Why AI struggles in marketing — and it isn't because it's dumb

There's a course for everything now. "AI FOR MARKETERS!! Grab it before your competitors do!!" Let me save you the money.

Models learn where the feedback is instant

AI works brilliantly where the loop closes in seconds: write it, look at it, fix it, repeat. That's exactly why copywriters, developers and designers caught the first wave — the "did it work?" answer arrives immediately, and the model learns from its own misses on the fly.

Marketing's feedback arrives in weeks

Now take marketing. Time to result isn't measured in seconds but in weeks, sometimes months. You do a thing, and the feedback shows up much later — if it shows up at all, assuming you were smart enough to set up analytics. First the warm-up, then the touchpoints, then delayed conversions, and only a month later does it dawn on you that the budget went down the drain back in week two.

There is nothing for the model to learn from quickly. Technically there is — but the signal comes so late that by the time "oh, it didn't work" arrives, the money is real and spent, the audience has changed its patterns, and the angle has burned out. Which is why vibe-coding your marketing the way you'd vibe-code a weekend landing page leads straight to an empty ad account and a very sad ROAS.

The unglamorous part: making the model actually work

I've spent an indecent amount of time wrestling models into producing not a stream of pretty water but a clean, defensible result: the model reaching for the right tool instead of poking at them at random, reading the context instead of inventing it, not hallucinating, not burning budget, not drowning the report in filler.

That work is far less photogenic than any "CLAUDE FOR MARKETERS" course. It's dotting every i, over and over.

What it looks like when it does work

Competitor research: 30 minutes. Market research: about 25. Assessing an idea: roughly an hour. And depth is a dial — thousands of keywords, hundreds of creatives, tens of thousands of mentions, and the output is a structured summary rather than porridge.

At real research volume that doesn't save hours. It saves weeks and thousands of dollars. That's what AI market research is actually for, and it's why AI competitor analysis stopped being a two-week project here.

The next step is triggering the research itself

The genuinely interesting part is automating the initiation — research that launches when it should, without you in the loop. For now that lives in my architectural wet dreams, but the shape is clear.

Bottom line

I'm all for AI in marketing. I'm firmly against firing blind and calling it a strategy. The whole difference between "asked a model and hoped" and "made a model actually work" is, by the look of it, the main profession of the next couple of years.