AI marketing tools: what to use them for and what to keep human
A clear-eyed guide to using AI in marketing: the tasks it does well, where it produces generic work, and how to choose tools without getting locked into hype.
AI has changed marketing work, but not in the way the hype suggests. It's very good at some tasks, mediocre at others, and actively harmful when used without oversight. Knowing the difference is the skill.
What AI does well
Research and synthesis. Summarising a hundred customer reviews, clustering support tickets, extracting common questions from forum threads. This turns hours of reading into minutes.
First drafts. Post variations, subject lines, ad copy options and scripts. Treat the output as raw material, not finished work.
Repetitive production. Resizing, reformatting, captioning, scheduling and building slide layouts.
Analysis. Spotting patterns in performance data and suggesting what to test.
Idea generation at volume. Producing 30 hooks to choose from is easy; picking the best three is your job.
Where AI falls short
- Original insight. It reflects what already exists. Your unique experience, data and opinions have to come from you.
- Brand voice. Without strong examples and editing, output drifts toward bland and generic.
- Accuracy. Models can state false things confidently. Verify facts, numbers, names and claims.
- Taste. Deciding what's good, what's on-brand and what's embarrassing still needs a person.
- Sensitive situations. Crisis communication, legal claims, health or financial advice.
A sensible division of labour
| Give to AI | Keep human |
|---|---|
| Research summaries | Strategy and positioning |
| Draft variations | Final editing and approval |
| Scheduling and formatting | Replying to real people |
| Reporting and data pulls | Deciding what to do with the data |
| Volume testing | Customer relationships |
Choosing tools
Ask these questions:
- What job does it do? Avoid tools that promise to "do all your marketing". Look for a clear task.
- Does it use your real data? Tools grounded in your brand, customers and current trends produce less generic output than blank-prompt generators.
- Can you edit everything? You should be able to change any output before publishing.
- Is there an approval step? Especially for anything that posts on your behalf.
- What does it cost at your volume? Watch per-generation pricing, especially for video and images.
- What happens to your data? Check privacy terms and where content is stored.
- How does it fit your workflow? A tool that needs constant copying between apps will get abandoned.
Quality control
- Read everything before it goes out.
- Keep a list of phrases and habits you don't want ("in today's fast-paced world").
- Check facts and links.
- Disclose AI-generated realistic media where platforms or law require it.
- Don't publish AI text that claims personal experience you didn't have.
Avoiding sameness
If everyone uses the same tools with the same prompts, the output converges. Differentiate by feeding the tool your own material: customer language, product details, opinions, and real examples. Add something only you could say.
Getting started
Pick one bottleneck (research, drafting, scheduling), pick one tool, run it for a month and measure time saved and quality. Then decide whether to expand. For a fuller version of this, see marketing automation and autopilot.