Start with the decision.
Then write the prompt.

“Write me a brand strategy” looks like an efficient starting point. It often asks the assistant to make too many decisions you have not made yourself.

Who is the buyer? What are they choosing between? What do you know about their behaviour? Where will the product be sold? Which economics constrain the offer? If those inputs are missing, the assistant can still produce a convincing document. The fluency of the answer does not tell you whether its foundations exist.

A better starting point is a decision that matters, stated narrowly enough that evidence could change your mind. Instead of asking for a finished strategy, identify the uncertainty standing between you and your next commitment.

1. Name the commitment.

A brand starts asking for resources before it has earned them: your evenings, a designer’s fee, stock, software, a lease or the cost of a first campaign. Record the next commitment you are considering. Then ask what you would need to know to make it responsibly.

For a physical product, the question might be whether the intended price leaves enough after the real costs of getting it to a buyer. For a service, the problem may be whether the working week can support the promise. For a venue, a cheap room may still be the wrong room if the occasions and catchment do not support it.

These are not interchangeable research briefs. Starting with the commitment changes what evidence matters and what the assistant should produce.

2. Separate a signal from a conclusion.

Search interest can help you understand a category’s direction. Reviews can show recurring complaints and valued qualities. Public conversations can reveal the words buyers use. Each is useful; none automatically proves demand for your particular offer.

Give every finding a source, date and status. Record whether it is something you observed, something a named source reported or something you are assuming. Then distinguish its evidence zone: opinion, reported evidence or behaviour. You can observe somebody saying they would buy without having observed them buy.

Ask your AI assistant to preserve that distinction. Require a gaps section. A closed source, a small sample or an unavailable tool should appear in the output as a limit. The assistant must not turn a missing fact into a plausible number because the template needs filling.

3. Read the alternatives as a buyer would.

Your business does not compete only with other businesses using the same category label. It competes with the alternatives a buyer can reach when a particular need or occasion arises. Begin with the five rivals closest to that real choice.

Read their offer as sold: the claim, price ladder, route to market and presentation. Record the praise as carefully as the complaints. A new proposition that fixes one irritation while removing the quality buyers love may not be an improvement.

Keep three lists: what you might beat, what you must match and the familiar claims or visual habits that make everyone look alike. Then test your own angle against those lists. The job is not to discover a difference at any price. It is to find a difference that matters to the person choosing.

4. Make the arithmetic explicit.

AI can help construct a model, but it cannot turn an unknown into a known input. Before asking it to calculate anything, identify the unit of value: an order, a subscriber month, a delivered engagement, an open day or a session.

Follow the money through the costs that belong to that route. A retail margin is not a direct-storefront payment fee. A booked hour is not the same as a deliverable hour once travel, cancellations and unpaid administration enter the week. A platform’s fee and rules can alter both the offer and its contribution.

Mark every input known, assumed or unknown. Run a low, base and high case with named levers. Keep customer-acquisition spending separate so that the contribution of a transaction and the cost of winning it remain visible. An illustrative model is useful when it reveals which fact you need next; it becomes dangerous when its finish makes you forget its assumptions.

5. Ask for work you can inspect.

A strong brief tells the assistant its role, objective, sources, output and quality bar. “Research the market” is open-ended. “Build a source-linked demand register, a five-rival comparison and a list of gaps that could change this decision” creates something you can interrogate.

Tell the assistant what to do when it cannot complete the work. Require it to say which source was inaccessible, which requested volume was not reached and which decision belongs to you. An assistant that knows when to stop is more useful than one instructed to sound complete.

Then ask it to argue against its recommendation. What evidence is weak? What alternative explanation fits the same findings? What would need to be true for the proposal to fail? This does not remove your review responsibility. It gives you a more useful object to review.

Keep the method ahead of the output.

The point of using AI in brand building is not to multiply documents. It is to make useful research, synthesis and creative exploration available to a founder who has to turn them into choices.

The order matters. Evidence before conviction. Insight before a proposition. Strategy before a visual system. A launch plan connected to what you can deliver, followed by a performance cycle that records what happened and changes the next decision.

You will still need real customers, credible sources, specialist judgement where the question requires it and the willingness to hear that something does not hold. A method is valuable because it keeps those requirements visible while the tools make producing an answer easier.