A Q3 strategy doc written in Markdown beats one trapped in a slide template
Strategy is an argument, not a bullet list. Write the reasoning first as one Markdown file, then render it as a doc for readers and a deck for the room.
Every quarter the same thing happens. Someone opens a slide template called Q3_Strategy_FINAL_v4.pptx, makes a title slide, and then spends two days fighting 16:9 boxes instead of thinking. The bets get compressed into three-word bullets. The tradeoffs vanish because there was no room for the sentence that said "we are choosing X at the cost ofY." By the time it's presented, the document is a performance, not an argument.
Strategy work is reasoning. Reasoning wants prose. It wants a claim, the thing that would prove it wrong, and one canonical place to point people. The slide template gives you none of that — it gives you a grid and a deadline. So the move is to invert the order: write the argument first as a single Markdown file, get it right, and only then render it into whatever surface the moment needs. This is the workflow I use for quarterly reviews, end to end, with the actual prompt scaffolding and the paste-ready skeleton.
Why prose first, format second
A strategy doc has a job no slide can do: it has to survive being read by someone who wasn't in the room. A deck is a prop for a live narrator. Strip the narrator away and the bullets are riddles. So the source of truth should be the readable version — the long-form argument — and the deck should be a view of it, generated for the 30-minute live review. Same content, two surfaces, one link. That's the whole idea behind the document is a link: you ship a URL that anyone can open, not a file that lives in seven inboxes at four different versions.
Markdown is the right source format for this because it is the format AI writes most reliably, it diffs cleanly when you revise next quarter (see diff-able documents), and it carries no layout opinions to distract you while the thinking is still wet. You write headings and paragraphs. The format comes later.
The skeleton: six sections that force honest strategy
Most quarterly docs are mush because they have no spine. Here is the skeleton I make every draft conform to. The sections are chosen to make it hard to hide: you cannot list a bet without stating what would falsify it, and you cannot claim progress without an anti-bet — the things you are deliberately not doing. Paste this and fill it in:
# Q3 Strategy — [Team / Product] ## Context Where we are right now, in three sentences. The single number that defines the quarter. What a reader needs to know before the rest makes sense. ## What changed What is different since last quarter's doc. Bets that paid off, bets that didn't, and the assumptions we now know were wrong. Be specific: "Activation rose from 31% to 44% after onboarding rework" — not "onboarding improved." ## Bets The 2–4 things we are spending the quarter on. Each bet: - **Bet:** one sentence, an action not a theme. - **Why now:** the reason this beats the alternatives. - **Falsifiable by:** the metric + threshold that tells us by [date] whether this was wrong. ## Anti-bets What we are explicitly NOT doing this quarter, and why. This section is the tell for whether a strategy is real. ## Metrics The 3–5 numbers we will be judged on. Current value → target → the date we check. No vanity metrics. ## Asks What we need from other teams / leadership to make the bets possible. Specific, owned, dated.
If a draft can't fill the Falsifiable by line, that's not a bet — it's a hope. If the Anti-bets section is empty, the strategy has no focus, it has a wish list. The skeleton does most of the enforcement for you.
The AI drafting loop
The mistake people make is asking an AI to "write our Q3 strategy." It will, and it will be generic sludge, because it has no idea what actually happened to you last quarter. The trick is to feed it the two artifacts that contain the truth: last quarter's doc and this quarter's numbers. Then it's not inventing — it's synthesizing.
Step one: drop last quarter's strategy doc and a metrics CSV into the chat. The CSV is just an export from your dashboard or spreadsheet — something like:
metric,q2_start,q2_end,q3_target mrr_usd,182000,214000,260000 activation_rate,0.31,0.44,0.55 weekly_active,4100,5200,7000 logo_churn,0.038,0.041,0.030 nrr,1.06,1.09,1.15
Step two: give it the skeleton above and a prompt that constrains it to the evidence. This is the scaffolding that matters:
You are drafting a Q3 strategy doc. I'm giving you two inputs: 1. Last quarter's strategy doc (below). 2. A CSV of this quarter's metrics vs. last quarter. Fill in this exact skeleton: Context / What changed / Bets / Anti-bets / Metrics / Asks. Hard rules: - Use ONLY numbers present in the CSV. If you want to claim a trend, cite the columns. Never invent a figure. - Every bet must have a "Falsifiable by: <metric> <threshold> by <date>" line. If you can't write one from the data, flag the bet as [UNVERIFIABLE] instead of faking it. - "What changed" must reconcile last quarter's bets against the actual numbers — name which bets paid off and which didn't. - No hedge words: "explore", "consider", "aim to", "look into", "leverage". Write commitments, not vibes. - Anti-bets: propose 2–3 things we should explicitly drop, based on what the numbers say isn't working. Output Markdown only. --- LAST QUARTER'S DOC --- [paste] --- METRICS CSV --- [paste]
What comes back is a real first draft — one that already reconciles your old bets against what happened, and flags the places it can't support a claim. That's the 70% an AI is genuinely good at: structure, reconciliation, and refusing to let you skip the falsifiability line.
