics that missed plan, with the stated driver
  4. Asks — numbered decisions the board needs to make
  5. Open Data Gaps — every "Not stated" item, listed for follow-up

The locked order is the whole point. Models left to their own devices will reorder sections to flatter the numbers, and a soft "include highlights and risks" instruction lets them blur the two. A numbered output contract holds.

Why the missing-data guard matters more than it looks

Here's the opinionated part: the single most valuable line in a board prompt isn't the structure, it's the Not stated guard. Without it, the model fills gaps with confident-sounding narrative that nobody fact-checked. That's how a board deck ends up asserting a churn driver the team never actually diagnosed. Tell the model to write Not stated and route every gap into a closing "Open Data Gaps" section, and the deck stops lying by omission. The gaps become a to-do list instead of fiction.

Step-by-step usage

1. Gather inputs

Pull the metrics table you already maintain. Plan, actual, prior period, in whatever rows your board expects. Don't reformat it; the model reads a plain pasted table fine.

2. Fill variables

Drop the table into {{metrics_table}}. Write three to six honest lines in {{management_commentary}} covering what moved the numbers. List the real decisions in {{board_asks}}. If you don't have a clean driver for a miss, leave it out. The guard handles it.

3. Run the prompt

Paste the whole thing into ChatGPT or Claude. Claude tends to honor a ## Output format heading and a numbered contract more literally on the first pass. GPT-4o sometimes needs the locked order restated on the final line of the prompt to stop it from collapsing highlights and lowlights into one section.

4. Post-process

Read the Open Data Gaps section first. Every Not stated line is either a number you forgot to paste or a story you haven't worked out yet. Both are worth knowing before the board does.

5. Iterate

If the highlights read thin, the commentary was thin. Add a driver line and rerun. The prompt is deterministic enough that the same inputs give the same shape, which is the entire reason to reuse it.

Prompt-craft patterns that make it hold

Role first, contract last. A two-line role ("You write board narratives for the leadership team") plus the numbered output contract at the very end beats a long persona. Models weight the most recent tokens, so the structure you want belongs at the bottom of the prompt, after the pasted table.

You write board narratives for a company's leadership team.
[ ...metrics table and commentary here... ]

Now produce the narrative in EXACTLY this order:
1. CEO Summary  2. Highlights  3. Lowlights  4. Asks  5. Open Data Gaps
For any figure not in the data above, write "Not stated".

Separate the two halves of the truth. "Highlights" and "Lowlights" as distinct, mandatory sections force the model to commit. A combined "key updates" section lets a 30% miss hide inside a paragraph that opens with a win.

The refusal boundary on numbers. Add one line: "Do not calculate variance percentages; report only the figures given." Models will happily do arithmetic, and a wrong variance in a board deck is worse than no variance. If you want the percentages, compute them in the sheet and paste them.

Variables you'll set

Getting started

1. Copy your current quarter's metrics into a plain table.
2. Write the honest commentary, gaps and all.
3. List the asks as actual decisions, not status updates.
4. Paste everything into the prompt with the locked contract last.
5. Run it in Claude or ChatGPT, then read the Open Data Gaps section first.
6. Fix the gaps, rerun, and move the narrative into your slide tool.
7. For a packaged version with the section structure already enforced, use the Board Pack Production Prompt Pack.

The narrative work is the same whether you're reporting to a startup board or a finance committee. If you also assemble the QBR for customers or leadership, the QBR Production Prompt Pack reuses the same highlights-and-asks discipline for that audience.

Treat the deck as a narrative problem before a design problem and the rest falls into place. If you're fundraising rather than reporting, the Fundraising Narrative Playbook handles the investor story, which is a different shape than a quarterly update. And if your board work bleeds into FP&A reporting, the variance write-up is its own job worth locking down. See the variance analysis prompt for the budget-vs-actual narrative, and the OKR drafting prompt for turning the same metrics into next-quarter targets.