Content Analytics Beyond Pageviews: Labels and User Needs

Pageviews tell you how many times a page loaded. They do not tell you whether anyone read it, why they came, or whether the piece did the job it was meant to do. To answer those questions, publishers label every piece of content by what it is about (topic), how it is presented (format) and what it does for the reader (user need), then analyse performance by label rather than article by article.

This guide covers why pageviews mislead, how to design a labelling scheme, how the User Needs model works, how to get labels into Google Analytics 4 and Chartbeat, and how to turn the reports into editorial decisions.

Key takeaways

• Pageviews measure page loads, not attention or value, and they reward publishing volume and lucky distribution as much as quality.
• Labelling content consistently by topic, format and user need lets you compare like with like across thousands of stories.
• The User Needs model, which grew out of BBC audience research and was later expanded to eight needs by smartocto and Dmitry Shishkin, gives newsrooms a shared vocabulary for why people read.
• In Google Analytics 4, labels become event parameters registered as custom dimensions; in Chartbeat, one option is to pass them as extra page-level section values.
• The most useful analysis compares each label's share of output with its share of engagement, then tests changes before rolling them out.

Why pageviews alone mislead

A pageview is recorded whether a visitor reads every word or leaves after two seconds. In a 2014 piece for TIME, Chartbeat's then chief executive, Tony Haile, reported that across 2 billion visits in a month, 55% of visitors spent fewer than 15 seconds actively on a page. He also found that among topics with similar traffic, the best performers captured about five times the attention of the worst, and that there was no relationship between how much a piece was shared and how much attention it received.

Pageviews also have structural biases:

• Volume. A desk publishing fifty short updates a day will usually out-pageview a desk publishing two explainers, even if each explainer is read far more closely.
• Distribution. Traffic swings with search, social and aggregator referrals that have little to do with the quality of the story.
• Purpose. A short service piece that answers a reader's question quickly may be a complete success with modest time on page.

Label-level analysis exposes patterns that article rankings hide. In one BBC World Service case study, BBC Russia was producing 70% "update me" stories that brought in only 7% of its page views, according to Journalism.co.uk.

No single metric is enough; labelling lets you read them against what each piece was meant to do.

Label content by topic, format and user need

A report by FT Strategies and smartocto on the User Needs model makes the case that user needs work best combined with well-thought-through topics and formats: the primary user need sets the purpose of a piece, the topic defines its subject, and the format determines how it is presented.

Practical rules for a scheme that holds up in reports:

1. Use controlled values. Every label comes from a fixed list with a definition and examples. For topics, a standard such as IPTC Media Topics is a sound starting point; see IPTC Media Topics and News Taxonomies Explained.
2. Give each piece one primary user need. An InPublishing feature on user needs advises identifying one need per article, so the team is clear about each piece's purpose. Secondary labels can sit alongside it.
3. Label at the source. Labels should be set in the CMS when content is created, so every downstream system receives the same values.
4. Backfill the archive. Without historical labels you cannot compare this quarter with last year. AI can label a back catalogue quickly, but check its accuracy against a hand-labelled sample first, as described in Evaluating AI Tagging: Precision, Recall and Human Review.
5. Version the scheme. When a label is added, split or retired, record the date so trends are not misread.

How to Build a Content Taxonomy covers designing the lists themselves.

The User Needs model: origins and version 2.0

The User Needs model grew out of BBC research into audience behaviour; Journalism.co.uk dates the BBC's user needs strategy to 2016. In December 2018 it reported that BBC World Service's 41 language services categorised articles by six needs: update me, give me perspective, educate me, keep me on trend, amuse me and inspire me. Dmitry Shishkin, then the World Service's digital development editor, described how his team analysed three months of an established service's content, and how journalists began pitching the user need along with the story.

User Needs Model 2.0, later created by smartocto and Shishkin, sorts eight needs into four groups, the drivers of why people consume news: to know (fact-driven), to understand (context-driven), to feel (emotion-driven) and to do (action-driven). Each group holds two needs; the table follows the grouping in Journalism.co.uk's guide to the model. The FT Strategies and smartocto report illustrates each need with a headline about a single news event; the examples below follow the same idea using a storm.

Newsrooms can adapt the model. InPublishing reported that the Bureau of Investigative Journalism works with user needs tailored internally to its mission, with first-person newsletter pieces falling under an "involve me" need. Adapting is fine if the labels stay consistent over time and each maps to a clear purpose.

Connecting labels to analytics tools

Labels are only useful in analysis if they reach the analytics tool with every page view, using the same values as the CMS.

Google Analytics 4

• Send the parameters. Add the labels to the page's tagging, either as event parameters in Google Tag Manager (typically read from a data layer variable) or in gtag.js. Google's developer documentation notes that parameters defined with gtag('set') are applied to all subsequent events on the page, so the call must run before the events fire.
• Register them. Custom parameters must be registered as custom dimensions in the GA4 interface; without that, the data is collected but not available in reports. Reporting can take 24 to 48 hours to start.
• Choose the scope. Content labels describe the page, so they are usually event-scoped. A standard property allows 50 event-scoped custom dimensions; a 360 property allows 125.
• Avoid high cardinality. Google warns against unnecessary high-cardinality dimensions such as unique IDs or timestamps, which can cause data to be grouped under "(other)". Topic, format and user need, drawn from short controlled lists, are low-cardinality by design.
• Use text codes. GA4 treats values that look like numbers as numbers, so use readable codes such as educate_me rather than 3.

