Straight answer: use AI tools for the parts of your work you’ve already mastered by hand, the repetitive first drafts, the summarising, the reformatting, and the admin you resent doing. Don’t use AI for the judgment calls, the relationship-based work, or anything you couldn’t do competently yourself if the tool vanished tomorrow. The mistake almost everyone makes isn’t using AI too much, it’s using it on the wrong 20 percent of the job.
The question I get asked most, and why it’s the wrong question
People ask me “which AI tools should I be using in my business” as if the tool is the decision. It isn’t. I’ve sat with clients who have ChatGPT, Claude, Gemini, Perplexity and three separate AI writing assistants all open at once, and they’re still less productive than they were eighteen months ago with none of it. The tool was never the bottleneck. The decision about when to reach for it was.
I rebuilt my own workflow around this the hard way. Five years ago I was drowning, doing everything myself after losing the big brand deals and the influencer income that used to fund a team. When AI tools got good enough to help, around 2023, I threw them at everything. Proposals, client emails, LinkedIn posts, even a keynote outline for a paying gig. Some of that went brilliantly. Some of it was a proper mess, and one client noticed.
The client email that taught me the rule
I had a warm lead, a fintech founder in Leeds who’d booked a call after seeing my content. I used AI to draft the follow-up email summarising what we’d agreed. It read fine. Grammatically flawless, well structured, on brand voice on paper. He replied asking if I’d written it myself, because it “didn’t sound like the call we’d had.” He was right. The AI had smoothed over three specific numbers he’d mentioned and replaced them with generic phrasing. I hadn’t caught it because I was skimming, trusting the tool to hold the details I should have been holding myself.
That’s the moment I built my own rule: AI drafts the shape, I own the substance. If the substance came from a real conversation, a real number, a real promise to a real person, I write that part myself or I edit it so hard the AI version barely survives.
Where AI earns its place in your workflow
These are the tasks where I’ve measured a genuine time saving, not a guess, an actual before-and-after comparison over several weeks of my own work:
- First drafts of repeatable content. LinkedIn posts, blog outlines, email newsletters. Before AI, a LinkedIn post took me around 35 to 40 minutes including research. With AI drafting the structure and me rewriting the opening line and closing thought, it’s down to 10 to 12 minutes.
- Summarising long documents or calls. I use transcription tools like Otter.ai for client calls, then feed the transcript into an AI tool to pull out action points. A 60-minute strategy call used to take me 20 minutes to write up. Now it’s 5.
- Reformatting and repurposing existing work. Turning a blog post into five LinkedIn posts, or a webinar transcript into a lead magnet. This is pure mechanical transformation of material you’ve already created and already trust.
- First-pass research. Getting an overview of a market, a competitor, or an industry term before you go deeper. Treat it as a starting point, not a citation.
- Admin you’d otherwise avoid. Drafting meeting agendas, structuring a proposal template, writing a first version of terms and conditions before your solicitor looks at it.
Notice what these all have in common: in every one, I could already do the task competently by hand before I let AI touch it. That’s not a coincidence, it’s the actual rule.
Where AI quietly makes things worse
Here’s the part most people skip past because it’s less flattering to the tools. AI is worst exactly where you’d most want to save time: on the work you’re least confident doing yourself. If you don’t know what a good sales page looks like, AI won’t tell you, it’ll give you a confident, fluent, wrong answer and you won’t be able to spot the difference. That’s the uncomfortable truth nobody selling AI courses wants to say out loud: the tool is only as good as your ability to judge its output, and if that judgment is weak, AI doesn’t fix it, it hides it behind good grammar.
Specific situations where I now refuse to use AI, or use it only with heavy restriction:
- Pricing and proposals for high-value clients. The numbers, the specific promises, the tone that matches how that particular person talks. This gets written or heavily rewritten by me.
- Anything involving a real disagreement or a difficult conversation. A refund dispute, a scope-creep pushback, a “we need to talk about the invoice” email. AI tends to soften these into mush that doesn’t resolve anything.
- Strategic decisions. Which market to enter, which service to drop, whether to raise prices. AI can lay out options; it has no skin in the game and no feel for your specific risk tolerance.
- Final client-facing copy where your name is on it. If a client can Google a phrase from your “personal” email and find it verbatim in ten other AI-generated emails, that’s a problem you created, not the tool.
I’ve also stopped using AI to write anything where the value is entirely in the fact that a human noticed something specific. A LinkedIn comment on someone’s post celebrating a launch. A birthday message to a long-term client. These take two minutes by hand and lose all their value the second they sound generated.
