Strategy document prompt — why AI-written plans read like consulting homework

What makes this hard

Ask for a strategy document and this often comes back: SWOT, 3C, PESTLE in the table of contents; three strategic options with pros-and-cons tables; and a conclusion of 'choose depending on circumstances.' The form is all there, yet after reading you don't know what it's proposing to do.

Two causes. Use frameworks as the outline and filling the framework becomes the goal, so the conclusion disappears. Frameworks are thinking tools, not a document's skeleton. And don't demand a decision and AI takes the safest path: listing. The decision gets shipped, whole, to the reader.

What to add to the request

1. Use the frameworks, delete them from headings

Specifying frameworks is good, since 'analyze' without one just makes longer lists. But add: don't put framework names in headings; make each heading the question that section answers. Not 'VRIO analysis' but 'what we have and what we lack.'

2. Demand a decision with its basis

Don't stop at options. Commit to one and justify it by impact, risk, and difficulty. In the execution plan's responsibility table, allow exactly one final approver per row. With two approvers, nobody approves.

3. Don't let solutions stop at a name

'Build an integrated dashboard' plus one line is something no recipient can build. Demand five things: what is being built · what the user sees and does · what goes in and what comes out · which step of the current workflow it replaces · what done looks like. With those five, a name becomes a plan.

Seven stages — and the question each one answers

The strategy instruction has the document written in seven stages. Frameworks are used inside a stage but never surface as headings, and tables are allowed only in stages 1 and 7.

1. 1Task analysis — background and purpose, the value chain from input to impact, key metrics, resources, risks. The first place a table is allowed
2. 2Project brief — problem definition, the friction users actually experience, root cause, goals and the metric system, what's in and out of scope, constraints
3. 3Environment — what we have and what we lack, the forces coming from outside, and three to five implications that tie the two together
4. 4Options and the decision — three to five options, then one is chosen and justified by impact, risk, and difficulty. The chosen strategy's initiatives and roadmap follow
5. 5Critical success factors — from eight to twelve candidates, the top one to three are chosen with reasons
6. 6Solutions — from ideas per factor, the contradiction is stated first and only then are inventive principles applied, ending in one to three committed solutions. Principles listed without a contradiction aren't TRIZ
7. 7Execution plan — roles, a RACI table with exactly one final approver per row, cost, deadline, and risk grades, change management. The second place a table is allowed

Metrics split into one and two

Performance metrics can't end as qualitative phrases. The instruction requires one north-star metric plus the leading indicators that move ahead of it and the lagging ones that confirm it afterward. 'Improve customer satisfaction' isn't a metric; 'monthly repeat-purchase rate' is.

The whole document also stays out of the imperative voice. 'Do this' and 'confirm that' are what we say to the AI, not the language of a plan. Anything that still needs confirming is written as a fact: 'the approval rules could not be confirmed; once they are, the sign-off steps change'.

Checklist for writing it yourself

1. 1Did you specify frameworks and also say 'not as headings'
2. 2Did you demand a committed choice with basis, not a list of options
3. 3Did you demand the five elements per solution
4. 4Did you make it mark unknown figures, org names, budgets for verification instead of inventing them
5. 5Did you keep undecided items out of the document's opening? A document that opens with an undecided list reads as a confirmation request, not a plan