AI in Marketing: What It Should Do, and What Must Stay Human

You can feel it now. Scroll any feed and you can spot the AI posts in two lines.

The same rhythm. The same three-part lists. The same words nobody actually says out loud. Brands that used to sound like people now sound like the same assistant wearing different logos. I wrote about why buyers notice in The AI Copy Tell.

I build almost everything with AI agents. Pages, emails, ad scripts, CRM workflows. So this is not an anti-AI post. It is the opposite. Because I use it every day, I know exactly where it helps and where it quietly destroys trust. The work can be automated. The judgment cannot.

What should AI do in your marketing?

Anything repeatable, where a good brief and a good reviewer can judge the output quickly.

The build work

Sales pages, checkout pages, email sequences, ad scripts, CRM workflows. This is where AI saves weeks. My agents build an entire revenue system around an offer in one working session, usually three to five hours, and the next three days are testing.

Speed in the inbox

Answering DMs, comments and texts in seconds, from answers you already approved. That is the job I give an inbox agent. Speed is something a machine is simply better at than a tired human at 11 PM.

Variation

Ad creative needs volume and range. AI is excellent at producing many genuinely different angles from one strategy, as long as a human picks which angles are true to the offer.

Summaries for faster decisions

Reading a breakdown report, summarizing a sales call, flagging which ads are burning money. AI can get the information in front of you faster. You still make the call.

What should AI never do in your marketing?

Four things. These are not preferences. They are lines.

Never invent proof

No AI-written testimonials. No made-up client stories. No results you cannot show on a sheet. When the FTC finalized its rule against fake reviews in August 2024, it said the rule would let it seek civil penalties and help deter AI-generated fakes.¹

On my own results page, every number comes from a client sheet, with first names only, and the weak months stay in. That is the standard. If you cannot prove it, do not publish it.

AI makes this temptation stronger, not weaker. It will happily write a glowing testimonial or a client story that sounds real. That is exactly why the rule has to be written down and enforced by a human.

Never make the final strategy call

Which offer to lead with. What to charge. Whether to scale. Which promise is honest for your audience. AI can model options. It cannot carry the consequences, and it does not know your clients the way you do.

Use it like a sharp analyst. Ask it to argue both sides, find the holes in your plan, or pull the numbers together. Then close the laptop and decide like an owner. The moment you let a model choose your positioning, you have handed the most valuable decision in your business to the one participant with nothing at stake.

Never publish claims nobody approved

Every claim that reaches a buyer should pass a human first (the wider principle is in Where AI Stops and a Human Starts). In my builds, I check every page, email and tracking event before it ships, and the client approves every claim before a dollar of ad spend goes out.

Never pretend to be human

If someone asks my inbox agent whether it is AI, it says so. An agent that lies about what it is will cost you more trust than it ever earns you in speed.

Why does robotic AI tone hurt an expert brand?

Because your buyers are not buying information. They can get information anywhere. They are buying you: your judgment, your style, your history.

Volume is cheap now

By Meta's own count, its generative AI tools turned out upward of 15 million ads in a single month.² When anyone can produce volume, volume stops being an advantage. What stands out is specificity, a real point of view, and proof.

The signs of robotic copy

You know them when you see them, and so do your buyers:

• Words nobody says in conversation, the kind that sound like a press release.
• Every paragraph the same length. Every list exactly three items.
• Big claims with no numbers, no names and no source.
• Questions the writer answers immediately, every time.
• No opinion. Nothing a reader could disagree with.

The fix is not to stop using AI. The fix is to give it something worth saying. A real point of view, real numbers, a real story you lived, and a human editor who cuts anything that sounds like it came from a template. When the input is specific, the output can be too.

Matt Wolfe, who covers AI tools for a living, hit this wall on camera in a video on the GPT-5 launch he posted on August 7, 2025. He had fed the model his own transcripts and asked for a script in his style: "Inevitably, the script never actually sounds like my voice."⁴ If a creator who tests these tools daily still gets that result, an editor who knows your voice is not optional.

How do you keep AI output in your own voice?

You train it on you, and you never let it ship without you.

My agents run on a private library I built: my design systems, motion graphics, niche by niche copy frameworks, offer logic, editing standards and tracking specs from 12 years of campaigns. That library is the difference between AI that sounds like me and AI that sounds like everyone.

What does a safe AI marketing workflow look like?

Anthropic's guidance on building agents favors starting simple and adding complexity only when it pays, with agents that stop at checkpoints for a human decision.³ That is the right instinct for marketing too.

If you want this built for your business with the checkpoints already in place, that is the system I build for clients. And if you run a team, the same agents and standards can be installed so your operators work this way.

AI did not make marketing easier. It made average marketing free. Which means everything that is not average, your voice, your proof, your judgment, just became more valuable.

Keep building.Ekai