The personalisation paradox: When brands know too much

And there it was... GPT-5, OpenAI’s shiny new release. Power users hailed its coding upgrades, casual users got modest quality-of-life tweaks, and many in between raged about losing model choice. The GPT-5 model has raised the roof without raising the floor. But that’s the story with AI: progress often means more - more capability, more automation, more intelligence. What it doesn’t always answer is whether more makes things better. Which brings us back to the old question: just because AI can, should it?

Where this question bites hardest is marketing personalisation. Done well, AI can make every interaction feel like it was crafted just for you. But the line between delight and discomfort is thin. Cross it, and the magic fades. What felt personal suddenly feels a little… creepy.

Right time, right place, not right up in your face

Sugar’s great when you want a sweet treat, but you don’t want it on everything all the time. Personalisation works the same way. People don’t crave the maximum dose; they want the appropriate one. That means the message respects the context, the channel, and the stage of the relationship.

Marketers should ask: Would a reasonable person feel helped by this message, in this channel, at this moment, given what the brand has earned the right to know?

When the value exchange is clear, people lean in. They see utility, feel some control, and stay open to future suggestions. But when data use feels opaque or oddly intimate, they pull away. The same message that looks fine on a dashboard can feel intrusive on a lock screen if trust hasn’t been earned, pushing marketers into the ‘Uncanny Valley’ of customer engagement.

Marketing’s uncanny valley 

Marketers usually meet the Uncanny Valley when three forces converge:

• Timing is too sharp, which suggests the brand is watching more closely than the user expected. 
• The inference goes beyond the social norms of the channel. 
• The brand cannot explain the why in plain language. 

For example, picture a visitor who browses credit services and receives a push notification minutes later that hints at a private detail, such as unpaid debt. Accuracy may be high, but trust, not necessarily. The remedy is not less intelligence. The remedy is better calibration of pace, tone, and explanation so intelligence feels like consideration, not surveillance.

Core risks of over-personalisation

The first casualty is trust. Overfamiliar touches spend what you might call ‘consent capital.’ Each unsettling message nudges up opt-outs and spam complaints, quietly shrinking the audience future programs can reach.

Next come the filter bubbles. Marketers often run into the explore-exploit conundrum: should the system keep serving what worked yesterday or take a chance on something new? When it overfits to yesterday’s clicks, customers stop seeing variety. They keep getting more of the same, never discovering new categories, seasonal ideas, or higher-value bundles. Growth stalls because curiosity is never sparked.

Layered on top is voice fragmentation. Too many one-to-one variations fracture tone across segments and channels. A brand that sounds different everywhere risks becoming less memorable and less trusted.

Beyond perception, the risks turn structural. Skewed data and proxy signals can underserve cohorts or gate offers in ways that feel arbitrary. Even with neutral intent, the outcome can still feel unfair. And complexity makes programs fragile: brittle rules, weak data contracts, and model drift can silently erode performance, creating more leakage than lift.

Finally, there’s the outside view. The regulatory and reputational risk. Surfacing sensitive inferences about health, finances, children, or life events without clear consent invites scrutiny that lasts far longer than any campaign.

A human-in-the-loop framework

AI in marketing should be more like a good concierge. Helpful when you need it, invisible when you don’t. The goal isn’t to predict every move or strong-arm a conversion; it’s to make life easier while respecting boundaries. Think intent over identity: knowing when to say something useful, and when to just say nothing. Success isn’t only clicks; it’s satisfaction, lifetime value, and a voice people recognise even without the logo.

• Purposeful data, living consent: Start with a real customer problem, then only gather the data you need to solve it. Keep consent simple, specific, and easy to change. Retire signals that don’t add value. A utility that feels earned always beats intimacy that feels assumed.
• Context-aware orchestration with built-in variety: Match depth of personalisation to the channel and stage of the relationship. Go light in ads, deeper in owned channels once trust is earned. Keep a small quota for exploration so people don’t get stuck in filter bubbles, and set tone guardrails so the brand sounds like itself, no matter who’s typing.
• Build a big red stop button: Test the brakes before you hit the gas. Run preflight checks for sensitive campaigns, watch early warning metrics, and always have a safe mode that falls back to simple, helpful messages when confidence is low. And whatever you send, explain in plain language why someone got it, plus an easy way to say no.

So when is personalisation too much, just right, and great?

Toomuch: When a brand oversteps into private territory. Imagine getting recommendations about something you haven’t shared or even fully acknowledged yourself, such as a private health condition. Even if the algorithm is right, it feels wrong… like the brand knows more than it has earned the right to. Accuracy without empathy erodes trust.

Just right: When personalisation is built on clear signals you’ve given, like items in your cart or preferences you’ve set, and the brand stays transparent about how it’s using them. The message feels useful, not invasive. It’s timely, relevant, and respectful.

Great: When personalisation widens horizons instead of narrowing them. It makes recommendations that feel fresh, explains why you’re seeing them, and gives you easy control to adjust. Instead of boxing you in, it sparks curiosity while reinforcing trust.

Earn connections, don’t assume it

In marketing, trust works like in any other relationship: you don’t start by finishing sentences, you start by listening. Over time, with clarity and consistency, you get to do more together.

AI makes it tempting to do more, faster, everywhere. But more isn’t always better. The brands that last will be the ones that pace themselves: clear about why they’re reaching out, careful about what they infer, and at times, willing to leave some things unsaid.

This article is penned by Jacob Joseph, Vice President - Data Science, CleverTap

Disclaimer: The article features the opinion of the author and does not necessarily reflect the stance of the publication.