Is "AI Personalization" a Marketer's Remedy, or Just Another Source of Anxiety?

A while ago, I opened a shopping app I use often and searched for "wool socks."

The next day, the ads in my social feeds, my inbox, my text messages — all wool socks.

The day after that, I opened a different app, and the home-page recommendations were wool socks too.

I didn't feel understood. I felt surveilled.

This is the most contradictory reality in marketing circles in 2026: on one side, vendors tell you "AI personalization will save everything"; on the other, users are voting with their feet — turning off tracking, installing ad blockers, blocklisting brands.

After doing some research and talking with several frontline marketers, I slowly worked out one thing:

AI personalization has been oversimplified in the conversation.

What do we even mean by "personalization"?

Let's get the concept straight first.

When vendors say "AI personalization," they usually mean several things bundled together: using machine learning to analyze your browsing, purchases, and location; predicting what you'll want next; and pushing content, products, and prices at you in real time.

Sounds lovely.

But think about it — this is really two things bolted together:

The first is understanding you — and AI is genuinely good at this part.

The second is making you feel understood — and this is the part where AI keeps getting it wrong.

A friend of mine runs user operations at a retail brand in China, and she vented to me:

"Our recommendation algorithm is actually pretty accurate. But users constantly complain to customer service that our pushes are annoying, too precise — like we're snooping on them."

Algorithmic precision and user comfort are not the same thing.

This is the paradox every brand running AI personalization will sooner or later have to face.

What the data tells us: what do users actually want?

I saw a set of numbers that genuinely surprised me.

52% of users expect the deals brands offer them to be tailored. In other words, the default assumption is that you understand them.

But at the same time, 76% of users switch communication channels depending on context. Which means they don't want to be bombarded all day by one app, one email, one SMS.

Put those two together and you get a counterintuitive conclusion:

What users want is not "to be tracked everywhere," but "wherever I speak up, you respond on that channel."

Most brands have it backwards — they read personalization as "I need to study you thoroughly, then disturb you around the clock."

Real personalization is "when you show up, I happen to be on the right channel."

It sounds abstract, but at the operational level it comes down to a few things:

• If a user didn't reply in email, don't immediately chase them with a text.
• If a user searched for A in the app, don't push A across every platform for the next 24 hours.
• If a user turns off your push notifications at 11 p.m., don't pop up a window at 2 a.m.

These are all common sense, but under the slogan of "fully automated AI personalization," they've been quietly ignored.

What vendors tell you vs. what's actually true

Let me walk through a few things in the vendor pitch that I think we should be wary of.

Red flag one: "AI learns automatically — the more you use it, the more accurate it gets."

This is technically true, but operationally it's a trap.

What AI learns isn't just user preferences — it also learns the user's "tolerance threshold" for you. If a user has clicked on three ads in a row without buying, the AI assumes he didn't see them and ramps up the ad spend. But in reality, he may have already decided never to buy anything from you again, for the rest of his life.

Learning goes both ways: the AI learns the user, and the user learns the brand. The more you annoy him, the better he gets at blocking you.

Red flag two: "Real-time bidding + dynamic pricing — revenue maximized."

It's sexy on paper; commercially, be careful.

A certain ride-hailing platform got blasted all over the internet for raising prices on rainy days. A certain airline was called in by regulators for showing different prices to users in different regions based on IP address.

Users' tolerance for a "sense of unfairness" is razor-thin. Personalized pricing is a minefield — play it well and you quietly make money; play it badly and you end up trending for all the wrong reasons.

Red flag three: "Unified omnichannel experience."

Sounds reasonable, but you have to distinguish between "omnichannel" and "always-on."

Omnichannel means: wherever the user opens a conversation, you can pick it up. That's a good thing.

Always-on means: you use every channel to the max, carpet-bombing the same user. That's a disaster.

What users want is "the option to choose," not "to be covered."

Four pieces of advice I'd give marketers

That's a lot of talk — let's bring it down to operations. I have four very practical pieces of advice.

First, separate "data collection" from "data use."

How much data you've collected is not the same as how much you may use. Most users are actually willing to give you data — provided you don't use that data to annoy them.

Build an internal rule: for any piece of data, first ask "after using this, will the user feel comfortable?" and only then decide whether to use it.

Second, do "contextual personalization," not "persona-based personalization."

Persona-based personalization is "I know who you are, so I push what you need." The problem is, user needs are dynamic.

Contextual personalization is "I know what you're doing right now, so I push what you might need in this moment." The former stares at the person; the latter stares at the situation. The latter is lighter — and far less likely to offend.

Third, a transparent data policy is your best brand asset.

Tell users what you collect, what you use it for, and how you protect it. Then actually do it.

It sounds basic, but 90% of brands haven't done it. The 10% that have will see their conversion rates slowly pull ahead, because users start to trust them.

Fourth, humans are the last mile of AI personalization.

AI can handle 80% of standard scenarios. The remaining 20% — the high-value ones like customer complaints, complex consultations, and repurchase decisions — still need people.

Plenty of brands pour money into AI and then gut their human customer service. The result: when a customer hits a complex problem, they bounce around a bot loop, the problem doesn't get solved, and the trust is gone.

The customer-service costs you save will eventually come back to you as higher churn.

In closing

Vendors will tell you AI personalization is the future.

They're not wrong — but what they don't tell you is: get it wrong, and users vote with their feet.

The core of personalization has never been technology. It's a sense of proportion.

It's knowing when to show up, and when to disappear.

It's knowing what data you may use, and what data is off-limits.

It's knowing how far AI can help, and at which step you have to bring a human back in.

The stronger the technology, the more valuable a sense of proportion becomes.

This isn't a new truth of the AI era; it's an old truth of the business world, centuries old.

It's just that everyone's been so busy chasing the new tools lately, they forgot it.

May you make choices with a sense of proportion.