Chartbeat

Chartbeat reads content metadata from the page's _sf_async_config settings.

• Sections. The sections variable accepts one or more comma-separated values, up to 10 per page and 63 characters each. Chartbeat uses the "|" character internally when splitting values, so avoid it in labels. Section values drive the section filters in Chartbeat's dashboards.
• Type. The type variable takes a single page-type value, such as 'article page' or 'gallery page', and is meant for teams using Chartbeat's Advanced Queries or Data Lab.
• Adding labels to sections. One option, a publisher convention rather than a documented Chartbeat feature, is to add format and user-need labels as extra section values alongside the topic, named consistently so each kind of label stays distinguishable.

A simplified example of the labels for one article, written into the page by the CMS:

// Chartbeat (one option): labels as extra section values
_sf_async_config.sections = 'Politics,Explainer,Need - Educate me';
_sf_async_config.type = 'article page';

// GA4: set before the events that should carry the labels
gtag('set', {
  content_topic: 'politics',
  content_format: 'explainer',
  user_need: 'educate_me'
});

Check the result in each tool's real-time or debug view before relying on the reports.

Turning insights into editorial decisions

Start with a simple comparison for each label: its share of output against its share of attention and conversions. The BBC finding above, where a large share of output brought a small share of traffic, is the classic pattern.

Publishers have reported concrete results from this kind of rebalancing. InPublishing described Metro UK achieving a year-on-year increase in page views while cutting the volume of content in its entertainment section by 30%, and a small, specially trained team at Berlingske Media producing 4% of content but accounting for 30% of conversions. In the FT Strategies and smartocto User Needs Lab, the Serbian news site Blic found that 80% of its poorly performing context stories in one section had negative emotions in their headlines, and changed its headline approach.

The same report adds cautions: small samples can mislead, shifts in traffic-acquisition channels can distort results, so data needs regular re-evaluation, and meaningful metrics such as average article reads (or the median, as samples grow) deserve the focus.

A workable routine:

1. Review label-level performance monthly with editors, not only the analytics team.
2. Pick one or two experiments per quarter, such as more explainers on a key topic.
3. Define success before starting, in the metric that matches the user need, and record what happened.

Common mistakes

• Too many labels. If editors cannot apply the scheme in a few seconds, it will be applied inconsistently.
• Inconsistent values across platforms. "Explainer" on the website and "explainers" in the app become two rows in every report.
• Never re-checking labels. Whether applied by people or AI, labels need periodic accuracy audits.
• Optimising for one metric. Chasing the highest-engagement label can crowd out reporting the organisation exists to do. A publishing consultant quoted by InPublishing describes the model as a tool to differentiate a publication, not a solution to everything.

How Druid Learning helps

Druid Learning is an AI archival processing platform, based in Dublin and New York, that turns content archives into AI-ready datasets. Enhanced Analytics feeds content labels into analytics tools such as Chartbeat and Google Analytics. For NPR, Druid processed more than 2 million assets with 20+ labels across 8 dimensions, including user needs, article type, location, topic and community, so audience analytics can be broken down by label in both tools. For the University of Bristol, enriched topic and format metadata was connected to analytics and research workflows. Read the NPR and University of Bristol case studies.

Frequently asked questions

Why are pageviews a misleading metric for content?

Pageviews count how often a page loaded, not whether anyone read it or found it useful. They favour teams that publish more items and stories that happen to receive search or social traffic. Chartbeat's 2014 analysis of 2 billion visits found 55% of visitors spent fewer than 15 seconds actively on a page. Engaged time, return visits and conversions, read by label, give a fuller picture.

What is the User Needs model for news?

The User Needs model is a framework for classifying content by what it does for the reader. It grew out of BBC audience research, and BBC World Service's language services used six needs, including update me and educate me. Later, smartocto and Dmitry Shishkin developed User Needs Model 2.0, with eight needs under four drivers: to know, to understand, to feel and to do. Newsrooms use it to balance their output.

How do I add content labels to Google Analytics 4?

Send each label, such as topic, format and user need, as an event parameter from your site's tagging, through Google Tag Manager or gtag.js. Then register each parameter as an event-scoped custom dimension in the GA4 admin settings; data appears in reports after 24 to 48 hours. Use short controlled values and avoid high-cardinality fields such as IDs or timestamps.

Can Chartbeat report on topics, formats or user needs?

Chartbeat reports on the metadata a site passes in its tracking configuration. Its sections variable accepts up to 10 comma-separated values per page and drives the section filters in its dashboards, so one option is to add topic, format and user-need labels as section values. That is a publisher convention, not a dedicated Chartbeat feature, so keep the names consistent.

How many user needs should one article have?

One primary user need per article is the simplest rule to apply and report on, and it is the approach an InPublishing feature on user needs recommends. A piece can carry secondary labels for analysis, but the primary need should decide how it is written, packaged and judged. If many articles seem to need two primary needs, the definitions probably need sharpening.

Sources

• What You Think You Know About the Web Is Wrong — TIME
• How BBC World Service engages younger audiences by fulfilling six reader needs — Journalism.co.uk
• How to get started with user needs: learnings from the Reuters Institute, Metro.co.uk and smartocto — Journalism.co.uk
• The User Needs Model: Driving Relevance, Resonance and Revenue in Modern Newsrooms — FT Strategies and smartocto
• Using user needs to improve content performance — InPublishing
• Custom dimensions and metrics — Google Analytics Help
• Event parameters — Google Analytics developer documentation
• Customize Tracking Settings — Chartbeat Documentation