When AI Personalises Too Well: The Uncanny Valley of Marketing

There’s a moment, if you’ve spent enough time watching CGI films or visiting robotics labs, where something shifts. You’re looking at a face that is almost human. The skin texture is right. The eyes move correctly. The proportions are anatomically precise. And yet, something is deeply, viscerally wrong. Not wrong enough to identify. Wrong enough to feel. Your stomach tightens. Your instincts fire. You look away before your conscious mind has even registered why.

Masahiro Mori identified this phenomenon in 1970 and called it the bukimi tani genshō, the uncanny valley. As a robot or animated character approaches human likeness, our emotional response rises from indifference to empathy, until it reaches a point where the resemblance is so close, yet so imperfect, that the response collapses into revulsion. The almost-human becomes more disturbing than the clearly mechanical.

Marketing has its own uncanny valley. And thanks to artificial intelligence, we are walking straight into it.

The Personalisation Arms Race

Let’s trace the trajectory, because the path matters.

Twenty years ago, personalisation meant putting someone’s first name in an email subject line. “Dear David, we thought you might like…” It was crude, transparent, and mostly harmless. Nobody felt surveilled by a mail merge.

Then came behavioural targeting. Cookies tracked what you clicked, what you bought, how long you lingered on a product page. Segments grew sharper. Instead of “men aged 25 to 34 in London,” we had “people who viewed running shoes in the last 48 hours but didn’t purchase, and who have previously bought from brands in the mid-premium tier.” Better. More useful. Still, broadly speaking, people understood what was happening. The mechanism was visible enough: you searched for something, you got ads for it. Cause and effect. Fair enough.

Now we are in the age of AI-driven personalisation, and the game has changed entirely. Machine learning models don’t just track what you do. They infer what you intend. They read context: the time of day, the device you’re using, the weather in your postcode, the patterns in your browsing behaviour that you yourself haven’t consciously registered. They predict. They anticipate. They know.

And here is the problem. The more accurately they know, the more uncomfortable you become.

The Spectrum of Precision

Think of it as a spectrum.

At one end sits the generic ad. A billboard for car insurance that everyone driving past sees. It is irrelevant to most people, mildly useful to a few, and emotionally inert to all. Nobody feels anything about it. It is marketing wallpaper.

In the middle sits the relevant ad. You searched for flights to Lisbon yesterday, and today you see a promoted post for a boutique hotel in Alfama. Useful. Timely. Perhaps even welcome. You don’t think twice about it. The mechanism is legible: you expressed intent, the system responded. The transaction feels fair.

Further along sits the personalised recommendation. Spotify’s Discover Weekly serves you a track by an artist you have never heard of, from a genre you didn’t know you liked, and it is exactly right. The song feels like it was made for you. You are delighted, briefly, before a quieter thought intervenes: how did it know? You never searched for this artist. You never explicitly signalled this preference. The algorithm read something in the pattern of your listening, the tempo you favour on Tuesday mornings, the key you gravitate towards when it rains, and it synthesised a prediction so precise that it feels less like a recommendation and more like mind-reading.

At the far end of the spectrum sits the ad that knows too much. The one that makes you put your phone down and stare at it, as if it might be listening. Because maybe it is.

This is the uncanny valley of marketing. Not a binary. A gradient. And the line between “helpful” and “unsettling” is thinner than most marketers realise.

The Famous Cases

You probably know the Target story. It has become something of an origin myth for this conversation.

In 2012, a man in Minneapolis walked into a Target store and confronted the manager. His teenage daughter had received coupons for cribs, baby clothes, and prenatal vitamins. She was, as far as he knew, a high school student. “Are you trying to encourage her to get pregnant?” he demanded. The manager apologised. The father, it turned out, was the one who had to apologise. His daughter was, in fact, pregnant. Target’s predictive analytics engine had identified her pregnancy from shifts in her purchasing patterns, changes so subtle that neither she nor her father had noticed them. The algorithm had detected the behavioural signal before the human had acknowledged the reality.

The story is unsettling precisely because it works as intended. The system did exactly what it was designed to do. It identified a life event, predicted a need, and served relevant offers. In marketing terms, it was a perfect execution. In human terms, it was a violation.

Spotify’s Discover Weekly has a different texture. It doesn’t feel like surveillance so much as telepathy. The playlist arrives every Monday, and week after week, it contains songs that feel uncannily right. Not just songs you like, but songs you didn’t know you would like. The algorithm has mapped the topography of your taste so precisely that it can predict where your ears will go next, even when you cannot. It is flattering and faintly disturbing in equal measure.

Amazon’s “customers who bought this also bought” engine is perhaps the most commercially successful example. It generates an estimated 35 percent of the company’s revenue. It works because it leverages a form of social proof that feels personalised without being invasive. It says, not “we know you,” but “people like you.” That small linguistic shift makes a significant psychological difference. It is personalisation at arm’s length, and it is effective precisely because it maintains that distance.

