How AP, Reuters and others write a newsroom AI policy

As of August 2023, most news companies allowed some use of generative AI, but only under human editorial supervision. Many guidelines prohibited using AI to write articles, and AI-generated images and video received extra scrutiny.^† AP and Reuters have written detailed standards, while smaller outlets are still drafting a newsroom AI policy of their own.

• AP treats any generative AI output as unvetted source material.
• Reuters requires journalists to independently verify facts, sources and claims produced by AI.
• Newsrooms show near-uniform agreement that AI use should be disclosed.
• Only about 20% of local news organisations have public AI usage policies.

What do AP's generative AI standards allow?

AP's standards say any output from a generative AI tool is unvetted source material. Staff must apply editorial judgement and AP's sourcing standards before publishing anything drawn from it. They may experiment with ChatGPT with caution, but they do not use it to create publishable content.

The rules for visual and audio material are stricter. AP does not use generative AI to add or subtract elements in its photos, video or audio. It will not transmit AI-generated images suspected or proven to be false depictions of reality, though it may use such an image when the image is itself the subject of a news story and is clearly labelled.

AP also urges staff to keep confidential or sensitive information out of AI tools and to check that material from outside sources is free of AI-generated content.^† If staff doubt a piece of material's authenticity, they should not use it. AP says AI can be used for menial tasks such as story digests for newsletters. It has experimented with simpler forms of AI for a decade, producing short stories from sports scores or corporate earnings reports.^†

How does Reuters handle verification and disclosure?

Guidance issued in 2023 by Reuters's editor-in-chief and ethics editor set out four principles:^†

• Experimentation is necessary.
• Journalists must approve AI-generated material.
• AI use is disclosed when it materially affects the result.
• Journalists must stay sceptical and verify.

The agency's published standards require journalists to independently verify facts, sources and claims produced by AI.^† They prohibit using generative AI to create or enhance news imagery. Editors and reporters remain responsible for AI-assisted content, and Reuters grounds its approach in the Reuters Trust Principles.^†

Reuters treats disclosure as a spectrum. Fully automated outputs, such as some automated translations, carry prominent labels, while an assistive use like an editor-reviewed headline suggestion does not necessarily require disclosure.^† The test is whether automation directly produces content for publication.

Will AI replace journalists?

These policies keep a journalist accountable for published work. Insider told its newsroom it may experiment with generative AI, but reporters remain responsible for the accuracy, fairness, originality and quality of every word.^† Its memo told staff that ChatGPT is not a journalist.

The Guardian said in June 2023 that it will use generative AI only with clear evidence of a specific benefit, human oversight and the explicit permission of a senior editor.^† Sahan Journal's policy includes a pledge not to use AI to replace staff.

Automation tools are built around a human review step. Reuters's FactGenie extracts key facts from corporate disclosures to suggest alerts, which a journalist must still confirm and publish.^† A 2026 newsroom report says it was deployed to 150 journalists and halved the average time to send non-corporate alerts.

Why newsrooms agree on disclosure

Disclosure of AI use is an area of near-uniform agreement among newsrooms.^† CNET's AI publishing experiment prompted more companies to include disclosure in their standards. A Reach-owned newspaper in Nottinghamshire trialled AI-generated bullet-point summaries on some longer articles; an editor checks them before publication, and a line at the bottom discloses the AI use.^†

Readers want the information, though it does not always raise trust. Trusting News research found that 93.8% of news consumers want AI use disclosed. In a test with 10 partner newsrooms, 42% of respondents said a disclosure made them less likely to trust the story and 30% said more likely.^† Trusting News still recommends public AI policies and in-story disclosures.

Who controls AI training on news content?

Licensing has developed alongside editorial rules. AP has a licensing agreement with OpenAI, and in July 2023 it became the first major news company to strike a deal letting OpenAI use its content to train AI models.^† The New York Times updated its terms of service on 3 August 2023 to forbid using its content to train machine learning or AI systems.^†

How common are written AI policies?

Only about 20% of local news organisations have public AI usage policies. Of 28 newsrooms in one 2025 AI cohort, four have published public policies, three have internal policies and six are drafting them.^†

A survey of 1,004 UK journalists, carried out between August and November 2024, found that 44% said their main outlet had rules for human oversight and control. Transparency protocols were reported by 42%, and guidelines on bias and fairness by 27%.^†

How to start drafting a newsroom AI policy

Poynter recommends beginning with an AI committee that includes leadership and the functions the policy will affect. It also recommends sorting decisions into three use cases: audience-facing editorial work, back-end reporting assistance, and business strategy and fundraising.^†

Sahan Journal spent about two months researching, drafting and approving its policy.^† The document covers AI disclosure in editorial content and sets out banned and permissible uses. It remains unclear whether the industry will adopt a uniform approach to disclosure.

Frequently asked questions