The number that has not moved in two years
Start with the most sobering figure. A large consultancy has been asking companies the same question for years: is the effect of AI showing up in your results? The answer has stood in the same place for two years. It is not falling and not rising — adoption is broad, measurable business impact does not follow it.
That does not mean there is no effect. It means the effect is not where the demos promise it, and not as large as our own experience makes it feel. Marketing is where this shows up most clearly, because marketing is where the most measurement happens.
What it genuinely changed
Four things where the change is real and measurable.
- The cost of draft zero. An outline, an ad copy variant, the first version of an email now costs minutes instead of hours. That is not a jump in quality, it is a lower barrier to entry: work that was never worth doing now gets done.
- Volume. Fifty ad variants instead of five. On its own that is not worth anything, but it is the precondition for testing.
- The start of research. Reading through reviews, support tickets, competitor material — what used to take days now takes hours.
- Translation and versioning. The same material in several languages and several formats. Here the gain is unambiguous, and nobody disputes it.
What it did not
- The creative idea. Models hand back the average they were trained on. What was average before is now average faster.
- Strategy. Who it is for, how it differs, why now — a model cannot answer these, because it does not know your market.
- Quality. In an industry survey 87% of marketers using AI for content report a productivity gain, but only 39% report better content performance — and 12% report quality getting worse. Volume went up, the hit rate did not.
- Credibility. Audience trust is the one dimension where measurements consistently show a decline once the machine origin comes out.
The experiment that says the most
There is a peer-reviewed 2026 study that put this question in the cleanest possible way: three pre-registered experiments, 27,500 real customers each, on live newsletter campaigns. One group got copy from human copywriters, the other got copy written by a language model.
The result: no difference could be detected between the two groups — not in revenue, not in open rate. The machine copy was not better, and it was not worse either. Where AI won was the part that has nothing to do with the writing: the annual cost. A licence runs around a thousand dollars a year; three copywriters run $370,000.
That is the most honest summary of this entire field. Quality does not go up, unit cost comes down — and commercially that is reason enough, it is just not the reason the marketing material gives.
Where the gain is real — but not where you think
Ad creative has one interesting exception. According to another peer-reviewed 2026 study, the click-through rate of AI-made creative was 0.98 percent against 0.65 for the human designer — a meaningful difference in AI's favour.
Except the same study measured 3.38 and 3.24 percent on those same two groups eighteen months later. The gap had essentially vanished. The explanation is obvious: the human side learned what works too. AI had a temporary lead, not a durable advantage — and that will hold for most areas.
The feeling that misleads you
This part is uncomfortable, but it matters. A tightly controlled experiment watched professionals working with an AI tool in their own field. The participants felt they had become twenty percent faster. The measurement says they were nineteen percent slower.
That does not mean the tool is bad. It means your own impression is not a measurement — and marketing is full of decisions made on impressions. If you bring AI in, the effect has to be measured like anything else: before and after, on the same task.
What to spend time on
- On strategy and positioning. There is no machine solution for this, and it decides what all the rest of the work is worth.
- On measurement. If you do not measure, your impression will drive you — and you just saw what that is worth.
- On your customers' own words. The best marketing copy is not generated, it is collected. Where to collect it is what the customer research from reviews post sets out, with the method.
- On AI search. This is the one area where real, structural change is under way — and where it is worth moving now.
Common misconceptions
- "Copy written with AI is worse." In the largest controlled study there was no detectable difference. The question is not quality, it is cost.
- "Copy written with AI is better." Not true either. Where there was an edge, it melted away within eighteen months.
- "If I feel more productive, I am." In a controlled measurement it came out exactly the other way round: self-estimates were off by 40 percentage points on average. What to measure instead is set out by the measuring the impact post, with four metrics.
- "Volume is a value in itself." Only if you have something to do with it: testing, selection, measurement. Without that you simply get more material.
What to take away
In marketing, AI brought unit cost down, it did not bring quality up — and that is reason enough to use it, just not with the expectation that your work will get better. The biggest real change is not in content production but in the fact that your audience searches differently: that is what the zero-click search post is about, and what you can do about it is set out by the citation in AI answer engines post, with the evidence.