TikTok, YouTube and LinkedIn are building the same discovery engines

For years, social platforms were distinct ecosystems with their own logic, quirks, ranking systems and cultures. TikTok was TikTok. YouTube was YouTube. LinkedIn was LinkedIn. Meta was Meta. Each had its own personality.

That era is over.

Today, every major platform is quietly converging toward the same underlying discovery architecture — AI-powered, behaviour-driven, multimodal, and emotionally aware.

It’s not obvious if you look at their marketing.

It’s glaringly obvious if you look at their engineering.

The “feeds” we see on TikTok, YouTube Shorts, Instagram Reels and LinkedIn aren’t separate inventions. They’re parallel evolutions — all orbiting the same objective:

Keep users in the app longer by predicting what they’ll want before they know it themselves.

And the way they do that is essentially identical.

Let’s break the myth of platform uniqueness once and for all — and walk through the evidence that TikTok, YouTube and LinkedIn have all built the same system Meta just finished rolling out.

1. The TikTok algorithm

The original blueprint for the Discovery Era

TikTok didn’t invent short-form video.

It invented the discovery engine.

Before TikTok, social platforms were distribution systems. You posted to your followers. Your followers saw your stuff. TikTok flipped the entire model on its head.

TikTok’s For You Page (FYP) was the first mainstream ranking system built entirely on:

• user embeddings
• content embeddings
• behavioural similarity
• predicted interest
• emotional resonance
• rapid feedback loops

TikTok engineered three crucial breakthroughs:

Breakthrough 1: user behaviour as the primary signal

TikTok tracks micro-signals with microscopic sensitivity:

• how long you hover
• which second you drop off
• the type of edit that makes you pause
• the emotional tempo you prefer
• whether you replay once or twice
• what you tap when you’re bored
• what content restores your focus

TikTok knows you better than you know yourself — and it updates your behaviour profile in real time.

Breakthrough 2: content embeddings, not “topics”

TikTok reads content through multimodal models that interpret:

• visuals (scenes, objects, motion)
• audio (tone, sentiment, rhythm)
• text (semantic meaning, intent)
• style (humour, sincerity, intensity)

This means TikTok doesn’t run on trends;

it runs on meaning.

Breakthrough 3: discovery before network

TikTok decoupled distribution from followers completely — the first major platform to do so.

This is exactly the system Meta has now duplicated.

TikTok was the domino that knocked the entire industry into the Discovery Era.

2. The YouTube algorithm

The sleeping giant woke up

YouTube has always been a recommendation-driven platform — but from 2022 onward, something fundamental shifted.

YouTube didn’t just build Shorts to compete with TikTok.

It rebuilt its entire ranking architecture to mirror TikTok’s discovery logic.

Here’s the proof:

Proof 1: YouTube’s 2023 multimodal deep learning upgrade

In 2023, YouTube rewrote its ranking models to include multimodal analysis: video frames, audio patterns, speech transcripts, creator history, channel identity, and viewer behaviour models.

This is the exact same approach TikTok uses.

YouTube openly admitted that its ranking system began prioritising:

• predicted engagement
• emotional response indicators
• behavioural patterns
• user energy and session depth
• personalised interest clusters
  

It moved away from topic-based recommendation and towards behaviour-led forecasting.

Proof 2: viewers watch more “from-nothing” content

YouTube reported that users were increasingly discovering videos from creators they don’t follow.

This is the hallmark of a discovery engine.

Proof 3: Shorts is now a training ground for long-form distribution

YouTube uses Shorts behavioural data to predict what long-form content viewers will want next.

That is exactly what TikTok does with multi-part series.

Proof 4: YouTube replaced “relevance” with “affinity” models

Affinity models are embedding-based AI systems that classify both content and viewers in multidimensional behavioural space — the same structure TikTok pioneered.

YouTube is not unique anymore.

It is operating under the same underlying logic as Meta’s Discovery Engine and TikTok’s FYP.

3. The LinkedIn algorithm

The most surprising transformation of all

LinkedIn used to be a network platform.

It is now a discovery platform wearing a corporate blazer.

Most people haven’t noticed because they’re still posting like it’s 2017 — but the architecture has changed dramatically.

Proof 1: LinkedIn no longer prioritises connections

If you scroll your LinkedIn feed today, half the content you see is from people you don’t follow.

Sound familiar?

LinkedIn now surfaces posts based on:

• your behaviour
• your recent emotional signatures
• your “professional interest clusters”
• semantic categories
• dwell time
• contribution value
  

They stopped pretending the network graph should define the feed.

This is the exact blueprint of TikTok, YouTube and Meta.

Proof 2: LinkedIn uses deep semantic classification (not topics)

LinkedIn publicly revealed that its ranking system identifies:

• the intent behind the post
• the emotional tone
• the format
• the audience likely to benefit
• the behavioural match for each user

This is multimodal classification — identical to Meta’s approach.

Proof 3: LinkedIn’s “quality score” mirrors a relevance score

In 2023–2024, LinkedIn introduced a hidden “quality” layer that penalised:

• vague content
• low semantic clarity
• emotionally mismatched posts
• off-niche rambling
• inconsistent contributor patterns

This is functionally the same as Meta’s relevance scoring, even if LinkedIn uses softer language.

Proof 4: LinkedIn started building user embeddings

This wasn’t a big announcement — it was quietly placed in an engineering blog.

LinkedIn now models users using behaviour vectors:

• time spent
• content type preference
• energy tolerance
• scroll depth
• niche oscillation
• emotional response patterns

This is the same system driving Meta’s Discovery Engine.

LinkedIn is now a business version of TikTok’s recommendation AI.

All platforms are converging on the same architecture

Here’s the simplest way to see the convergence:

TikTok invented the discovery engine.

YouTube upgraded itself into a discovery engine.

Meta rebuilt itself into a discovery engine.

LinkedIn quietly transformed into a discovery engine.

And they are all powered by:

• user embeddings
• multimodal content embeddings
• behavioural prediction models
• interest learners
• emotion inference systems
• semantic similarity ranking
• cross-surface personalisation
• unified organic/paid relevance logic

This is not coincidence.

This is the future of social.

Every platform realises that attention is finite.

Growth depends on prediction accuracy.

Prediction accuracy depends on AI models.

AI models depend on embeddings.

Embeddings require semantic clarity.

Semantic clarity requires consistent creators.

This is the new universal truth.

The implication: Social Discovery Optimisation isn’t a Meta thing — it’s an industry transformation

This convergence means something huge:

SDO isn’t a Meta discipline.

SDO is a cross-platform discipline.

If discovery works the same way everywhere, then the future of growth depends on:

• emotional consistency
• semantic clarity
• predictable brand signals
• behavioural alignment
• funnel-stage sequencing
• relevance strengthening
• keyword coherence
• stable content identity
• recognisable creative patterns

Creators who optimise for these signals will win across platforms because all platforms now reward the same thing.

This is the most powerful moment of opportunity in social media since SEO was invented.

The industry just hasn’t realised it yet.