Personalization is not a feature. It is a new way of making journalism.

Personalization is one of the most overused words in this industry. Every publisher claims to do it. It shows up in every strategy deck I have ever seen. 

And yet, when I look at what most publishers are actually doing under that label, it tends to be one thing: a recommendation widget based on what you last read. That is not personalization. That is a related-articles module with a better name.

Real personalization in 2026 runs across three dimensions: Format (the medium and shape of content delivery), Content (what the journalism is actually about), and Audience (who your readers are and what they need). 

Publishers doing this seriously are not just building better features. They are rethinking the entire way journalism is produced and distributed.

By the numbers

800K+
Page views for a single YLE investigative story distributed across text, video, audio and social simultaneously

13 years
Since YLE began building personalized news algorithms, and they say topic-based personalization was only ever the first step

1,600+
Audience attributes in IAB Tech Lab's Audience Taxonomy available to publishers for privacy-safe, standardized targeting

Format: The content must fint the reader

The first dimension is the one most publishers have thought about least: not what the content says, but what shape it takes when it arrives. For most of journalism's history the format was chosen before the reporter filed a word. 

YLE's argument, and that of leading publishers in INMA's research, is that this is the single biggest constraint on reaching new audiences. 

The same underlying journalism should take multiple forms: a long read for someone with twenty minutes on a Sunday, an audio summary for a morning commute, a short vertical video for someone arriving from social.

At INMA World Congress in Berlin, Mika Rahkonen, Head of Strategy at YLE, and Riina Malhotra, Head of Newsroom, presented their work on "liquid content". 

Their argument: the article is no longer the atomic unit of journalism. Information is.

YLE and the Liquid Content Model

Mika Rahkonen told Berlin that YLE was building its first personalized news algorithm thirteen years ago, understanding that topic-based personalization was only the first step. What they now call "liquid content" goes further: content that adapts its form to the person consuming it.

The image is deliberate. Liquid takes the shape of whatever container it occupies. As Bruce Lee said: "Be water my friend." Journalism should do the same. The underlying reporting - the interviews, the data, the verified facts, the editorial judgment - is the substance. The format is the container, determined by the reader's context, not the production workflow.

"The contents find their audience. The audience don't have to find the content. There is less friction with the customer's schedule, not ours."
- Mika Rahkonen, Head of Strategy, YLE

Riina Malhotra brought this from philosophy to practice. 

A months-long investigation into police violence in Finland was atomized: multiple articles, videos, social media content, radio features, broadcast segments. More than 800,000 page views, 1.4 million Finns watching TV coverage, over three million social media impressions. 

The next challenge is making the workflow systematic and scalable.

IAB Tech Lab's Content Taxonomy, updated in late 2024, allows publishers to signal content type and format through programmatic channels in standardized terms. A vertical video story carries meaningfully higher CPMs than a standard display placement -- getting taxonomy signaling right is the difference between your format investment being visible to buyers or invisible.

What format personalization requires in practice

Modular content architecture. 
Build production workflows around information units - facts, context, data points, quotes, core narrative - that can be assembled differently for each delivery context. Separating story substance from format is an editorial design change before it is a technology change.

Automated format versioning. 
Versioning must not be a journalistic job. AI-assisted versioning - summarizing, converting text to audio scripts, adapting for different screen orientations - frees journalists to do journalism, not reformat it.

Adaptive delivery infrastructure. 
Detecting device type, connection speed, and viewing context and serving the appropriate format automatically has direct commercial impact - from adapting a video player for mobile to deciding when to push a notification.

IAB Content Taxonomy alignment. 
Aligning format signals with IAB Tech Lab's Content Taxonomy connects editorial format decisions to programmatic demand. For publishers investing in video or audio, this is the plumbing that makes the investment legible to buyers.

Content: The right story for the right person

Content personalization is the dimension most publishers have at least started on - but most have started on a narrow version. A "you might also like" widget is not content personalization. It is a traffic recirculation tool.

INMA's Jodie Hopperton, in "Best Personalisation Practices for News Media," separates personalization into two types. Active personalization is where users explicitly tell you what they want: topics, journalists, verticals, newsletter preferences. Users who have actively declared preferences generate higher engagement and better retention. 

Passive personalization infers preferences from behavior - it works across a much broader audience but requires a robust data layer; without it, it defaults to amplifying whatever is already popular.

The Empathy Gap

Behavioral data tells you what someone consumed. It does not tell you what they needed. 

Some publishers are experimenting with counter-personalization: proactively surfacing content readers are unlikely to have sought out - a counterargument, a different perspective, a story from a beat they have never visited.

Mika Rahkonen put the editorial responsibility plainly: "AI can merge. AI can summarize. AI can optimize. But AI cannot decide what matters." 

The curation judgment - what to emphasize, what a person needs to know even if they did not know they needed to know it - is the irreducible human function in journalism.

Personally I would take the "Content" aspect one step further. When I studied media design at the university the optimal design delivery was the "expected unexpected", where what you deliver isn't what they asked for but what they actually need. 

I would argue this is the personalization we need to deliver.

This requires a deep understanding of your audience - what others similar to the individual consume that might be relevant by unexpected vectors. I think Meta is world class at this, driving engagement in a way that is almost scary.

By segmenting users in cohorts you can transfer interests between users anonymously and drive increased engagement to build loyalty and ultimately sell more subscriptions. 

