A founder at a twelve-person HR consultancy types into ChatGPT: “Help me decide whether to hire another consultant.” The model returns a list of general considerations. She reads it, closes the tab, and carries on without making the call.
She gave the model a question with no brief. What came back matched.
What is a thinking-partner prompt?
A thinking-partner prompt is a structured sequence, not a single open question. You give the AI your context first, your role, your business model, and your audience, then a goal, then the constraints you are working within, then the format you need for the output, then you ask it to challenge its own answer before you refine. That sequence is what separates a useful deliberation session from a generic text dump.
The six steps come out of practical playbooks developed for small businesses, including HaloTech Lab, a UK-focused AI consultancy, and from Microsoft’s guidance for Copilot users in Office tools. They map directly to the way you would brief any skilled contractor.
- Context prompt: describe your role, your business model, and your audience.
- Goal prompt: state the decision or outcome you are working towards.
- Constraints prompt: name the limits you are working inside, budget, regulation, timeline, market.
- Structure prompt: specify the format you need back, a pros/cons table, three scenarios, a numbered list of risks.
- Critique prompt: ask the model to list five reasons its answer might be wrong or incomplete.
- Refinement: give your reaction to the answer and ask for a revision.
The Scottish AI Playbook, produced with support from the Data Lab and the Scottish Government, recommends treating AI tools as “a colleague whose work always needs review”. A thinking-partner prompt is what gives that colleague a full brief to work from.
Why does prompt structure change the quality of what comes back?
Research from MIT Sloan Management Review tested 444 professionals using ChatGPT with and without a structured prompting approach. Those given the scaffold completed complex tasks 25% faster and produced outputs rated 40% higher quality by independent evaluators. For a services firm founder spending thirty minutes a day with AI tools, a 25% speed gain returns meaningful time each week.
The mechanism is straightforward. AI language models generate the most statistically plausible continuation of whatever you give them. A vague question leaves an enormous space of possible answers open. A structured brief narrows that space toward something useful.
The UK Department for Science, Innovation and Technology’s 2024 study on AI adoption found that 68% of businesses adopting AI use it for information search, synthesis, and decision support, not just content generation. A Sage survey of over 2,000 UK SMEs found that 51% use generative AI to support decision-making or planning. The gap between founders who get real value from this and those who do not almost always comes down to how specifically they brief the tool.
Unstructured use produces a subtler problem too. When the output is vague, founders tend to read it as the model being unhelpful. The prompt is where the value is set.
Where will you actually use these prompts?
The six-step sequence earns its keep wherever a founder-level decision is currently sitting unresolved. A pricing question for a new service line, a choice between two candidates with different profiles, a difficult client conversation to rehearse before it happens, a quarterly planning session where you want a structured challenge to your own assumptions. Those are the right situations.
These are not rare moments. They are the daily rhythm of running a services business. What changes with the prompt structure is that each one becomes a disciplined session rather than an open-ended chat.
The World Economic Forum’s 2023 briefing on generative AI for SMEs notes that building a library of reusable, tested prompt templates and sharing it with a small team reduces inconsistent outputs and lets non-technical staff adopt the approach consistently. For a founder who has developed a reliable context prompt for their business model, reusing it across hiring, pricing, and client decisions saves the setup time each time.
The value shows up quickly. A founder using a critique prompt for the first time typically gets back three or four objections they had not named themselves. That is the tool working as a thinking partner: surfacing the blind spots in a decision before it is made.
When does the six-step approach reach its limits?
Two limits need naming before you build this into your working week. The first is data: pasting client personal information, financial detail, or confidential commercial material into a public chatbot creates a data-processing question the ICO’s generative AI guidance addresses directly. The second is accountability: in regulated sectors, a thinking-partner prompt gives you decision support, not a decision. Accountability stays with the senior manager.
The National Cyber Security Centre’s guidance on the secure use of generative AI is direct on the data point. Assume that anything submitted to a public AI system could be stored and used to train the model, unless enterprise controls are in place. The practical test both the NCSC and the Scottish AI Playbook recommend is the same as for email: if you would not send it to an external supplier, do not paste it into a public AI tool. The solution is to brief at the level of abstraction the model needs, role descriptions and sector context rather than real names, real figures, or real client details.
Research from the University of Cambridge and Google DeepMind on language model reliability found that asking a model to show its step-by-step reasoning reduces hallucination rates by 20 to 30% compared with direct-answer prompts. That is why the critique step is not optional. It is the mechanism that turns a plausible answer into one you can act on.
For regulated businesses there is a third constraint. The Competition and Markets Authority’s 2023 review of foundation models flagged that using AI to analyse pricing could raise competition law concerns if outputs effectively coordinate market behaviour with peers. The FCA’s Senior Managers and Certification Regime makes the parallel point in financial services: the AI provides analysis; the senior manager carries the decision.
What to read alongside this
Three ideas make the thinking-partner prompt more reliable over time. A written AI policy sets the rules on what data goes into any tool, which the ICO advises and the Scottish AI Playbook describes as a senior-management responsibility. A shared prompt library turns your tested templates into a team resource. And building the critique step into every session turns AI from a text generator into a genuine check on your thinking.
The ICO’s 2024 AI and data protection risk toolkit explicitly warns against automation bias: the tendency for staff to accept AI outputs without question. Building a challenge prompt into every thinking session is the operational response to that warning, and the habit that keeps the tool in the analyst role rather than the decision-maker role.
If you want to build the library that makes these prompts reusable across a team, the companion post on standing prompts and the personal prompt library sits directly alongside this. For decisions with higher stakes, the post on using AI as a sparring partner for hard decisions covers the deliberation discipline in more depth.
If using AI as a genuine thinking partner is a direction you want to take more seriously in your business, a conversation with Dave is the best starting point. Book a conversation.
