What is AI marketing?

Where the term actually comes from

It does not have a founding paper. Unlike positioning or brand equity, AI marketing is a trade term that formed around a set of tools between roughly 2022 and 2025, which is why definitions of it vary so much depending on who is selling what.

That matters for how you read any article about it, including this one. There is no canonical source to appeal to. What exists is a set of practices, some well evidenced and some not, grouped under a name that got popular faster than the practices got tested.

What it covers in practice

Four things, and they are genuinely different jobs.

Research and analysis. Clustering customer data, summarising interviews, reading unstructured feedback at a scale a person cannot. The least glamorous and the most reliable of the four.

Creative production. Drafting, variants, resizing, localisation. Real productivity gains, and the place where quality control fails most often, because volume is easy and judgement is not.

Targeting and bidding. Largely automated already, and has been for years. Most of what platforms now market as AI here is machine learning that predates the current wave.

Being present in AI answers. The genuinely new one. When a buyer asks an assistant to recommend a supplier, something decides which names appear. See How do AI models choose which brands to recommend?.

What changes for a brand, and what does not

The mechanics change. The job does not.

Buyers still have to know who you are before they need you, still have to be able to recognise you, and still buy mostly from a small set of names they can recall without effort. That is Byron Sharp’s mental availability, set out in How Brands Grow in 2010 on decades of Ehrenberg-Bass purchase data [1], and none of it is repealed by a language model sitting between the buyer and the search box. If anything it is sharpened, because an assistant that has never encountered your name cannot recommend you, and will not say so. It will simply name somebody else.

The part that genuinely changes is who you are being understood by. A search engine indexed your page. An answer engine builds a description of your company and then speaks on your behalf. If that description is wrong, every answer it gives about you is wrong, consistently, to everyone who asks.

The honest state of the evidence

Two things are worth saying plainly.

Most published claims about AI marketing effectiveness come from companies selling AI marketing tools. Treat a vendor benchmark the way you would treat a supplier’s own case study, because that is what it is.

And the field is moving faster than the evidence can follow. A result measured on one model in one quarter may not hold on the next version. Anything in this section carries the date it was measured for that reason, including our own research.

What to do about it

1. Separate the four jobs above before you buy anything. A tool that is good at creative production tells you nothing about whether you appear in answers.
2. Ask what an AI assistant currently says about your company. Not what it says about your category, what it says about you by name. Most executives have never checked.
3. Check whether the description is accurate before you try to improve your ranking in it. A wrong description that ranks well is worse than no presence.
4. Treat vendor effectiveness claims as unsourced until you see the method.
5. Keep doing the ordinary work. The brands that AI assistants name confidently are mostly the brands that were already well known and consistently described, which is not a coincidence.

If you run the company rather than the marketing

The question to ask is not “what is our AI strategy”. It is “what does ChatGPT say when somebody asks it to recommend a company like ours, and are we in the answer”. That takes two minutes and you can do it yourself, today, without anyone’s help.

Three outcomes. You are named and described correctly, in which case protect it. You are named and described wrongly, which is the most urgent and the least noticed, because the error is being repeated to every person who asks. Or you are absent, which is the most common and the slowest to fix.

Do not let this become a budget line before you have done the two-minute version. A supplier who proposes an AI visibility programme without first showing you what the engines currently say about you is selling before they have diagnosed.