A four-step test before you add any AI tool to your workflow
I run every new tool through this before it gets anywhere near my actual work, and I’d recommend the same discipline before you add another subscription to a marketing tech stack that’s already bloated:
- Step 1: Can I already do this task well by hand? If no, don’t automate it yet, learn it first, then automate.
- Step 2: Is the output checkable in under five minutes? If checking the AI’s work takes as long as doing it yourself, you haven’t saved anything.
- Step 3: Does this task happen at least weekly? One-off tasks rarely justify learning a new tool’s quirks. Save AI tools for things you’ll repeat.
- Step 4: What’s the actual cost of a mistake here? A slightly generic LinkedIn post costs you nothing. A wrong number in a client contract costs you the client.
If a task passes all four, use AI on it without guilt. If it fails step 4 badly, keep it entirely human, no exceptions, regardless of how much time it would save.
The number that changed how I audit my own stack
I tracked my own time for six weeks last year, logging every task where I used an AI tool and how long it took start to finish versus my old average. Across 40-odd tasks, AI saved me a combined 11 hours a week. But three tasks, all client-facing final drafts, cost me more time than doing them manually, because I was editing out generic phrasing that took longer than writing from scratch. The lesson wasn’t “AI doesn’t work.” It was “AI doesn’t work everywhere, and pretending it does is expensive.” If you want a rougher version of this exercise for your own business, the AI marketing calculator is a decent starting point for working out where your own AI spend is paying you back.
How to tell if you’re using AI as a crutch rather than a tool
A crutch replaces a skill you should be building. A tool speeds up a skill you already have. The test is simple: take the AI away for a week. If your output quality drops because you’ve forgotten how to structure an argument, write a clear sentence, or spot a weak proposal, you were leaning on the tool to cover a gap, not to save time on a strength. I see this most in newer business owners who’ve never written a proposal without AI help, and who cannot tell you why one paragraph is stronger than another. That’s not a workflow problem, that’s a skills problem wearing a workflow costume.
This matters more the smaller your business is. If you’re a solo consultant or a team of three, your judgment is the product. Outsource the typing, never the thinking.
Where this fits with the rest of your tools
AI tools don’t sit in isolation, they sit inside whatever else you’re already running, and most small businesses I work with are paying for overlapping software they’ve forgotten they own. Before you add another AI subscription, it’s worth doing a proper pass through your existing setup using something like this guide to finding productivity tools that save you time, because half the “AI problem” I see in client audits is really a “nobody’s cancelled the trial from March” problem.
The same goes for content specifically. Not every AI writing tool is worth the monthly fee once you factor in editing time, and I’ve written a fairly blunt breakdown of which content creation tools are worth paying for and which to skip based on what my own team has kept versus cancelled.
When it’s worth getting outside help instead of guessing
If you’ve read this far and you’re still not sure where the line sits in your specific business, that’s normal, because the line is different for a solicitor’s practice than it is for an ecommerce brand. This is the most common reason clients bring me in: not to teach them what AI is, but to sit inside their actual workflow for a day and mark up which tasks are safe to automate and which aren’t. If that’s where you are, an AI implementation coach can save you months of trial and error that would otherwise cost you client trust while you work it out yourself.
The bottom line on timing
Use AI when the task is repeatable, checkable in minutes, and low-stakes if it’s slightly wrong. Keep it away from anything where your specific judgment, your relationship with the person, or a real number is the whole point of the task. Most people get this backwards, they automate the parts that matter most because those are the parts that feel most tedious, and they hand-craft the LinkedIn caption nobody will remember by Thursday. Flip that, and your workflow gets faster without getting worse.
Frequently asked questions
Should small businesses use AI tools for customer emails?
Yes for routine confirmations, order updates, and FAQ responses, no for complaints, refund disputes, or any email where the customer needs to feel heard rather than processed. Draft with AI, edit the specific details yourself before sending anything that matters to a relationship.
How do I know if an AI tool is saving me time?
Track it for two weeks. Time yourself doing a task the old way, then time the AI-assisted version including all your editing and checking. If the AI version isn’t at least 30 percent faster once editing is included, it isn’t earning its place yet, and you should either change how you’re prompting it or drop it for that task.
Is it unprofessional to use AI for client work?
It’s unprofessional to send AI output your client can tell is generic, or that gets a fact wrong. It’s not unprofessional to use AI to structure a first draft you then rewrite. Clients generally don’t object to the tool, they object to feeling like a template.
What’s the biggest mistake people make when adding AI to their workflow?
Automating the task they’re worst at rather than the task they’re best at. People assume AI should help them with weak spots, but AI needs your judgment to check its work, and you can’t judge what you don’t already understand. Start with your strengths, not your weaknesses.
