This is the single biggest use of AI at work, and nobody talks about it because it feels too ordinary. Here is how to do it properly, and where it stops being a five minute job.
Not agents. Not automation. The most common thing people actually do with AI at work is paste in a rough, blunt, or half-finished thought and ask for a version they can send. A fast email that came out too sharp. A cover letter. An awkward text to a client about a late invoice. Nobody wants to write these from scratch, and nobody needs to any more.
A paid ChatGPT or Claude subscription and a person checking the output before it goes out. That is genuinely the whole answer for most small business writing. The skill is not the tool, it is what you ask it to do. A few habits make a real difference:
Give it the outcome, not just the topic. "Write a polite chase-up email for an invoice 14 days overdue, firm but not threatening" gets a far better result than "write me an email about an unpaid invoice."
Keep a few good prompts written down. If two staff members are both writing customer replies, having one shared prompt for tone means the business sounds consistent, not like two different people improvising.
Always read it before it goes out. This is not optional. A general AI tool will confidently write something factually wrong or oddly worded if you do not check it. The person stays in the loop, every time.
The trouble starts when the volume gets big enough that nobody is reading every output any more, or when the writing has to sound exactly the same across a whole team without someone rewriting each one. A few examples:
A brand voice that keeps drifting. If five people are all pasting into a general chatbot, you get five slightly different tones, and customers notice. A written brand voice guide, or a small custom setup that applies it automatically, fixes this without anyone doing more work.
High volume, low review time. Once you are generating dozens of replies a day and a person cannot realistically check every one, you need either fewer categories that get pre-approved templates, or a properly reviewed automation with guardrails, not a general chatbot running unsupervised.
It needs to pull real data in. An email that has to reference the actual invoice number, the actual amount owed and the actual due date from your accounting system is not a writing job any more. That is connected automation with a writing step at the end.
If a person is sitting there anyway, checking the result before it goes out, a subscription is doing the job. The moment nobody is checking, or the tone has to be perfectly consistent across a team without extra effort, that is a legitimate small build, not a subscription problem.
Written by the AI Builds team. AI Builds is operated by Media Mantra, ABN 89 594 851 152.
She checks the job, writes the scope, and never sells you a build.