How do newsrooms use audience data?
Constantly, and more carefully than critics assume — the useful distinction is between using data to decide what to cover and using it to decide how to present and distribute what has been judged worth covering.
What they actually measure. Not simply page views. Modern newsroom analytics track attention time, scroll depth, completion rate, whether a reader returns, referral source, and — for subscription businesses — which articles cause someone to subscribe or to stay. That last metric has changed newsroom behaviour more than any other, because it rewards distinctive and demanding journalism rather than volume.
Where it is used legitimately:
Headline and image testing, running variants to see which communicates better.
Timing and placement, deciding when to publish and how long to keep a story prominent.
Identifying unmet demand, where readers repeatedly search for something not covered.
Resource allocation, showing which beats readers value.
Follow-ups, since sustained interest justifies continued reporting.
Where it distorts:
Chasing traffic towards cheap, aggregated and outrage-driven content, which performs on raw page views and builds nothing.
Abandoning necessary but unpopular coverage — local government, courts, and complex policy reliably underperform and are exactly what public-interest journalism exists to provide.
Headline inflation, where the version that tests best overstates the article — a documented source of the gap between headline and content.
Homogenisation, as every outlet optimises against similar signals and converges.
Why subscription models changed the incentives. Advertising rewards volume of attention from anyone; subscriptions reward depth of attention from people willing to pay. That shift is the main reason several major outlets invested in investigations and specialist reporting rather than aggregation.
What good newsrooms do: publish some things regardless of performance because they matter; separate editorial judgement from commercial targets; avoid displaying live traffic rankings to reporters, which distorts behaviour; and treat data as one input among several, not an editor.