Most of the AI-for-marketing content I see is about the big swing. Replace your agency. Generate a whole campaign. Ten times the output. That has never really matched my experience.
What has actually changed my week is smaller and duller than that. Three jobs I used to do by hand, that now take a fraction of the time. None of them would make a good LinkedIn headline, which is probably why nobody writes about them.
1. The weekly numbers, turned into words
Every week there is a pull of the same metrics, and every week someone has to turn that into a paragraph a human will actually read.
The pull itself was never the slow part. The slow part was writing "here is what moved and here is what it probably means" for the fifth week in a row, in a tone that suits the person reading it.
So now I paste the export in and ask for a first pass: what changed, by how much, and what stands out as unusual. I get back a plain summary in about a minute.
Then I do the actual job. The AI can tell me a number went up. It cannot tell me that it went up because we shifted send time, or because a promotion was running, or because last week's comparison was distorted by a public holiday. That context only exists in my head, and it is the entire value of the summary.
What used to be forty minutes of writing is now about ten minutes of editing.
2. One piece of content, all its channel versions
Write something decent once, and you immediately owe four more versions of it. The short social version. The email subject line and preheader. The cut-down ad copy. The internal one-liner explaining why it exists.
This is the automation I would recommend to almost anyone, because the source material is already good. I am not asking AI to invent an idea. I am asking it to reshape an idea I have already approved, into formats with known constraints.
I give it the finished piece, the channel, the character limit, and two or three examples of how we normally sound. The examples matter far more than the instructions do. Without them you get generic copy. With them you get something that needs a trim rather than a rewrite.
I still rewrite the subject lines myself, most of the time. They are short enough that the human version is usually better, and they are the bit that actually decides whether anything else gets read.
3. Messy notes into a usable brief
Half an hour of meeting notes, a few Slack messages, and a vague sentence about what someone wants. That is what a brief usually starts as.
Turning that into a structured brief - objective, audience, message, deliverables, what success looks like, what is out of scope - used to be a job I put off until the afternoon.
Now I paste the mess in and ask for it back in that structure, with a list of the things that are still unclear. That last part is the useful bit. The gaps it flags are almost always the questions I should have asked in the meeting.
It is rarely right first time. It does not need to be. Editing a wrong-but-structured draft is much faster than starting at a blank page.
What I don't hand over
Anything where being confidently wrong is expensive.
I don't let it decide who the audience is. I don't let it write the reasoning behind a result. I don't let it near anything I could not personally defend in a meeting, and I don't paste in anything confidential.
The pattern across all three is the same. AI does the reshaping and the first pass. I do the judgement.
It is not doing my job faster. It is doing the parts of my job that were never the job.
Where to start
Pick the task you do every single week, that you already know the shape of, and that you resent slightly. That is the one worth automating first.
Not the most impressive one. The most repetitive one.