How to Write a Board Memo With AI That Still Sounds Like You

TL;DR: You can write a board memo with AI, and the first draft will read like it was written by anyone. That happens because "write it like a CEO" is a tone label, and a tone label is not a voice. Your voice lives in sentence length, qualifier placement, transitions and punctuation, and your board has learned all four from years of your memos. AI does not know your tone; it knows words that describe tone. The fix is to read your own habits off three past memos, write them as rules, and hand the rules to the model. A prompt describes how you hope to sound. A profile built from your own writing is evidence of how you actually write, and every new model executes that evidence better than the last one.

You paste the quarter's numbers into ChatGPT, ask for a board memo, and get one in eight seconds. Clean structure. Correct tone. Reads like anyone's.

Your chair has read forty of your memos. She knows you put the number in the first sentence and the caveat in the second. This draft does neither, and she will read that before she reads the number.

Can You Write a Board Memo With AI?

Yes, and the structural part is the easy part. Give any current model the agenda item, the numbers and the decision you need, and it will return a memo with a purpose line, context, options, a recommendation and a risk section, in the right order and at the right length. The structure of a board memo is well documented, and the model has read thousands.

The hard part is that the memo has to sound like the person whose name is on it. A board is the one audience that has read your writing for years, in a single register, at a predictable cadence. Directors do not consciously analyse your style. They have simply learned it, the way you learn a colleague's footsteps in a corridor. When a memo arrives with the same numbers and a different gait, the reaction is not "this was written by AI". The reaction is "something is off this quarter", and that is a worse thing for a board to feel.

So the question is not whether AI can draft the memo. The question is what you hand it so that the draft carries your gait.

Why Does an AI Board Memo Read Like Anyone's?

Because the instruction you gave was a description, and the model turned it into the average of everyone that description fits. "Professional, confident, appropriate for a board" describes thousands of people. The model does not know what you mean by confident. It knows which words appear near "confident" in the text it was trained on, and it produces the register those words point to. AI does not know your tone; it knows words that describe tone. We wrote about this in Adjectives Are Not Your Voice; the board memo is the case where the cost is highest, because the reader knows you best.

The register it produces is measurable. Our Average AI baseline, built from 320 samples across five models and four languages, puts English AI output at a formality score of 58 and a consistency score of 53. Formality in our scale weighs hedge frequency per thousand words, function-word density and semicolon use. Consistency measures how evenly sentence length is distributed. A score of 53 means moderate, uniform variation: no six-word verdict, no thirty-word sentence carrying three conditions. How we built that baseline has the method. The short version is that the model writes the middle of the distribution, on every request, for every executive.

Here is what the middle looks like on a board memo, and what a director who knows you notices in each row:

Every row is a placement or rhythm choice. None of them is a grammar error. A director cannot tell you which row bothered her, and she does not need to.

Why Doesn't "Write It Like a CEO" Fix It?

Because every common fix is a better description, and the problem is that descriptions carry almost no information about you. Three fixes get tried in roughly this order, and each one fails for the same reason.

A longer adjective list. "Direct, decisive, data-led, warm but not casual, no jargon." Each adjective narrows the register a little and describes you not at all. Two CEOs can both be direct and data-led and differ on every countable feature: one opens with the figure, the other with the decision; one hedges forecasts with "on current trajectory", the other with "assuming no change in churn"; one writes a five-word recommendation, the other a twenty-word one with two conditions attached. The adjective list selects a register and the model fills it with the average occupant.

One example memo pasted in. This is closer, and it fails in a subtler way. The model imitates the surface of the example on that request and reverts on the next one. Worse, one memo is one occasion. If the example was a good-news quarter, the model learns to open with the number even when the number is bad and you would have opened with the plan. Style in the moment is not a profile. A profile is what stays constant across eight memos of different weather, and a single sample cannot tell the model which parts are constant.

A newer model. The expectation is that GPT-5 or the next Claude will simply know how a CEO writes. It will know that better than the last one did, and it still will not know how you write. Model capability and user-data conditioning are separate layers. The lab improves the first with every release and cannot supply the second, because the second comes from your memos and the lab has never seen them. Given a vague adjective, a stronger model produces a smoother average. That is why better models still don't know your voice, and why the gap between generic and yours widens rather than closes as models improve.

The pattern across all three: a prompt describes how you hope to sound. What the model needs is evidence of how you actually write.

What Should You Hand the Model Instead?

Your habits, counted, written as rules the model can check its own output against. A stylometric profile built from your own writing is evidence of how you actually write, and evidence is what a model can execute. This is the same principle a linguist uses in authorship attribution: people are identified not by the words they choose to describe themselves but by function-word density, sentence-length distribution and punctuation habits they never think about. Your board has been doing informal authorship attribution on you for years. A profile makes it explicit.

