A clip reaches your newsroom just before a deadline. The image looks plausible. Its caption fits the story. A generated summary confidently describes who appears in it. None of that answers where it came from or whether the account is accurate.
Your team still has to make those decisions. As production and distribution involve more AI systems, preserving the reasons for an editorial decision matters as much as preserving the version that was published.
AP has expanded AI guidance while keeping people accountable.
On 23 July 2026, The Associated Press updated its newsroom AI standards. Its announcement allows assistance with tasks including research, transcription, translation and headline suggestions, while requiring journalists to review and edit output before publication.
It keeps reporting, sourcing, verification and editorial judgement with people. AP also maintains its prohibition on generative AI creating, altering or enhancing news photography. The distinction is useful: a newsroom can define specific permitted uses without surrendering responsibility for what it publishes.
For your own team, the practical question is whether a colleague can understand which use was permitted and how the resulting material was checked. A policy becomes much easier to apply when the decision is attached to the work.
An answer can sound certain and still mishandle its sources.
The EBU and BBC published an international study in October 2025 in which journalists assessed more than 3,000 news answers from four AI assistants. It reported at least one significant issue in 45% of assessed answers and serious sourcing problems in 31%. The work involved 22 public service media organisations across 18 countries.
These are results for the tested assistants, questions and research period. They are not a current error rate for every AI system or a test of whether journalists created a video. They do establish why a fluent response and a source label deserve closer examination.
When a summary names a source, check the original passage and its date. When a clip carries an account name, establish whether that person captured it, reposted it or added a new claim. Preserve the distinction in the review record.
Use verification tools without outsourcing the editorial decision.
AP introduced its Verify dashboard in December 2025, bringing established digital verification tools together with AI features for checking online material. The product is another example of a newsroom treating verification as a workflow with several sources of evidence.
Reuters similarly commits in its published standards to editorial accountability and transparency when using generative AI. Its rules apply sourcing and verification expectations to online material. Tools can organise or accelerate a check, while the publishing organisation remains responsible for the decision.
Keep your review practical: the source, the checks performed, unresolved questions, the editor responsible and the approved version. If evidence changes, the team should be able to find the affected publication and explain a correction.
Show who worked on the package you delivered.
After the source checks, your team still edits, produces sound, prepares graphics and approves a final package. DigiBridge connects those contributors and permitted creative activity to a registered project and version review.
Its strict programme checks for prohibited AI assistance and software generated control that replaces the contributor. A dedicated verification agent supports device, system and creative tool checks, with evidence and creator context available for review. Your organisation can assess that creative record alongside its own editorial checks.
Keeping those records distinct prevents a misleading claim: evidence that a human edited a package does not establish that the event it depicts happened. Both the production process and the reporting need scrutiny.
“Let the next editor see what was checked, who decided and which version was approved.
How DigiBridge helps
Keep your team’s contribution visible without exposing its sources.
DigiBridge records permitted activity without collecting private source files or typed content. It supports review of the creative process behind a news package. Fact checking, source protection and the decision to publish remain with your newsroom.
In the strict human creation programme, permitted device, operating system and creative tool evidence feeds layered checks and an AI verification agent. These detect prohibited AI assistance and software control, warn creators with supporting evidence, and inform publishing decisions after confirmation and review. A warning or uncertain result alone does not count as an incident.
Read the human creation standardSources and further reading
Primary and authoritative sources reviewed for this article. Statistics remain attributed to the organization that published them.
- 01Updated newsroom AI standardsThe Associated Press23 July 2026. Permitted assistance, human review and news photography rules.
- 02News integrity research findingsEuropean Broadcasting Union22 October 2025. Findings from more than 3,000 evaluated news answers, with the study scope stated in the article.
- 03AP Verify introductionThe Associated Press15 December 2025. A dashboard combining digital verification tools and AI features.
- 04Journalistic standardsReutersSourcing, online verification and editorial responsibility for AI use. Checked 30 September 2026.
Continue reading
Browse all insights