And as trust builds, you can execute more personal content experiences that require more user information, and ask for this information accordingly.

What content personalization requires in practice

A progressive audience data profile. 
Start with the minimum data point that creates genuine value, earn trust, and expand the profile over time. Ask for topic preferences before demographic information. Each step should feel worth it to the reader, not like a toll.

Editorial guardrails in recommendation logic. 
Build rules into your recommendation systems: minimum topic diversity, mandatory inclusion of public interest journalism, limits on how much any single topic can dominate. An engine with no editorial rules is an engagement maximizer, not a journalism product.

Audience understanding as an editorial skill. 
Every journalist and editor needs to understand who they are writing for and why that person should care. Publishers who do this well have analytics-literate journalists and data-literate editors - not just a data team that sends weekly reports nobody reads.

Metrics that measure need-satisfaction. 
Did this reader come back? Did they convert? Did they share? Did they complete the story? These are better proxies for genuine value than time on page and click-through rates.

Audience: Knowing who you are actually talking to

Most publishers know their average reader. Very few know their actual readers.

The average reader is a composite fiction from aggregate analytics. The actual readers are individuals with jobs, ages, locations, interests, and loyalty. 

The gap between publishing for a demographic and publishing for a person is the gap between knowing the average and knowing the actual.

The infrastructure runs through first-party data. I covered this extensively in the last edition of The Samhub Signal. Without a registered user base, a consent-compliant data layer, and the infrastructure to activate that data, audience personalization is built on sand. 

Knowing that a significant segment of your audience is under thirty-five and arriving from social media is not just an advertising insight - it is a product architecture insight about navigation, content density, and the loyalty behaviors needed to turn visitors into subscribers.

I think that in order to create a seamless experience that caters fully to the individual, understanding your audience is not an option. It is an imperative.

The IAB Tech Lab Infrastructure

IAB Tech Lab's Audience Taxonomy provides more than 1,600 standardized attributes for describing audience segments - the shared language that allows a publisher's first-party knowledge to be understood by programmatic buyers. 

The Seller-Defined Audiences (SDA) specification lets publishers define audience cohorts from first-party data and communicate them to buyers through OpenRTB without passing any personal identifiers. 

Audience Taxonomy v1.1 added privacy safeguards minimizing the risk of sensitive inferences - meaningful for publishers under GDPR.

I have personally seen editorial and commercial teams describe the same audience in completely different ways. If both sides view the audience through the same framework, it has the potential to uncover monetization opportunities and synergies between the two.

What audience personalization requires in practice

An identity strategy before a segmentation strategy. 
You cannot segment an audience you cannot identify. The first priority is a registration architecture with a meaningful logged-in user base. The value exchange - what does the reader get for registering? - must be designed as carefully as any editorial product.

Both active and passive audience signals. 
Declared preferences - subscription type, newsletter selections, explicit topic choices - are your most reliable signals. Behavioral signals - reading patterns, device habits, referral sources, time-of-day behavior - fill in the picture for the majority who have not actively declared preferences.

IAB Audience Taxonomy alignment. 
If your audience segmentation lives in proprietary internal language, it cannot be efficiently communicated to buyers. Using the IAB Tech Lab Audience Taxonomy as the shared vocabulary connects editorial knowledge to commercial activation through Seller-Defined Audiences.

Audience data driving product decisions, not just ad targeting. 
Use audience knowledge to make decisions about product architecture, content investment, format priorities, and distribution strategy. The question to keep asking: what does knowing this about our audience change about how we make journalism?

The harder truth: Why the real obstacle is not technology

The technology exists. The standards are in place. The case studies are compelling. 

So why are most publishers still treating personalization as a product feature rather than a fundamental rethinking of journalism?

"Liquid content fails because newsrooms are still built to produce articles, not information that can flow."
- Riina Malhotra, Head of Newsroom, YLE

Most newsrooms are structured around formats: print desk, digital team, video unit, social team. Content moves through separate silos. 

Adaptations are afterthoughts, created by different people in different systems, without coordination with the original reporter. Real personalization requires a different model: one where the information asset, not the article, is the primary output. Where "who needs this and in what form?" is asked at the start of production, not the end.

YLE's prescription: Format follows need - start every editorial decision with what the person needs to know, not what format to produce. Versioning is infrastructure, not labour. Every hour a journalist spends reformatting is an hour not spent reporting.

Audience understanding belongs in the newsroom. Every journalist and editor should understand who they are making journalism for; this changes hiring, onboarding, and daily editorial conversations.

The technology is the easy part.

What this actually means

Personalization in 2026 is not a feature on a product roadmap. It is a different answer to the oldest question in journalism: who is this for, and why does it matter to them.

What was different about INMA Berlin was not that anyone said something I had never heard before. It was that the practitioners on stage - journalists and newsroom leaders, not product managers - were describing personalization as an editorial imperative rather than a commercial one. 

Mika and Riina were not talking about yield optimization. They were talking about reaching people that journalism currently fails to reach, and about what it means for public discourse when those who most need good information cannot find it in a form that works for them.

That reframing - from commercial opportunity to journalistic obligation - is the shift that makes personalization genuinely possible to implement. Because once it becomes a question of what journalism is for, the organizational changes that feel expensive in a product conversation start to feel necessary in an editorial one.

That is what personalization actually means in 2026.