Table of Contents
There’s a particular kind of email I’ve started to notice. Technically fine. Grammatically perfect. Completely hollow.
You know the one. A supplier apologises for a delay in three tidy paragraphs, and somehow says nothing at all. No explanation, no ownership, no sense that a person on the other end felt bad about it.
That’s the thing that worries me about AI in small business. Not robots taking over. Just a slow flattening of everything that made small businesses worth dealing with in the first place.
So here’s how I’d think about using these tools properly — keeping the time savings without hollowing out the bit customers actually value.
First, be honest about why people choose you
Small businesses rarely win on price or scale. They win on responsiveness, judgement and the feeling that someone cares whether the job goes well.
That’s the asset. Everything below is really about not spending it.
Which leads to a simple test I keep coming back to: would this customer be annoyed if they found out AI wrote it?
If the answer is no — a booking confirmation, a delivery update, a product description — go ahead. If the answer is yes, that’s your line. Condolences, complaints, apologies, bad news. Write those yourself.
The stuff AI should never touch
I’ll be opinionated here, because I think the middle ground gets people into trouble.
- Apologies. If something went wrong, the customer wants to know a human is bothered by it.
- Complaint responses. A polished non-answer is worse than an awkward real one.
- Anything involving grief, illness or hardship. Obvious, I’d hope. It still happens.
- Firing, warning, or difficult staff conversations. Just don’t.
- Reference checks and hiring decisions. Use AI to sort logistics, not to judge people.
The pattern is emotional weight. Where the message carries feeling, a human needs to carry it too.
Where the human touch actually survives
The good news is that most day-to-day AI use doesn’t threaten anything.
Nobody has ever felt a connection with a calendar reminder. Nobody treasures your invoice reminders. If AI handles the scheduling, the transcription, the data entry and the first draft of routine copy, your customers won’t notice and shouldn’t care.
In fact, here’s the argument I find most persuasive. Used well, AI gives you more human time, not less. The tradie who stops doing quotes at 9pm has energy for an actual phone call the next morning. That call is where the relationship lives.
It only goes wrong when businesses bank the time saving and don’t reinvest any of it.
Tell people when it’s AI
This one’s becoming less optional.
The Australian Government’s Guidance for AI Adoption, published through the National AI Centre, builds on the Voluntary AI Safety Standard. A recurring theme in those guardrails is giving people a way to understand and challenge how AI is being used.
There’s a legal edge to it too. From 10 December 2026, organisations must disclose in their privacy policies where personal information is used in substantially automated decisions that could significantly affect someone’s rights or interests. Formal OAIC guidance on this has been expected around September 2026.
But forget compliance for a second. Disclosure is just good manners.
A line on your chatbot saying it’s automated, with an easy route to a real person, costs you nothing. Customers are far more forgiving of AI they were told about than AI they discovered.
What disclosure looks like in practice
You don’t need legalese. Something like:
“You’re chatting with our automated assistant. Type ‘human’ anytime and we’ll put you through to the team.”
That’s it. Clear, unfussy, respectful.
Keep a human in the loop — and mean it
“Human in the loop” gets said a lot and practised rarely. In many small businesses it means someone glances at the output and hits send.
That’s not review. That’s rubber-stamping.
Real oversight means someone is genuinely responsible for what goes out, and has both the time and the authority to change it. If your team member feels they can’t override the AI’s draft, you don’t have oversight. You have a formality.
One thing I’d look at first: who owns this in your business? The government’s guardrails are blunt on the point — accountability for safe AI use sits with leadership and can’t be delegated away. In a five-person business that’s probably you.
Protect customer information properly
This is the unglamorous risk, and the one I see handled worst.
People paste client details, medical notes, contracts and payroll data into free AI tools without a second thought. The basic rule is straightforward: don’t put personal, sensitive or confidential data into public AI tools without safeguards.
A few practical habits:
- Use paid business or enterprise plans for anything involving customer data. They generally exclude your inputs from model training and give you admin controls.
- Strip identifying details before pasting. “Client in Brunswick” instead of a name and address.
- Write down which tools are approved. Shadow AI use is the norm, not the exception.
- Update your privacy policy to reflect AI use, particularly ahead of the December 2026 automated decision-making requirements.
None of this is difficult. It just requires deciding once, rather than leaving it to whoever’s on the keyboard.
Don’t let AI write in a voice that isn’t yours
Here’s a frustration I’ll admit to. Most AI writing has a tone — slightly corporate, slightly eager, allergic to a short sentence. If you publish it unedited, your business starts sounding like everyone else’s.
The fix is boring but effective. Feed the tool examples of how you actually write. Old emails, past newsletters, your website’s better pages. Then edit hard.
Cut the throat-clearing. Delete the phrase “we’re excited to announce” every single time. Put back the small specifics only you would know — the street the job was on, the weather that week, the thing that went slightly wrong.
Specifics are what make writing sound human. AI is generic by default.
Where it gets a little tricky
Not every call is clean.
Reviews. Should you use AI to draft responses to Google reviews? I’d say yes for the positive ones, no for the negative ones. A complaint answered by an obvious template reads as contempt.
Sales calls. AI-generated call summaries are great. AI-generated voice agents making outbound calls are, in my view, still a reputational risk for a local business. Different story for a national brand.
Quoting. Fine for the formatting and the follow-up. Not fine for the judgement call about scope and price. That’s your expertise and it’s what you’re being paid for.
There’s no single right answer for any of these. It depends on your industry, your customers and how much personal contact they expect.
A short policy beats a long one
You don’t need a governance framework. You need two pages your team will actually read, covering:
- Which tools are approved
- What data must never be entered
- What always needs human review before it goes out
- Who to ask when something’s unclear
Stick it in your onboarding. Revisit it twice a year. Done.
For those who want more structure, the Voluntary AI Safety Standard and the Guidance for AI Adoption on the INDUSTRY site are free and reasonably readable. Most small businesses sit in the “deployer” category — you’re buying and using commercial tools, not building models, so a lot of it won’t apply to you. Skim accordingly.
My honest take
I don’t think AI is going to strip the humanity out of small business. I think laziness might, and AI makes laziness easier.
The businesses that get this right seem to treat AI as backstage crew. It sets up, clears away, handles the repetitive work. It doesn’t go on stage.
Keep the awkward, warm, specific parts of how you communicate. Those were never inefficiencies. They were the product.
And if you’re ever unsure whether something needs a human, it probably does.
WebFix Business Blog
Latest posts by WebFix Business Blog (see all)
- How Small Businesses Can Use AI Responsibly Without Losing the Human Touch - September 19, 2026
- How Small Businesses Are Navigating Rising Costs and Other Challenges in Australia - August 25, 2026
- The Australian Ecommerce Starter Guide to Buying From China - May 28, 2026
