How AI Recommendation Algorithms Decide What to Show You

You open Netflix and the top row feels eerily personal. You linger on a product on Amazon and it follows you around the internet for a week. YouTube somehow knows you’ll watch that exact video next. It’s not coincidence — it’s recommendation algorithms, some of the most sophisticated AI systems ever deployed, quietly shaping what billions of people see every day. Understanding how AI recommendation algorithms decide what to show you is essential digital literacy, because these systems influence what you buy, what you watch, and increasingly what you believe.

The Basic Idea: Predicting What You’ll Engage With

At their core, recommendation algorithms answer one question: given everything we know about this user, what are they most likely to click, watch, or buy next? They learn from three kinds of data:

• Your behaviour: what you’ve clicked, watched, bought, searched, liked, and how long you lingered. Dwell time is gold — pausing on a post tells the algorithm more than a like.
• People like you: collaborative filtering finds users with similar tastes and recommends what they enjoyed. “People who bought X also bought Y” is the simplest version of this.
• The content itself: content-based filtering analyses the attributes of items — genre, keywords, style, topics — and matches them to your demonstrated preferences.

Modern systems blend all three, plus hundreds of other signals, in deep learning models trained on billions of interactions. The BBC’s technology coverage regularly reports on how these systems evolve — and the controversies they generate.

How the Big Platforms Differ

Netflix and streaming

Netflix’s recommendations drive the majority of viewing on the platform. Its system combines your viewing history with detailed content tagging (every title is tagged across hundreds of attributes — mood, pacing, setting, character types) to predict what you’ll watch next. It even personalises the artwork you see — the thumbnail for a film might emphasise the romance for you and the action for someone else. Our deeper look at AI in streaming services covers this in detail.

TikTok and short video

TikTok’s For You Page is widely considered the most powerful recommendation engine ever built. Its secret: incredibly dense feedback. Every swipe, rewatch, pause, and share is a data point, and with dozens of videos per session, the algorithm learns your preferences with frightening speed — often within a single sitting. This is why new accounts feel generic for an hour, then uncannily personal.

Amazon and shopping

E-commerce recommenders optimise for purchase probability: your browsing and buying history, what similar customers bought, seasonal patterns, and profit margins (yes, the algorithm has commercial incentives — it’s not a neutral adviser). See our piece on how AI is changing online shopping for the full picture.

News and social feeds

Facebook, X, and news apps rank content to maximise your time on the platform. This is where recommendations get controversial: outrage and strong emotion drive engagement, so algorithms can amplify divisive content — not out of malice, but because the optimisation target is attention, not truth or wellbeing.

The Optimisation Problem Nobody Talks About

Here’s the crucial insight: these algorithms don’t optimise for your goals. They optimise for the platform’s goals — usually engagement, watch time, or ad revenue. The system doesn’t care whether the video makes you happy, informed, or anxious; it cares whether you keep watching. This misalignment explains most of the harms attributed to recommendation algorithms: rabbit holes, filter bubbles, doomscrolling, and the amplification of extreme content.

You’re not the customer of these systems — you’re the product being optimised. The sooner you internalise that, the more deliberately you can use them.

Taking Back Some Control

You can’t turn recommendations off entirely, but you can shape them:

• Curate actively. Use “not interested,” “don’t recommend,” and hide options — the algorithm does listen to explicit feedback.
• Break the pattern deliberately. Search for and engage with content outside your bubble; the algorithm follows your behaviour, so change the behaviour.
• Use chronological or following-only feeds where platforms offer them.
• Limit the data. Review ad personalisation settings, limit tracking where possible, and consider opting out of AI training on social media.
• Notice the nudge. Simply being aware that your feed is a constructed, optimised artefact — not a neutral window on the world — is half the defence.

Children and the Algorithm

Recommendation algorithms deserve special attention where children are concerned. Young users are the most susceptible to algorithmic influence and the least equipped to recognise it. A child’s YouTube autoplay can drift from cartoons to disturbing content; TikTok’s For You Page learns a teenager’s vulnerabilities with alarming speed. The harms — from exposure to harmful content to distorted body image and addictive design — are well documented and ongoing.

Practical protections: use YouTube Kids rather than main YouTube for younger children; enable restricted modes (imperfect but helpful); keep devices out of bedrooms at night; and — most importantly — watch with them sometimes. Co-viewing lets you see what the algorithm serves and discuss it in real time. “Why do you think it showed you that?” is a powerful question that builds algorithmic awareness naturally.

For teenagers, the conversation shifts from control to literacy. They’ll encounter algorithmic feeds regardless of your restrictions, so the goal is critical understanding: how the feed works, why outrage spreads, how influencers are incentivised, and what the algorithm doesn’t show them. Our parents’ guide to fake news and what children should learn about AI complement this — together they form a family media literacy toolkit. The algorithm isn’t going away; raising children who understand it is the durable defence.

The Bottom Line

Recommendation algorithms decide what to show you by predicting your engagement from your past behaviour, similar users’ behaviour, and content attributes — optimised for the platform’s goals (your attention), not yours. Understanding this lets you use these systems deliberately rather than being used by them: curate actively, diversify your inputs, limit your data where you can, and never mistake your personalised feed for an objective picture of the world.