The Disclosure Obligation
Artificial intelligence tools are increasingly embedded in journalistic workflows — from transcript summarisation and data analysis to image generation and headline optimisation. The ethical question is not whether AI can be used, but when and how its use must be disclosed to maintain reader trust and compliance with editorial standards.
IPSO has not yet published a standalone AI clause, but Clause 1 (Accuracy) creates an obligation that AI use implicates directly. Any AI-generated content published as fact that is inaccurate could breach Clause 1. IPSO’s guidance (updated 2024) confirms that editors bear responsibility for verifying AI-generated content before publication, and that significant AI involvement should be disclosed to readers.
The NUJ AI policy (2024) requires disclosure where AI plays a “significant” role in content creation. Reuters AI Trust Principles establish that AI-generated content must be clearly labelled and subject to the same editorial standards as human-generated content. The C2PA standard provides a technical mechanism for attaching verifiable provenance metadata to digital assets. See also: /ethics/ai-generated-content.
When AI Disclosure Is Required
AI-generated article text
Any article substantially written by an AI model should be clearly labelled. "Substantially" means the AI generated the primary draft, not merely suggested edits. A journalist who prompts an AI to write a 500-word news report and publishes it after light editing must disclose AI generation, not merely AI assistance.
AI-generated or AI-modified images
Images produced by generative AI (Midjourney, DALL-E, Stable Diffusion) must be labelled. Images that have been significantly altered by AI — not just colour-corrected — require disclosure. The C2PA standard allows publishers to attach verifiable provenance metadata. Unverified images from social media should be checked for AI generation before publication.
AI data analysis and summarisation
Using an AI to analyse data and summarise findings before a journalist writes the article is AI assistance, not AI generation — disclosure is not mandatory but is good practice. If the AI analysis is quoted or cited directly, the AI origin should be stated.
AI headline or SEO optimisation
AI-optimised headlines are AI-assisted editorial decisions. Disclosure is not typically required, but newsrooms should have internal policies. AI that changes the substance or framing of a headline (not just its length) requires editorial oversight.
Red Flags
- ⚠Publishing AI-generated text or images without verification against primary sources
- ⚠Using AI-generated images without labelling, particularly where they could be mistaken for photographs
- ⚠Failing to disclose AI involvement when the AI produced the primary draft of an article
- ⚠Using AI to create synthetic quotes or statements attributed to real people
- ⚠Relying on AI-generated summaries of documents without reading the original documents
- ⚠No newsroom policy on AI use disclosure, verification, or editorial sign-off
- ⚠Using deepfake video or audio without verification and prominent disclosure
AI Disclosure Checklist
AI Disclosure & Verification Tools
Use the AI Disclosure Checklist tool to document AI involvement in a story before publication, and the Social Media Verification Checklist for AI-generated image detection.
Common Mistakes
- ›Conflating AI-assisted with AI-generated: Using Grammarly or an AI headline tool is AI assistance. Prompting an AI to write an article is AI generation. These require different disclosure levels. Newsroom policies should distinguish between the two.
- ›Treating AI detection as definitive: No AI detection tool is 100% accurate. Detection outputs should inform editorial judgement, not replace it. Combine tool output with source credibility assessment, metadata examination, and contextual analysis.
- ›Omitting disclosure for AI images because they look photographic: Photorealistic AI-generated images require disclosure precisely because they can be mistaken for photographs. The more convincing the image, the greater the disclosure obligation.
- ›No verification of AI-generated facts: Large language models hallucinate — generating plausible-sounding but false information. Every factual claim in AI-generated content must be verified against a primary source before publication, regardless of how confident the AI output appears.