Using AI honestly: what to automate and what not to
AI promised your small business a part-time team for the price of a coffee subscription. A year in, most owners have a folder of forgettable, slightly robotic content and exactly the same number of customers as before.
The tool isn't the problem — the targeting is. Here's a short, current framework for what to automate, what to keep human, and where the line should sit in 2026.
1. Use AI to draft, never to publish
Treat every AI output as a first draft. Your job is to edit it into something only your business could have written. Anything that ships unread is a liability — both for accuracy and for the bland sameness Google has started filtering out of search results.
2. Automate the busywork, not the thinking
Transcribing a call, summarizing a long email thread, reformatting one announcement for five channels, generating ten subject-line variations — these are perfect AI jobs. The stakes per output are low and you'll still read the result.
3. Use it to break blank-page paralysis
"Give me five rough angles for a post about X" gets you out of the chair. You'll throw away four. The fifth is a real starting point. That alone is worth the subscription.
4. Keep a human on every customer-facing reply
Frustrated customers, complaints, refund requests, discovery calls — automating these transfers a small, slow cost (your attention) into a large, sudden one (a lost client, a one-star review you'll be answering for years).
5. Fact-check anything specific to your industry
Modern models still invent statistics, misquote regulations, and describe features your software doesn't have. For pricing, legal, medical, or technical pages, a human who knows the subject reads every word before it ships.
6. Train it on your own voice
Generic AI output sounds generic because the prompt is generic. Paste in three of your best past emails or articles and ask the model to match that voice. The lift in quality is immediate.
7. Be transparent when it matters
If an article was drafted with AI and edited by a person (like this one!), saying so builds more trust than pretending otherwise — especially as AI-disclosure norms tighten across publishing and search.
8. Measure inquiries, not output volume
Ten AI posts a week that bring zero leads is a worse outcome than one human-edited post a month that brings two. Track what actually moves the needle and cut the rest.
Want help drawing the line for your business? Book a free 20-minute call — we'll map the three places AI will save you the most time without costing you trust. Contact us to schedule.