Marketers, Put Your AI Anxiety on Hold

A while back, I had dinner with a friend who works in marketing. Halfway through the meal, he suddenly put down his chopsticks and asked me: "Run, my company brought in generative AI last month. Some people on the team are so excited they can't sleep, and others are worried they're about to be laid off. Tell me — those of us in marketing, what are we supposed to do with it?"

I said: don't rush to be anxious, and don't rush to be excited either. Let me show you a few numbers first.

In 2022, MIT Technology Review ran a survey in which only 5% of marketing organizations considered generative AI "critical" to their business. By 2025, the share of marketing executives planning to make it a core capability of their department had climbed to 20%. Another 44% intend to roll it out across all kinds of use cases.

Three years — from "what is this thing?" to "how do we use it well?" That's how fast things changed.

Or look at Salesforce, which surveyed 1,000 marketers in 2023: more than half were already using it, and another 22% planned to adopt it within a year. In a Statista survey the same year, the figure was 73%.

Seventy-three percent. In other words, of the marketing colleagues around you, seven out of ten have already put it to work.

The question isn't whether to use it. It's how.

What exactly is generative AI?

Let's get the concept straight first.

AI in the past was mainly built to "see": analyzing data for you, spotting patterns. Generative AI is different — it's built to "do": writing copy, creating images, adding audio, editing video, even writing code.

It's a branch of machine learning. The principle, once you strip away the mystique, is simple — using an approach called "deep learning," it imitates the way the human brain forms associations, and from oceans of data it has learned the patterns of how people speak, write, and draw.

So the work it can do overlaps almost perfectly with a marketer's daily grind.

Think about it: what does a marketing department do every day? Come up with ideas, write content, personalize, plan campaign pacing. Which of those isn't something generative AI is good at?

But every coin has its other side.

What can it actually do for you?

I've grouped the most common marketing use cases into three categories.

Category one: doing the work.

This is the most obvious. Give it an outline or a single prompt, and it can produce first drafts of blog posts, emails, and social media updates. And not just text — generating images from text descriptions, translating content into multiple languages, creating charts from data, text-to-speech, adding royalty-free music, summarizing long articles, researching SEO keywords.

What a content team used to produce in a week might now come out in a single day.

Category two: thinking things through.

When the words won't come, many content marketers use it as a sounding board. Throw your target audience and campaign goals at it, and let it toss back a pile of ideas. The ideas may not be directly usable, but they're more than enough to yank you out of your rut.

Category three: understanding people.

This, to me, is the most valuable category of all.

Personalization used to mean "adding the customer's surname to an email." Personalization now means generating content in real time, just for that one person, based on what they've browsed, what they've bought, and every interaction they've had with the company. Same with A/B testing: you used to top out at two or three versions; now you can have AI generate dozens of versions in one go and test them all.

Customer service has changed too. Old-school chatbots could only recite scripts and answer the wrong question. A customer service system powered by generative AI can give human-like answers in dozens of languages, and can even use "sentiment analysis" to hear the anger in a customer's voice and switch to a different tone in response.

Then there's market research. Let AI chew through mountains of unstructured data, distill insights, predict churn, forecast demand, and predict how a campaign will perform. That work used to require a dedicated data scientist.

It's startling, isn't it? Have you noticed? Taken together, these three categories cover roughly half of what a marketing department does.

Exactly. That's why Deloitte surveyed a group of business leaders in 2023 and found that 91% believed generative AI would boost their organization's productivity.

And yet, in that same survey, only 29% of organizations were using it at the strategic level.

That gap is where the opportunity lies. More on that later.

Don't go all in just yet

I've listed a lot of benefits, so now I have to pour some cold water on this.

Using generative AI carries real risks. And for marketers, the risks land precisely on the most critical thing — the "content" it generates.

Here are a few you may not have thought of:

It can spit out training data verbatim, word for word. That's plagiarism.

Its training data may contain misinformation and bias, which will seep straight into the copy it writes for you.

It may have been trained on data the company never got rights to. That's a copyright exposure.

The prompts users type in may be collected without their knowledge. That's a privacy problem.

And then there's regulation. The EU has already drawn red lines: AI-generated content must be labeled, models must be prevented from generating illegal content, and summaries of which copyrighted materials were used in training data must be disclosed. And EU rules don't just govern European companies — American and British companies doing business with the EU have to comply too. In the US, the Federal Trade Commission polices false and deceptive business practices, and deepfakes are bound to end up in its crosshairs sooner or later.

So far, the lawsuits have mostly targeted the companies building the AI — tech giants like Microsoft, Google, Amazon, OpenAI, Apple, Nvidia, and IBM — with the litigation concentrated on the copyright and licensing of training data. No brand using AI has been sued yet.

But "not yet" doesn't mean "never." Once brand reputation takes a hit from one plagiarized piece of copy, that little efficiency dividend won't come close to covering the damages.

AI can do the work for you. It won't do the jail time.

So what's the right way to use it?

The answer is almost embarrassingly simple: a human has to be in the room.

Any AI-generated content must be reviewed by a person before publishing. Review for what? Bias, facts, and copyright risk. Before adopting a model, seriously investigate whether its data sources are clean and lawful. Pilot it on a small scale first, and only roll it out company-wide once it's proven. Add a disclosure when you publish AI-generated content.

One step up from that is data strategy. Some companies are even thinking about training their own model, using brand guidelines and historical campaign data as the training set — content that fits better and risk that's more controllable. Amazon SageMaker is the platform built for exactly that. But for most companies, that's beyond the syllabus. The more realistic path is to build on public models like GPT-4 or Gemini, and keep tuning them with prompt engineering and human feedback.

Don't agonize too much over the tools. For writing, ChatGPT, Gemini, and Claude each have their devotees; for images, DALL-E and the open-source Stable Diffusion will do the job; for strategic analysis, platforms like Alteryx, DataRobot, Skai, and Braze have turned what used to be a data scientist's work into a few clicks for a marketer.

Tools will keep changing. The method won't:

Give it a clear goal, give it clean boundaries, and give it a pair of human eyes.

The real dividing line

Back to that dinner at the beginning.

What did I finally tell my friend? I said: the question you're anxious about — "will I be replaced?" — is the wrong question.

The vast majority of companies using AI today are in it to save time and cut costs. Nothing wrong with that. But what you do with the time you save — that's the dividing line. That's the subtext of the Deloitte survey: nine in ten companies expect it to lift efficiency, fewer than three in ten use it strategically — to find new market segments, optimize media-buying decisions, and rebuild customer journeys.

Efficiency is defense. Strategy is offense.

Once AI drives the cost of "making content" down to floor price, what's scarce is no longer content — it's judgment: judging who to create for, what to create, and why.

That judgment, no machine can give you.

What replaces you won't be AI. It'll be the colleague who figured that out first.

Here's hoping you're not the one left behind.