Straight answer: AI is excellent at speed, volume and pattern-matching: first drafts, repurposing, summarising, variations on a theme. It is poor at judgement, timing, taste and knowing what your specific audience will forgive or love. Most businesses have the split backwards, using AI to make the big calls and humans to do the typing.
I found this out the expensive way, not the theoretical way
Last year I handed a client's entire LinkedIn content calendar to an AI tool for a month. Not the ideas, the ideas were still mine, but the drafting, the editing, the final polish. Reach on that page dropped from an average of around 8,000 impressions per post to just over 2,200. Same topics, same posting times, same audience. The only thing that changed was who wrote the actual sentences.
That's the part nobody wants to say out loud when they're selling AI courses. The tool didn't fail because it can't write. It failed because it can't tell the difference between a sentence that's technically correct and a sentence that sounds like a real person who's annoyed about something specific. My client posts are full of small, ugly, true details. AI drafts are full of smooth, plausible, forgettable ones. Readers can feel the difference even when they can't name it.
Where AI earns its keep, earns it
I'm not anti-AI, I use it every single day and I'd be lying if I said otherwise. But the tasks where it's brilliant are narrower than the marketing world admits.
- First drafts of anything repetitive. Product descriptions, meta descriptions, ad variations, email subject lines. I ask for 15 subject line options in ten seconds, keep two, bin the rest. That's a genuine win.
- Repurposing long content into short content. Turn a 40-minute podcast transcript into 8 LinkedIn posts, 20 tweet drafts and a newsletter summary in under 15 minutes. I did this with a client webinar last month, a task that used to take my assistant three hours took 25 minutes, including the edit.
- Research summarising. Feed it a competitor's last 12 months of blog posts and ask what themes they hit. It's fast and reasonably accurate at spotting patterns humans would miss on a skim read.
- Data structuring. Turning messy survey answers or customer feedback into themed categories. useful for pulling insight out of a pile of comments nobody was ever going to read by hand.
- SEO grunt work. Meta titles, alt text, internal linking suggestions, keyword clustering. Boring, mechanical, exactly what it should be doing.
Notice the pattern. Every one of those tasks has a right-ish answer that a human can check quickly. There's no taste required, just competence.
Where it falls apart, and where I've watched it fall apart
Now the uncomfortable list, the one that doesn't make it into most webinars because it's harder to sell.
- Positioning and strategy. AI can describe ten positioning frameworks. It cannot tell you that your actual problem is you're too scared to say what makes you different, because it doesn't know you're scared. Positioning is a decision, not an output.
- Timing and restraint. Knowing when NOT to post, when a joke lands and when it's tone deaf given what's happening in the news that week. AI has no sense of the room because it isn't in the room.
- Brand voice consistency over months. It can mimic a style for one paragraph convincingly. Ask it to hold that voice across 50 posts without drifting into generic corporate warmth, and it drifts. I've caught it sliding my client's sharp, sarcastic voice into something that sounds like a bank's Twitter account by post 15.
- Anything that requires knowing a specific human. What made your best customer buy. What your angriest customer meant in that one-star review. AI can guess. It cannot know, because knowing requires having sat with these people.
- Judgement calls under pressure. A product recall, a PR mistake, a competitor doing something aggressive. These need someone with scars, someone who's been wrong before and remembers exactly how it felt.
Here's the bit that makes marketers uncomfortable: the tasks AI is bad at are the tasks that make a brand worth remembering. Oatly's marketing works because of a specific, weird, slightly annoying voice that a committee, let alone a language model, would never approve. Depop's whole approach is built on community judgement calls about what feels authentic to teenagers, a thing no model trained on last year's internet can predict about next year's teenagers. You cannot automate the parts of a brand that make people feel something. You can only automate the parts that get the message out the door.