For a board memo specifically, the countable habits that matter are a short list:

• Opening move. Number first, decision first, or context first. Most executives have one default and one exception for bad news.
• Where the ask sits. Sentence one, or after the recommendation, or in a separate line at the end.
• Hedge vocabulary and count. Which phrase you use to qualify a forecast, and how many times per forecast. "On current run-rate" and "we believe" are different hedges, and most people have exactly one.
• Recommendation length. The number of words in the sentence that says what you want the board to do. Five and twenty are both legitimate, and they are not interchangeable.
• Risk ownership. Whether you name a person against each risk, and whether the risk comes before or after the ask.
• Pronoun. "I recommend", "we recommend", or "management recommends". Boards notice a switch.
• Sentence rhythm. Your shortest sentence in a typical memo, your longest, and how often a short one follows a long one.
• Close. Decision and date, a question to the chair, or nothing after the ask.

Written out, a profile for one executive's board memos looks like this:

Board memo, quarterly update:
- Open with the headline figure and its delta to plan in sentence one. No purpose line.
- Bad quarter: open with the corrective action, figure in sentence two.
- One hedge per forecast, always "on current run-rate". Never "may potentially".
- Recommendation sentence under 10 words, first person: "I recommend we approve X."
- Every risk names an owner and a date. Risks come before the ask.
- Vary rhythm: at least one sentence under 8 words per section, none over 30.
- No semicolons. No "please find". Close with the decision required and the meeting date.

Every line can be checked against the output. Every line came from something you already sent to your board. That is the difference between a prompt and a profile: the prompt is a hope, the profile is a record.

How Do You Write a Board Memo With AI, Step by Step?

Spend twenty minutes once on the profile, then two minutes per memo. The objection every executive raises is that there is no time to feed an AI writing samples. The samples already exist, they are sitting in your sent folder, and three of them are enough.

Step 1: pull three past memos and two emails to your chair (5 minutes). Choose memos from different quarters, ideally one good and one bad, so the profile captures what stays constant when the news changes. Redact the figures and names if you are pasting anywhere outside approved tools; the profile measures how you write, and the numbers are irrelevant to it. How to select writing samples covers the general rules; for board material the specific rule is that a memo you rewrote four times with the CFO counts less than one you sent as written.

Step 2: count, do not describe (10 minutes). Go down the list in the previous section and write a number or a phrase next to each item. Not "I am direct" but "figure in sentence one in 3 of 3 memos". Not "I hedge carefully" but "one hedge per forecast, always 'on current run-rate'". You will be surprised by at least one of the counts. Most executives are surprised by how short their recommendation sentence is.

Step 3: write the counts as rules (5 minutes). Use the block above as a template. Then place it where the model reads it on every request: ChatGPT custom instructions or a Project, a Claude Project, a Gemini Gem. Placing it once is what makes it a profile rather than a prompt; the rules apply to next quarter's memo without you remembering to paste them.

Step 4: draft with the rules, then run the five-minute check. Give the model the agenda item, the numbers and the decision required. When the draft comes back, put it next to your most recent real memo and check four things:

1. Circle the first figure in each. Same sentence position?
2. Count the hedges on the main forecast in each. Same count, same phrase?
3. Count the words in the recommendation sentence in each. Within three words of each other?
4. Underline every sentence under eight words in each. Roughly the same number per section?

If the draft fails two of the four, the rules are not in place, or the model is not reading them. If it passes, your chair reads a memo with the same numbers and your gait, which is the only outcome that matters. ChatGPT for business leaders covers the wider set of documents where the same profile applies: investor updates, all-hands notes, the two-line email that moves a decision.

The same separation that makes a newer model useless with an adjective makes it more useful with a profile. "Risks before the ask, one hedge per forecast, recommendation under ten words" is a nuanced instruction. Each new generation follows nuanced instructions more faithfully, so each one executes your profile more accurately than the last. The model is the layer that improves on its own. Your profile is the layer only you can supply, and it is the one your board is listening for.

MyWritingTwin does the counting. A Writing DNA analysis measures your sentence rhythm, hedge placement, formality and punctuation from the memos you already have, against the 320-sample Average AI baseline, and turns the measurements into a Style Profile you carry to ChatGPT, Claude, Gemini or whatever your company has approved. The Executive tier builds separate sections for the board memo, the investor update and the internal note, because the same person writes those three differently and a single profile should not average them.

---

If you would rather see what a board-memo profile looks like before deciding anything, the executive profile page walks through one. The free Writing DNA Snapshot measures a few of your samples against the Average AI baseline and shows you, on one chart, where the model would have flattened you.