For several years I built Retargeting with Rareș Bănescu, a marketing automation and personalisation platform for online stores. Our job, in practice, was to decide what a visitor sees: which product to show, which email to send, at what moment. So when I read today about “the algorithms that control our lives”, I don’t see a mysterious black box. I know what they look like on the inside, at least their smaller, better-behaved cousins.
That’s exactly why I think this conversation deserves to be had calmly. Algorithms are neither demons nor neutral. They are tools built by people, chasing a goal chosen by people. The real question isn’t whether they influence what we see online, because they certainly do, but who picks the goal and in whose interest.
How it works, briefly and without jargon
Every recommendation system does essentially the same thing: it collects signals and tries to guess what you’ll do next. The signals are mundane. What you clicked, how long you stayed on a page, what you scrolled past quickly, what you bought, what people similar to you did. From these, the system scores every possible piece of content and puts the highest-scoring one in front of you.
The part that matters isn’t the maths, it’s what the score is trying to maximise. In an online store, the goal is usually an order. On a social network or a video platform, the goal is most often attention: how long you stay, how many reactions you leave, how often you come back. I’ve written before about the principles behind e-commerce personalisation that delivers real recommendations. Here, though, the stakes are different.
Why optimising for engagement rewards outrage
An algorithm that measures reactions can’t tell a good reaction from a bad one. An angry comment counts as much as an enthusiastic one, or even more, because anger breeds more angry comments. Content that outrages us, frightens us or confirms that “the other side” are idiots is, sadly, exactly the content we share fastest.
Nobody wrote “promote polarisation” into the code. But if you ask a machine to maximise time spent on a platform and let it learn on its own what works, it will discover quite quickly that strong emotions keep people glued to the screen. I’ve seen this from the opposite side too: with Știri de Bine (a Romanian positive-news site I ran) we spent years trying to make room for good news, and I wrote about it in what I learned about positive journalism. Good news is harder to sell than scandal, and the algorithms learned that lesson faster than we did.
The bubble we live in without knowing it
The second problem is subtler. If the system always shows you what you already liked, you end up seeing more and more of the same. Your opinions look like the majority view, because everyone in your feed thinks alike. People who think differently seem either non-existent or caricatures. This is what’s been called the “filter bubble”: not censorship, but a slow narrowing of the world, done with the best personalisation intentions.
Video platforms have faced growing criticism in recent years that automated recommendations push users towards ever more extreme content in order to keep them watching. Early in 2019, YouTube announced it was changing its recommendations to reduce such content. I don’t know how effective the change is, but the fact that it was needed says enough.
When personalisation helps and when it manipulates
It would be hypocritical of me to say personalisation is bad in itself. I made a living from it. If you visit a diving gear shop and get recommendations for masks and fins rather than evening dresses, everyone wins: you find what you need faster, the shop makes a sale. An email reminding you of an abandoned cart is useful, not sinister.
For me, the line is fairly clear. Personalisation helps when it serves you, the user, and when you’d agree to it if you knew exactly how it worked. It becomes manipulation when it exploits weaknesses: fear, insecurity, anger, compulsion. Showing someone a relevant product is good service. Showing an undecided voter a message built specifically to scare them, based on a psychological profile they know nothing about, is something else entirely.
What Cambridge Analytica and GDPR taught us
The Cambridge Analytica scandal in 2018 was the moment the wider public learned that data from a social network could end up, via seemingly harmless apps, with firms running political campaigns. The data wasn’t just about what we buy, but about who we are.
The same year GDPR came into force, and I wrote a practical GDPR guide for online stores. The core idea is sound: your data is yours, you should know what’s done with it, and you should be able to say no. Also in 2018, Facebook announced it was changing its News Feed to favour posts from friends and family over content from pages and publishers. And last month Twitter announced it would stop running political ads. These are signs that the platforms feel the pressure. Whether the changes are substantial or cosmetic remains to be seen.
What we can do as users
We can’t rewrite the algorithms, but we don’t have to let them do all the work. A few habits that genuinely make a difference:
- Diversify your sources. Read a few publications you respect directly, including one you disagree with, instead of waiting for them to show up in your feed.
- Use the chronological feed where it exists. Twitter, for example, lets you see the latest tweets instead of the “top” ones.
- Follow people, not just pages that churn out outrage. And unfollow, guilt-free, anyone who systematically winds you up.
- Before sharing something that made you angry, wait a minute. That split-second reaction is exactly what the system is betting on.
- Check your privacy and ad settings now and then. You’ll be surprised how much the platforms know about your interests.
What platforms and regulators should do
Responsibility can’t rest with users alone. Platforms should explain, in plain words, why a particular post or ad appears in front of you, and give you real control: the option of a chronological feed, the ability to switch off personalisation, the ability to see the profile they’ve built of you. For political ads, transparency should be the bare minimum: who paid, how much and whom they targeted.
Regulators, for their part, face a hard balance. Rushed regulation can stifle innovation and end up favouring the big players who can afford the lawyers. No regulation leaves everything to the goodwill of companies that, naturally, have their own interests. GDPR showed that Europe can enforce rules. The next step would be transparency about algorithms, not just about data.
What we owe our customers, those of us who use personalisation
For entrepreneurs who use recommendations, automation and retargeting, and there are more of them every year, my rule is simple: only personalise in ways you wouldn’t be embarrassed to explain to the customer face to face. Tell them what data you collect and why. Make it easy to opt out. Don’t use fake urgency and fear as tactics. I’ve written before about what artificial intelligence in marketing can concretely do; that power comes with matching responsibility.
In the long run, trust is the most valuable asset of any online business. A customer who feels manipulated doesn’t come back. One who feels understood does.
Algorithms will keep getting better at guessing what we want. The question is whether they’ll be used to serve us or to keep us captive. How do you choose what you read online? Do you let the feed carry you, or do you have your own rules?