The uncomfortable truth almost nobody selling AI courses will say
Everyone's speed has gone up. Which means average is no longer impressive, it's the new baseline everyone can hit for free. Five years ago, a well-written email newsletter took effort, so effort itself was a signal of quality. Now anyone can produce a competent-sounding newsletter in four minutes. The bar hasn't lowered. It's moved. Competence is now worthless because it's everywhere, and taste has become the only thing that's scarce.
I see businesses celebrating that AI let them triple their content output, and I want to ask them: did engagement triple too, or did it stay flat, or did it drop like my client's did? Volume without judgement is just more noise, faster. I'd rather post half as often and have every single post edited by a human who knows the difference between clever and try-hard.
A simple test I use to decide what to hand over
When I'm working out what to give to AI and what to keep, I run every task through three questions:
- Is there a checkable right answer? Grammar, structure, factual accuracy, yes. A joke landing, no.
- Would a stranger who's never met my audience get this right too? If yes, hand it over. If the answer depends on knowing your specific customers, keep it human.
- Does getting this wrong cost me trust or just time? If it costs time, let AI have a go and edit fast. If it costs trust, a customer relationship, a reputation, do it yourself or get someone senior on it.
Draft an email subject line, hand it over. Decide how to respond publicly to a customer complaint that's gone viral, that's yours, every time.
Where this leaves small businesses without a marketing team
If you're running marketing alone, or with one part-time person, the split above is your instruction manual. Use AI for the drafting graft you'd otherwise pay an agency £60 to £120 an hour for. Keep the strategy, the voice, the judgement calls close, because that's the part clients and customers notice and remember. I've seen founders spend a fortune on an AI marketing consultant expecting them to hand over full control of the brand voice, and that's backwards. The right consultant helps you figure out exactly this split, what to automate and what to protect, rather than automating everything because it's fast.
If you don't know where your own line should sit, that's worth getting outside eyes on early rather than after six months of forgettable content. I've walked several small businesses through exactly this kind of audit as an AI implementation coach, and it's almost never a tools conversation. It's a "which five things must only ever sound like you" conversation.
Want AI doing the heavy lifting in your marketing?
I build the systems that handle the boring 80 percent, so you get your week back. Done properly, with the human kept in.
Look at any brand that's built something people care about, Shopify's approach included, and you'll find the mechanical stuff, the emails, the ad copy variations, the product feeds, running on tooling and automation for years now. What never got outsourced was the point of view. That's the bit that was true before AI existed and it's still true now, it's just got more expensive to skip.
What I'd do this month
Pick one repetitive task eating your week, product descriptions, social captions, email drafts, and give it fully to AI for four weeks. Track one number before and after: open rate, click rate, engagement, whatever matters for that task. Then pick one task you've been tempted to automate because it's tedious, your brand voice on customer replies, your strategy calls, your big campaign decisions, and deliberately keep it human, even when it's slower. Compare how each one feels after a month. You'll know within a fortnight which list each task belongs on.
Frequently asked questions
What is AI best at in marketing right now?
Speed tasks with a checkable right answer: first drafts, repurposing long content into short formats, subject line and ad copy variations, meta descriptions, and summarising research or customer feedback. Anything where a stranger could do it just as well as you can.
Can AI replace a marketing strategist?
No. AI can describe strategy frameworks and summarise competitor activity, but strategy is a decision that requires knowing your specific market, your risk tolerance and your customers. That judgement doesn't transfer to a model, however good the prompt is.
Why did my engagement drop after I started using AI for content?
Almost always because unedited AI drafts sound generic, and audiences can feel generic even when they can't name it. The fix isn't dropping AI, it's using it for the first draft and having a human do a heavy, specific edit before anything goes out.
How do I decide what marketing tasks to automate with AI?
Ask three things: is there a checkable right answer, would a stranger get it right too, and does getting it wrong cost you time or trust. Time-cost tasks with checkable answers go to AI. Trust-cost tasks that need to sound like you stay with a human.
Related reading: How Does SMS Marketing Work for Small Businesses? A Straight Answer With Real Numbers and How Do You Pick the Right Online Booking Tool for Your Business.
For the bigger picture, see my full guide to AI marketing.