Then there are the stories people tell each other at dinner parties, the ones that blur the line between anecdote and urban legend. You mention a product in conversation, and an ad for it appears on your phone within hours. You think about booking a holiday, open Instagram, and there it is: a promoted post for exactly the resort you had been considering. The explanations are usually mundane: someone else on your Wi-Fi searched for it, or the algorithm predicted the seasonal intent, or confirmation bias is doing the heavy lifting. But the feeling persists. The machines are listening. They know.

The Paradox of Accuracy

Here is the paradox that most marketers have not fully reckoned with.

The better AI gets at predicting you, the more it reveals about how much it knows. And the more it reveals, the more uncomfortable you become. Accuracy breeds suspicion.

This is not a technical problem. It is a psychological one, and it runs deeper than most marketing literature acknowledges.

Human beings operate on an implicit social contract with the brands and platforms they interact with. The terms of that contract are, roughly: I will give you some information about myself, and in exchange, you will use it to make my experience marginally better. The contract holds as long as both sides stay within the boundaries of what feels proportional. A clothing retailer knows my size and suggests similar items. A music app knows what I have listened to and suggests adjacent artists. The exchange feels balanced. The knowledge feels earned.

But when the output of personalisation becomes so precise that it could only come from a depth of surveillance that exceeds what the consumer consciously agreed to, the contract breaks. The consumer did not sign up to have their pregnancy predicted from shifts in their shopping basket. They did not consent to having their ambient conversations parsed for purchase intent. They agreed to cookies. They did not agree to clairvoyance.

The paradox is that the consumer often benefits from this level of precision. The predicted pregnancy came with genuinely useful coupons. The Spotify playlist was genuinely enjoyable. But benefit and comfort are not the same thing, and when they conflict, comfort tends to win. People will choose a slightly worse recommendation that feels respectful over a perfect recommendation that feels intrusive. This is irrational, commercially suboptimal, and entirely human.

Different People, Different Valleys

Here is where it gets interesting, and where most generic advice about “finding the right balance” falls apart.

Not everyone has the same uncanny valley threshold. The point at which personalisation shifts from helpful to creepy is not a fixed line. It varies, sometimes dramatically, based on the psychological profile of the person receiving it.

Some people are delighted when an algorithm “just gets them.” They experience precise personalisation as a form of being understood, even valued. When Spotify nails their taste, they feel seen. When Amazon surfaces the exact product they were about to search for, they feel efficient. These are people whose relationship with technology is fundamentally collaborative. They see the algorithm as a partner, not a spy. They are comfortable with data exchange because the value they receive in return feels worth the vulnerability.

Others experience the same level of precision as surveillance. For them, every eerily accurate recommendation is a reminder that they are being watched, analysed, and predicted. The more accurate the prediction, the more violated they feel. It is not that they don’t appreciate relevance. It is that relevance, when it arrives without an obvious and proportionate cause, triggers a threat response. Someone knows something they shouldn’t. That is not a service. That is an intrusion.

Some people don’t notice at all. Personalisation operates below their threshold of awareness. They see ads, they click or they don’t, and they never pause to wonder how the system knew. Their uncanny valley is effectively infinite because they never perceive the mechanism behind the output.

And some people actively resent it, not because the personalisation is creepy, but because it is reductive. They bristle at being categorised, at being reduced to a data profile, at having their complexity flattened into a segment. For them, the problem is not surveillance. It is oversimplification. The algorithm thinks it knows them, and it is wrong, and the wrongness feels like an insult.

The implication for marketers is significant. A single personalisation strategy, applied uniformly, will delight some customers, unsettle others, and bore the rest. The uncanny valley is not a universal constant. It is a personal threshold, and it shifts based on personality, context, and the individual’s psychological relationship with technology.

There is a concept in user experience design called the “consent gap.” It describes the distance between what a user has technically agreed to and what they understand themselves to have agreed to.

Every time you accept a cookie policy, you are technically consenting to data collection. The terms are there, in the privacy policy that nobody reads. Legally, the consent is valid. Psychologically, it is meaningless. You clicked “accept” because the alternative was a degraded browsing experience, not because you made an informed decision about how your behavioural data would be used to train predictive models.

The consent gap widens as personalisation becomes more sophisticated. When the output of data collection is a vaguely relevant banner ad, the gap is easy to ignore. When the output is a coupon for prenatal vitamins sent to a teenager who hasn’t told her family she’s pregnant, the gap becomes a chasm.

This is the point where abstract consent collides with visceral experience. Consumers know, in theory, that their data is being collected. They have agreed to it, technically, dozens of times. But they have not agreed to the emotional experience of being predicted with unnerving accuracy. They consented to tracking. They did not consent to the feeling of being known.

