The State of Journalism Quality 2026
Measurement, verification, and the gap between standards and practice
A specification-stage market report. Findings are illustrative of the method and will be replaced with measured data.
This is a specification-stage draft. The structure, method, and recommendations are real; the figures are illustrative and must not be cited as measured results. The final report will publish its dataset and limitations.
Executive summary
Journalism has more standards than it has measurement. Newsrooms can cite a code of ethics, a style guide, and a corrections policy, but most cannot say how consistently those standards are applied across desks, formats, or languages. That gap is the subject of this report.
We propose a measurement framework built on seven dimensions, claim-level evidence, and human review. We argue that the goal is not a single quality score but a structured conversation: where quality breaks down, why, and what would improve it.
Five findings are presented illustratively: quality is described in style-guide terms rather than measured; fact-checking capacity is concentrated; AI adoption is outpacing verification policy; readers value sourcing signals but rarely see them; and procurement increasingly asks for evidence of editorial controls.
1. Why measure quality?
There are three common objections to measuring journalism quality. The first is that quality is inherently subjective. The second is that measurement will be gamed. The third is that resources are better spent on reporting.
Each objection contains a truth. Quality is partly subjective, metrics are gamed, and reporting matters more than dashboards. None of them is a reason to abandon measurement. They are design constraints.
Subjectivity is managed by separating dimensions and publishing definitions. Gaming is managed by measuring processes and patterns rather than individual rankings. Resource constraints are managed by starting small: one desk, one content type, one question.
1.1 The cost of not measuring
Without measurement, quality debates become anecdotal. A senior editor says standards are slipping; a desk editor says the sample is unfair. Neither can prove the claim. Training budgets are allocated by intuition. Corrections are treated as isolated incidents rather than patterns.
The cost is not abstract. It shows up in reader trust, in correction volume, in the time senior editors spend on low-risk copy, and in the inability to answer a client or regulator who asks, “How do you know?”
1.2 What measurement is not
Measurement is not grading individual journalists. It is not a substitute for editorial judgement. It is not a compliance certificate. It is a way to make patterns visible so that people can act on them.
2. Method
The proposed method has four components: a dimension framework, a content-type weighting scheme, an evidence standard, and a review protocol.
2.1 Dimensions
Accuracy, Evidence, Sourcing, Fairness, Context, Clarity, and Accountability. Each is defined independently and mapped to established codes. The definitions are published on the assessment page and versioned.
2.2 Content types
Different formats carry different expectations. A breaking brief and an investigation should not share weights. We specify 54 content types, each with its own weighting and minimum evidence thresholds. Type is a suggestion that editors can override; overrides are logged.
2.3 Evidence standard
Every score must be traceable to evidence: a quoted passage, a linked source, a named entity match, or a detection signal. If evidence is unavailable, confidence is lowered and the absence is recorded. A score without evidence is not published.
2.4 Review protocol
High-stakes findings require a named reviewer. Reviewer decisions, overrides, and reasons are stored with the finding. A sample of decisions is independently re-reviewed to estimate agreement and drift.
3. Findings (illustrative)
3.1 Quality is described, not measured
Most newsrooms can describe quality in prose. Few can produce a distribution of evidence quality across a month of coverage. The gap is not laziness; it is the absence of a shared instrument.
3.2 Fact-checking capacity is concentrated
Verification capacity is clustered in a small number of organisations and languages. During elections, demand spikes while supply is fixed. The result is triage by intuition, not by consequence.
3.3 AI adoption is outpacing policy
Tools are being adopted at the desk level before governance exists at the organisation level. That is not inherently bad, but it makes accountability unclear when something goes wrong.
3.4 Readers value sourcing signals they rarely see
Readers say they want to know who said what and how it was checked. Most articles provide links, but few provide a structured account of evidence. The gap is an interface problem as much as a reporting problem.
3.5 Procurement is asking new questions
Buyers increasingly request evidence of editorial controls alongside security controls. “How do you prevent a wrong verdict from being published?” is becoming a standard question.
4. Recommendations
- Start with one question. Pick a dimension and a content type. Measure for a month. Publish the method internally.
- Separate individual and system metrics. Report desk-level patterns; avoid byline league tables.
- Make evidence visible. Require a source or a caveat for every material claim in review.
- Record overrides. Treat reviewer disagreement as data, not failure.
- Write the AI policy before the AI pilot. Define review, disclosure, and incident response first.
- Publish limitations. A method without limitations is marketing.
5. Limitations
- The findings are illustrative and not based on a completed survey.
- The framework is built primarily from English-language editorial codes.
- Content-type weights require validation per newsroom.
- Measurement may create incentives that distort behaviour; monitoring is required.
- Small samples cannot support outlet-level comparisons.
6. Conclusion
Quality is not a number. It is a set of judgements that can be made more consistent, more transparent, and more contestable. Measurement is worth doing if it improves those judgements. It is not worth doing if it replaces them.
Appendix: proposed survey questions
- How does your newsroom define quality in practice?
- Which quality checks are mandatory before publication?
- How do you record that a check happened?
- What happens when a check fails?
- How do you measure whether quality is improving?
- Who owns the corrections process?
- How do you review AI-assisted output?
- What evidence would you need to trust an external quality tool?
Get the full report
The download will be available when the study is published. Leave your email and we will send it to you, with the method and limitations included.