There is a particular pleasure in walking into a café and hearing, “The usual?” Nobody has run an algorithm. The barista has simply paid attention, and you leave feeling slightly more like a regular and slightly less like a transaction.
Now consider the other version. You search once for flights to Mauritius, and for the next fortnight every website you visit seems to be offering you luggage, sun cream and travel insurance. Technically, this is the same thing: a business noticing what you might want. Emotionally, it lands somewhere between impressive and faintly unsettling.
Brands are being told that AI makes this kind of personalisation available to everyone. Predictive tools promise to anticipate what each customer needs, adapt content in real time and match messages to moods. Some of that is genuinely useful. But it skips an awkward question: why does one kind of attention feel like care, while another feels like being followed around the shop?
The answer has less to do with technology than with psychology. And it matters for small businesses, because the tools are now cheap enough that anyone can cross the line without noticing.
One Like insight: Personalisation works when customers feel remembered. It fails when they feel watched.
Why feeling known feels good
The café works for two simple reasons. It saves you effort, and it signals that you matter. You don’t have to explain yourself again, and someone has thought about you when they didn’t strictly need to. That combination is one of the oldest forms of customer loyalty, and it predates the loyalty card by a few centuries.
Personalisation at scale promises the same thing. The trouble is that most customers don’t seem to be receiving it. In a June 2026 survey of 800 US adults by Site Impact, 77% said they noticed personalised marketing constantly or often, yet only 27% felt brands understood their interests very well. The word respondents chose most often to describe personalised marketing wasn’t relevant or helpful. It was repetitive.
That gap is revealing. Customers aren’t complaining that brands know too little. They’re complaining that brands have collected plenty and understood very little: one search becomes a fortnight of adverts, and a purchase you’ve already made keeps being offered back to you. The café owner would never do this. Nobody serves you a second flat white because you once ordered one in 2024.
When personalisation does land, the same survey suggests it still works. The recommendation people appreciated most was refreshingly unglamorous: a discount on something they had already been researching. Not a clever prediction about their hobbies, just a helpful response to something they had made clear.
Highlight: Customers aren’t asking to be predicted. They’re asking to be understood.
Where helpful turns creepy
So what separates the barista from the advert that follows you around? Researchers Tami Kim, Kate Barasz and Leslie John offered a useful answer in a 2019 study for the Journal of Consumer Research. They drew on the unwritten social rules about how information should pass between people, and found that customers apply much the same rules to businesses.
Their experiments identified two distinctions that mattered. People were more comfortable when personalisation used information gathered on the same site rather than tracked across other websites. And they were more comfortable with information they had stated themselves than with information the business had inferred about them. When an advert revealed one of the less acceptable routes, it became less effective, because privacy concerns began to outweigh the appeal of relevance.
This maps neatly onto everyday life. If you tell a friend you’re training for a marathon, you won’t mind when they ask how the running is going. If they work it out from your shopping bags and bring it up unprompted, it feels odd, even though they’re right.
[Infographic placement] The comfort grid. A simple 2×2 grid. Horizontal axis: “You told us” versus “We worked it out”. Vertical axis: “Here, with us” versus “Somewhere else”. The top-left square (told us, here) is labelled “Remembered” in a warm colour; the bottom-right square (worked it out, somewhere else) is labelled “Watched” in a cool colour. Add one everyday example in each square, such as “The usual?” for Remembered and “Ads for something you searched elsewhere” for Watched. Based on Kim, Barasz and John (2019).
AI raises the stakes because it is very good at inference. That’s its whole appeal. Yet the more a business relies on guessing, the closer it moves to the uncomfortable corner of that grid. Recent survey data suggests customers have noticed. The Usercentrics State of Digital Trust 2026 study of 11,000 consumers in seven countries found that 71% worried AI-driven personalisation felt intrusive. Its authors put the finding neatly: “Personalization isn’t the problem. Unexplained personalization is.” Among people who understood how their data was used, comfort with personalisation was nearly three times higher.
It’s worth noting that Usercentrics sells consent-management software, so it has an interest in that conclusion. But the finding echoes the academic research: people don’t object to being known. They object to not knowing how.
Anticipating versus presuming
The newest promise in this space is anticipatory design: experiences that predict what you want before you ask, and quietly remove the decision. The reorder button that appears just as you’re running low. The playlist that seems to know you’ve had a long day.
There’s real value in this. Most of us have enough decisions to make without choosing between seventeen nearly identical washing powders. But a group of researchers led by INSEAD’s Quentin André and Ziv Carmon pointed out an uncomfortable trade-off back in 2018. Automated choice, they argued, can make decisions easier and more efficient while also eroding people’s sense of autonomy. And that sense of being in control is part of what makes people feel good about what they buy. Their paper is a review that sets out open questions rather than a final verdict, but the tension it describes is easy to recognise.
Think of the difference between a good waiter and an overbearing one. The good waiter notices your glass is nearly empty and catches your eye. The overbearing one refills it without asking, chooses your dessert and seems surprised when you hesitate. Both are anticipating. Only one leaves you feeling looked after.
That suggests a practical test for any personalisation you’re considering, whether it runs on AI or a well-kept notebook. Does it offer a shortcut, or does it take a decision away? Does the customer understand why they’re seeing it? And can they easily say, “Not this time”?
Highlight: The best anticipation feels like a suggestion. The worst feels like a decision someone made for you.
What this means for a small business
Here’s the slightly cheering part. Small businesses have always been good at the kind of personalisation customers actually like. They remember names, preferences and the story behind an order. They use what customers tell them, in the place the customer told them. That’s the comfortable corner of the grid, and big brands spend fortunes trying to recreate it.
AI tools can help with this. They can remind you that a client’s contract renews next month, or that a regular hasn’t visited in a while. Used that way, they extend your memory. The risk comes when the tool starts doing your guessing for you, using signals the customer never knowingly shared, in places they didn’t expect to see them.
A few questions can keep you on the right side of that line:
- Did the customer tell us this, or did we work it out?
- Would they be comfortable if we explained how we knew?
- Are we responding to what they want now, or to something they glanced at once?
- Have we made it easy for them to change their mind?
The Site Impact report ends with a line from its chief executive, Jennifer Gressman, that applies well beyond email marketing: “Consumers don’t reward brands for knowing everything.” They reward businesses for getting the moment right.
Which brings us back to the café. The barista who says “The usual?” isn’t impressive because of what they know. They’re impressive because of what they do with it, and because they’d happily make you something different if you asked.
Sources
- Site Impact, “Consumers say personalized marketing knows too much and understands too little”, 17 August 2026. Survey of 800 US adults via Pollfish, June 2026.
- Tami Kim, Kate Barasz and Leslie K. John, “Why Am I Seeing This Ad? The Effect of Ad Transparency on Ad Effectiveness”, Journal of Consumer Research 45(5), 2019.
- Usercentrics and Sapio Research, State of Digital Trust 2026. 11,000 consumers in seven markets, March 2026.
- Quentin André, Ziv Carmon, Klaus Wertenbroch and others, “Consumer Choice and Autonomy in the Age of Artificial Intelligence and Big Data”, Customer Needs and Solutions, 2018.
