Generative AI for the reading and writing your team does every day
Generative AI is very good at a first draft and a summary. It's not good at knowing your business. I set it up so it gets the first part right and a person handles the second.
Where generative AI earns its keep
Most teams spend hours a week on reading and writing that follows a pattern:
- Summaries. Turning a call transcript into notes, decisions and to-dos.
- Drafts. The follow-up email, the weekly update, the first version of a report.
- Sorting. Tagging feedback as a feature request, a bug, an onboarding problem or a pricing signal.
- Matching. Working out which person or client an incoming email is about.
The pipeline I built for a software startup does all four, three times a day.
Where it goes wrong
Generative AI writes confidently whether it's right or not. Left alone, it will invent a detail, promise a date, or send a cheerful email to someone who just complained.
The fix isn't a better prompt. It's a better process:
- Give it the judgement calls in writing. "If the numbers dropped, say so in the first line." "Never promise a date."
- Show it good examples. Two or three pieces of finished work you were happy with.
- Keep a person on anything public. The startup's AI drafts every email and sends none. The founder reads and sends each one.
Outreach that sounds like a person
One part of the startup work was prompts that draft one personalised email per person on the waitlist, for the founder to review. Her verdict:
"The prompt worked beautifully. Thank you!"
Which model
ChatGPT, Claude, Gemini, Copilot, or what you already pay for. The differences between them matter much less than how clearly the work is defined.
How I work
Define the process with your team, build it on your tools, test it on one real project, then train your team to run and change it.
Common questions
What does a generative AI consultant do?
Sets up AI to do a team's reading and writing work, like drafts, summaries and notes, inside a defined process with a person checking the output.
Which generative AI model is best?
The one that fits the job and that you already pay for. A clearly defined process works on any of the major models.
How do you stop generative AI making things up?
Write the judgement calls down, give it good examples, and have a person check anything that leaves the company.
Can generative AI write in our voice?
Yes, if you give it real examples of your writing and clear rules. A person should still review anything that goes to a customer.