The question I get asked most often isn’t “should I use AI in my marketing” — most people have already decided yes, broadly. It’s “where do I actually start.” Start in the wrong place and you either waste time on something low-value, or hand over something that needed a person and didn’t get one.
Here’s how I’d think about the order, and the three jobs I’d leave alone for now regardless of how tempting the time saving looks.
Start with tasks that are high-effort and low-risk
The best first candidates share two things: they take a genuinely long time to do by hand, and getting them slightly wrong doesn’t cost much, because a human reviews the result before it goes anywhere.
First drafts of routine written content. Meeting summaries, first passes at blog outlines, social post variations from a brief you’ve already written, draft responses to common enquiries. The AI does the slow part — getting from blank page to something workable — and you do the part that actually needs judgement, which is deciding if it’s right.
Research and summarising. Pulling together a competitor scan, summarising a long report, turning raw notes from a client call into a structured brief. These are tasks where speed genuinely matters and the output is a starting point for your own thinking, not the finished product.
Repetitive formatting and variations. Resizing one piece of copy for five different ad placements, adapting a single message across different audience segments, turning a long piece into a short one. Mechanical work that used to eat an afternoon.
Then move to tasks that are high-effort and higher-risk, with a tighter review step
Drafting client-facing documents. Proposals, reports, onboarding material. Useful, but needs a proper read-through against the checklist in what “good enough” AI marketing output actually looks like before anything goes out — these carry more weight than an internal draft.
Data-led insight. Spotting a pattern in your numbers worth looking into. Good for raising the question — “conversion dropped in week three, want to know why” — bad for being trusted as the final answer. Treat it as a prompt to go and check, not a conclusion.
Three to leave alone
Not because the tools can’t technically do them, but because the cost of getting them wrong is higher than the time saved, or because the thing they’re meant to produce specifically requires a human to be real.
1. Anything claiming to be a genuine testimonial, review or case study. If it didn’t happen, it shouldn’t read as if it did. This one isn’t really a judgement call — fabricated social proof is a trust problem waiting to surface, not a marketing shortcut.
2. Strategic decisions about where the business is heading. Which market to go after, what to stop doing, how to price something. AI tools can help you think through a decision — lay out options, pressure-test an assumption — but the decision itself needs to be yours. It’s your money and your risk.
3. The first reply to a genuinely upset customer or a sensitive situation. Complaint handling, anything involving a real grievance, anything where tone matters more than content. People can usually tell when a response has been generated rather than felt, and it tends to make things worse, not better, at exactly the moment it needed to go right.
A rough way to decide for anything not on this list
Ask two questions: how much time does this actually take me by hand, and how bad is it if the first draft is wrong and nobody catches it before it goes out? High time, low risk — good candidate, start there. Low time, high risk — leave it, the saving isn’t worth it. Anything in between needs the review step to be real, not a formality.
FAQs
What marketing task should I automate with AI first?
Start with high-effort, low-risk tasks that a human reviews before anything goes out — first drafts of routine content, research summaries, and repetitive formatting or variations. These save real time and a wrong first attempt costs little.
What marketing tasks should I never hand to AI?
Fabricated testimonials or case studies, major strategic decisions about the business’s direction, and the first reply to a genuinely upset customer. In each case the risk of getting it wrong — or the fact that it specifically needs to be human — outweighs any time saved.
How do I decide if a task is safe to automate?
Ask two questions: how much time does it take by hand, and how bad is it if a wrong first draft slips through unreviewed? High time and low risk is a good starting point; low time and high risk usually isn’t worth automating at all.
Is AI-generated client-facing content safe to use?
With a proper review step, yes — but it carries more weight than an internal draft, so it needs a closer read against a checklist before it goes out, not a quick skim.
Once you’ve picked a task worth trying, the AI Marketing Opportunity Scorecard walks through a controlled 30-day way to test it properly rather than just diving in. Or get in touch if you want a second opinion on where to start.