And that feeling, that visceral response to a prediction that seems to come from nowhere, is what drives the uncanny valley response. It is not rational. It is not proportionate. It is not even consistent: the same person who is delighted by a perfectly curated playlist on Monday may be disturbed by a perfectly targeted ad on Tuesday, depending on context, mood, and the perceived intent behind the recommendation.

What Marketers Should Actually Do

So where does this leave us? If the uncanny valley of marketing is real, and if it varies by individual, and if crossing it doesn’t just lose a sale but erodes the trust that makes future sales possible, what is the strategic response?

First, understand that less personalisation can be more effective. This is counterintuitive in an industry obsessed with precision, but it is commercially sound. A recommendation that is 80 percent accurate and feels helpful will outperform a recommendation that is 99 percent accurate and feels invasive. The marginal gain in accuracy does not justify the marginal loss in comfort. Deliberate imprecision is not a failure of the algorithm. It is a strategic choice to prioritise trust over optimisation.

Second, make the mechanism visible. One of the reasons Amazon’s “customers like you” framing works is that it explains itself. It says, not “we have been watching you,” but “people with similar patterns chose this.” The same data, reframed as collective intelligence rather than individual surveillance, feels fundamentally different. Transparency does not require revealing the entire algorithm. It requires providing a plausible, non-threatening explanation for why the consumer is seeing what they are seeing.

Third, give people control that actually works. Not buried settings pages with opaque toggles. Real, accessible, immediate control. The ability to say “I don’t want this kind of recommendation” and to have that preference honoured instantly. Control is the antidote to the uncanny valley because it restores the sense of agency. The consumer is no longer a passive subject of prediction. They are an active participant in the exchange.

Fourth, know your audience. Not just their demographics and behaviours, but their psychological relationship with technology. Some segments will welcome deep personalisation. Others will recoil from it. The same strategy applied uniformly across both groups will alienate the second while barely registering with the first. The uncanny valley is not a fixed line on a chart. It is a moving target that requires the same level of understanding that we apply to any other consumer insight.

Fifth, and perhaps most importantly, recognise that in an AI-mediated world, trust is the only currency that matters. The models are getting more powerful. The data is getting richer. The predictions are getting more precise. Every one of these advances makes it easier to cross the uncanny valley, and every crossing makes it harder to win back the ground you have lost. The brands that will thrive in this environment are not the ones with the best algorithms. They are the ones that understand where the line is, and have the discipline to stay on the right side of it, even when the algorithm is screaming that more precision means more revenue.

The Commercial Risk

Let me be blunt about the commercial stakes, because this is not an abstract philosophical debate. It is a balance sheet issue.

Cross the uncanny valley and you do not just lose a sale. You lose trust. And trust, once lost in the context of data and surveillance, is extraordinarily difficult to rebuild. A consumer who feels that a brand has overstepped the boundaries of acceptable personalisation does not simply switch to a competitor. They tell other people. They post about it. They warn their friends. The reputational damage compounds faster than the revenue gain.

In the student accommodation and build-to-rent sectors where I operate, this is particularly acute. Our prospective residents are not anonymous clicks. They are young people making one of the most significant financial decisions of their lives. If our marketing feels like it knows too much, if the personalisation crosses from helpful into invasive, we do not just lose a booking. We lose the trust of an entire cohort, and in a world where those cohorts talk to each other constantly, on group chats, on TikTok, in WhatsApp threads, that trust deficit spreads faster than any campaign can contain it.

The uncanny valley is not a theoretical concern for futurists. It is here, now, and the brands that ignore it do so at their peril.

The Way Forward

The answer is not to abandon personalisation. That would be like responding to the uncanny valley in robotics by building only cartoon characters. The technology works. The predictions are valuable. The consumer, when treated with respect, genuinely benefits from relevance.

The answer is to personalise with restraint. To treat the uncanny valley not as a line to be pushed past, but as a boundary to be respected. To understand that the goal of marketing is not to demonstrate how much you know about someone, but to make them feel understood without making them feel watched.

That distinction, between understanding and surveillance, is the entire game. And in an age where artificial intelligence can know you better than you know yourself, the brands that hold that distinction will be the ones that survive.

The machines are getting better at knowing us. The question is whether we are getting better at deciding what to do with that knowledge.

David Chadderton spent his twenties and thirties teaching people how to make life-or-death decisions at forty thousand feet. He now applies the same principles to consumer psychology, which, depending on the brief, can feel equally high-stakes. He’s the creator of the STAR Framework and the author of The STAR Framework: Rewriting the Rules of Consumer Engagement (NYC Big Book Award 2025), The STAR Operating System: Decode Mindset, Understand Motivation, Transform Human Behaviour, and Dear Algorithm, It’s Not Me, It’s You. By day, a Chief Marketing Officer. By night, a behavioural science obsessive who writes The Unoptimised Human because he can’t stop thinking about why people do what they do.