For Media and Publishing

How Media Organisations Can Use AI Without Losing Editorsial Judgement

AI tools like ChatGPT and Claude now generate content faster than reporters can write it. This pressure is real: optimised AI drafts often outperform human-written pieces in engagement metrics. The risk is that editors choose the AI version not because it is better journalism, but because the numbers say it will get more clicks. Your readers cannot tell which stories came from reporting and which came from a language model trained on internet text.

These are suggestions. Your situation will differ. Use what is useful.

Download printable PDF

Keep Reporting Separate From AI Drafting

The jobs AI does well (summarising press releases, structuring data, filling in public information) are exactly the jobs that used to train junior reporters. When you use Claude to draft a company earnings story from SEC filings, you save time but you lose the moment a reporter would have asked a follow-up question. Use AI for assembly of known facts, not for the work that develops journalistic instinct. Protect the time your team spends on reporting that AI cannot do: interviews, observation, investigation.

›Use AI to turn source documents into structured outlines, then assign human reporting to fill the gaps
›Have reporters review all AI-generated copy before publication and mark what they would have reported differently
›Track which stories require the most editor corrections from your AI tools, as this shows where AI is weakest

Recognise When AI Creates Editorsial Homogenisation

When your newsroom and five other outlets all use the same AI model for the same news, you all publish the same story at the same time with the same framing. This kills the competitive advantage of having better reporters. Perplexity and Google Gemini will tell your competitors the same thing they tell you. Editorss who rely on these tools for story ideas and angles are following the crowd, not leading it. Your readers chose your outlet for your voice and your judgment, not for someone else's facts read through a language model.

›Assign one reporter to manually find stories that your usual AI research tools miss entirely
›Create an editorial rule: if three stories this week came from the same AI summary of the same news, stop using that tool for that topic
›Review your homepage monthly and note which stories look similar to competitors, then ask if AI was involved in the original reporting decision

Maintain The Distinction Between Reporting And Generation

Readers trust journalism because they know a human made judgement calls about what matters and what is true. When you publish a story that came mostly from AI, you have an obligation to be clear about that. AP Automated Insights works well for earnings summaries and sports scores because readers expect those formats and understand their limitations. But readers who see a byline expect human reporting. The moment a reader discovers that a story they thought was reported was mostly generated, you have damaged trust. That damage spreads because your audience will now question which of your stories are real.

›Mark AI-assisted stories with a clear note about what was generated versus what was reported
›Never run AI-generated stories under a reporter's byline unless that reporter did the original reporting
›Do a quarterly audit of your most-read stories and check which ones involved AI generation, then measure reader trust signals for those pieces

Build Editorsial Standards For When To Use AI

You need a rule book, not a principle. Tell your team exactly which tasks AI is allowed to do. Use Claude to draft the first section of a council meeting story from your notes, but not to interview sources. Use ChatGPT to generate alternative headlines and then have an editor choose. Use Google Gemini to find public data on a topic, but require a reporter to verify and contextualise it. These rules should be different for different story types and beats. The rule should be specific enough that a reporter can apply it without asking an editor every time.

›Write a one-page guide that lists each AI tool and exactly what story stages it can be used for
›Include examples from your own published work showing where AI helped versus where it failed
›Review this guide with your legal team to make sure you are not using AI in ways that could create liability for accuracy or plagiarism

Measure What Matters Beyond Engagement

Your analytics probably show that AI-optimised headlines and structures get more clicks. But engagement metrics miss what journalism is actually for. You need to also measure: how many sources a story used, whether it changed anything in the community, how many corrections you published for it, and whether readers still cite it months later as important. A story that got millions of clicks but turned out to be wrong is worse than a story that got fewer clicks and was right. When you measure only engagement, AI will make you worse at journalism because it optimises for what makes people click, not what makes them informed.

›Create a monthly report that compares engagement, accuracy rate, and reader trust signals for AI-assisted versus fully reported stories
›Ask five readers per month: which stories from your outlet changed how they understood something important, then check if those stories used AI generation
›Set a rule that no story can be published if it relies on AI generation for core facts that you could have reported yourself

Key principles

1.AI is best at assembling known facts, not at the reporting that develops journalistic judgment and uncovers what others missed.
2.When multiple outlets use the same AI tool for the same news, editorial advantage disappears and readers get the same story everywhere.
3.Your readers trust your byline because they know a human made decisions about what is true and what matters, so you must protect that distinction.
4.Engagement metrics optimise for clicks, not for accuracy or impact, so measure reader trust and lasting influence alongside traffic.
5.Use AI for the work that saves time without damaging training, not for the work that used to teach your reporters how to think like journalists.

Key reminders

When you use Claude or ChatGPT to draft a story, have the reporter who would have written it manually fact-check the AI output line by line so they learn where AI goes wrong
Create an editorial calendar rule: no more than one generated-assisted story per reporter per week, to protect time for real reporting
Track the correction rate for AI-assisted stories separately from fully reported stories, and publish this data to your team every month
Before using any new AI tool, identify which jobs it would replace in your newsroom and decide if you can afford to lose the training that job provided
Ask your audience directly: how would you feel learning that a story you read and shared came mostly from AI, and use their answer to guide your policy