What AI does really well
- Turning notes into a method. You describe how you build a sales proposal; it turns that into a clear, step-by-step process.
- Producing a first draft. A job description, a client charter, a meeting agenda: starting from a draft is always faster than a blank page.
- Running an interview. It asks the expert one question at a time and writes from the answers. Often the best way to get a busy expert to document what they know.
- Summarising. Reading scattered documents, a website, meeting notes, and pulling out what matters.
- Spotting inconsistencies. Two documents that contradict each other, a role with no owner, an offer described differently on different pages.
- Rewriting in the company's voice. Aligning documents written by different people.
What AI doesn't do for you
Understanding what's at stake
AI only knows what you give it. It doesn't sense tension in a team, doesn't know a client's history, doesn't see what goes unsaid. Its summary can be right in form and wrong in substance.
Deciding
Choosing a target market, dropping an offer, giving someone a responsibility: these are leadership decisions. AI can lay out options and their consequences; it shouldn't make the call. A good rule: AI suggests, a person approves or sets it aside.
Getting buy-in
A method perfectly drafted by AI, but not built with the team, won't be followed. Buy-in is still human work.
The risks to know about
- Everyone has their own context. If each person gives the AI their own version of the offers and methods, the documents it produces contradict each other. The most common risk.
- Generic documents. Without precise context, AI produces a textbook process that doesn't match your reality.
- Confidentiality. Pasting sensitive information into a consumer tool raises the question of what happens to it. Check the tool's terms, and keep sensitive data for business plans and settings.
- The illusion of being done. A document produced in five minutes feels finished. It isn't until it's been reviewed, approved and shared.
More in the risks of an unstructured business.
The method: structuring with AI in 5 moves
- Give it context. Your website, decks, existing documents. The more precise the context, the less generic the answer.
- Have it interview you rather than asking for a finished document: "Ask me the questions you need to write our invoicing process, one at a time."
- Review and decide. Fix what's wrong, choose between the options. This is where the real work happens.
- Get it approved by the people who practise the method, then by leadership.
- Store the approved version in a single place that the team and its AI use as the reference. Skip this, and everyone goes back to their own version.
Example prompts
For the strategy
"Here's our website and our brochure. Summarise on one page who our customers are, what we sell them and what sets us apart. Flag anything that seems vague or contradictory."
For a process
"Interview me to write our new-client onboarding process. One question at a time. At the end, write the process as steps, each with an owner and a timeframe."
For a job description
"From these notes on our account manager's role, draft a job description: mission, responsibilities, decisions she can make alone, indicators."
To check
"Compare these three documents and list where they contradict each other."
Which AI tool?
A general assistant (ChatGPT, Claude) is enough for drafting. The hard part comes after: keeping an approved version, sharing it with the team according to each person's access, and making sure everyone's AI draws on that same version. The comparison of solutions to structure your business reviews the options. For the overall approach, see how to structure your business.