The human edit pass that actually matters
The remaining 30% is the whole game, and it is not "polish the wording." It is two specific operations.
Kill the hedge words.Even when you tell it not to, an AI smooths toward safety: "we aim to improve activation," "we'll look into expansion revenue." Every one of those is a sentence that can never be wrong, which makes it useless. Rewrite each into a commitment with a number. "We aim to improve activation" becomes "We will lift activation from 44% to 55% by shipping the guided-setup flow in week 3." A strategy doc you cannot be held to is a diary.
Make the bets falsifiable, and restore the tradeoffs. This is the thing AI is worst at, and I'll be honest about it below. Go through each bet and ask: what observation, by what date, would make me admit this was the wrong call? Write that line. Then go to the Anti-bets and make sure each one names what it costs you. Real strategy hurts a little on the page — if every section reads like good news, you've edited the conflict out.
What AI gets wrong in strategy docs — and how to catch it
Two failure modes, every time, and they're worth naming because they're subtle.
It invents confidence.Models are trained to sound assured, so a shaky bet and a solid one come out in the same crisp voice. The draft will state "expansion revenue is our highest-leverage bet" with the exact tone it uses for things you actually have evidence for. Catch it by checking every assertive sentence against the CSV: if there's no number behind the confidence, downgrade it to a hypothesis or cut it. The[UNVERIFIABLE] flag in the prompt helps, but you still have to read for the confidence that snuck through unflagged.
It smooths over tradeoffs.An AI's instinct is to make everything fit together nicely. Real strategy is a set of painful exclusions — doing the email bet meansnot doing the mobile bet this quarter. The model will quietly imply you can do both. The fix is the Anti-bets section: if the draft won't commit to dropping anything, that's the tell, and you write the exclusions yourself. The most valuable sentence in most strategy docs is the one that starts "We are deliberately not…," and it's the one an AI will never volunteer.
One source, two surfaces
Now the payoff for writing prose first. You have one Markdown file holding the full argument. The async readers — the people who weren't in the review, the exec who reads it on a Sunday — get the scroll-doc: the long-form version where the Context paragraph and the Falsifiable-by lines are all there to read in full. Nothing was compressed away.
For the live 30-minute review, the same Markdown renders as a deck: one section per slide, the bet as the headline, the falsifiable line as the supporting point, the metrics as a chart. You didn't rebuild anything. You changed the render, not the content. When someone in the room asks "wait, what's the actual target on churn," you don't flip through slides — you send the doc link and they read the paragraph. This is the practical meaning of treating HTML as the artifact: the deck and the doc are both just views of the source, and the link is what you hand around.
A rough heading map, so you can see how one file becomes two surfaces:
Markdown heading → Doc surface → Deck surface ----------------------------------------------------------- # Q3 Strategy → title block → cover slide ## Context → intro section → 1 slide ## What changed → prose w/ numbers → 1 slide + chart ## Bets → each bet expanded → 1 slide per bet ## Anti-bets → list w/ reasons → 1 slide ## Metrics → table → dashboard slide ## Asks → owned list → closing slide
The deck for the room and the doc for everyone else never drift, because they are the same file. Next quarter you open that file, diff it, and write Q4 against what Q3 actually claimed — which is the only way a strategy doc compounds instead of resetting to a blank template every three months.
The shape of the workflow
Pulling it together: export last quarter's doc and a metrics CSV, hand both to the AI with the six-section skeleton and the constraint prompt, and get a draft that reconciles bets against reality. Then do the two edits that matter — kill the hedge words, make every bet falsifiable and every anti-bet honest about its cost. Read once more for invented confidence and smoothed-over tradeoffs, the two things the model will always get wrong. Then render the one Markdown file as a scroll-doc for the readers and a deck for the room.
The point isn't that AI writes your strategy. It can't — it has no skin in your quarter. The point is that AI gets you a structured, evidence-anchored first draft in minutes instead of days, so your time goes to the part only you can do: deciding what you're willing to be wrong about, and what you're willing to give up. Write that argument in plain text, and let the slide template be the last thing you think about instead of the first.