I’ve reviewed AI policies from over a dozen newsrooms in the past year. Most are either too vague to be useful or so restrictive that they prevent legitimate experimentation.

Creating a good AI policy is harder than it looks. You need to permit beneficial uses while preventing harmful ones, all while acknowledging that the technology is evolving faster than any policy can track.

Here’s a framework I’ve developed for newsrooms creating or updating their AI policies.

Start with Principles

Before writing rules, establish principles. These guide decisions when specific situations aren’t covered by explicit policy.

Transparency. Readers have a right to know when AI significantly contributed to content. This doesn’t mean disclosing every AI-assisted typo fix, but meaningful contributions require disclosure.

Human accountability. A human journalist must take responsibility for all published content, regardless of what tools contributed. AI doesn’t byline stories; humans do.

Quality standards don’t change. Content must meet the same standards for accuracy, fairness, and quality whether AI assisted in creating it or not. AI is a tool, not an excuse for lower standards.

Innovation within boundaries. Staff should be encouraged to explore AI capabilities while respecting guardrails. Fear-based policies that prohibit all experimentation serve no one.

These principles create a foundation. Specific rules flow from them.

Define Use Categories

Blanket approval or prohibition of “AI” isn’t practical because AI encompasses vastly different applications. I recommend categorizing uses:

Green Light Uses (Generally Permitted)

These applications have proven value and limited risk:

• Transcription of interviews, press conferences, and recordings
• Translation of foreign-language sources for research purposes
• Summarization of lengthy documents to identify relevant sections
• Grammar and spell-checking in draft content
• Research assistance to identify sources, find background, or understand technical topics
• Scheduling and administrative tasks unrelated to editorial content

These uses should require no special approval, though staff should be trained on best practices.

Yellow Light Uses (Permitted with Review)

These applications have potential value but require oversight:

• Draft generation of routine content (calendars, event listings, routine data reports)
• Headline or social copy suggestions that will be reviewed and edited by humans
• Data analysis where AI identifies patterns that humans then verify
• Image editing (cropping, enhancement) that doesn’t alter factual content

These uses should require editor awareness and possibly approval depending on the specific application. Human review of output is mandatory.

Red Light Uses (Generally Prohibited)

These applications pose unacceptable risks:

• Publishing AI-generated text without substantial human involvement
• Using AI to fabricate quotes, sources, or facts
• AI-generated images presented as photojournalism
• Analysis or conclusions published without human verification
• Any use that misleads readers about what they’re consuming

Prohibitions should be clear and consequences for violations explicit.

Address Specific Scenarios

Beyond categories, address scenarios journalists will actually face:

Interviewing AI

Can journalists quote AI chatbots as sources? Generally yes, if clearly attributed. “When asked about X, ChatGPT responded…” is legitimate if the AI response is accurately quoted and the limitations of AI as a “source” are understood.

AI-Assisted Investigation

When AI helps identify patterns in large datasets, the AI’s role should be documented in methodology notes. The findings must be verified through traditional reporting before publication.

Images and Multimedia

AI-enhanced images (brightness, cropping) differ from AI-generated images. The former may be acceptable for news photography; the latter generally isn’t. Clear guidelines should distinguish enhancement from fabrication.

Social Media

Using AI to help write tweets or social copy is different from using it to write articles. Social platforms may have lower stakes, but consistency with broader policy matters.

Disclosure Requirements

When must AI involvement be disclosed to readers?

I recommend this threshold: disclose when AI meaningfully contributed to the substance of the content. A few specific guidelines:

• Routine tools require no disclosure. Spell-check, basic grammar assistance, transcription.
• AI-assisted drafts require disclosure. If AI generated substantial portions of text that were then edited by humans.
• AI analysis requires disclosure. If AI identified patterns or connections that inform the reporting.
• Methodology notes suffice for most cases. “Reporting for this story was assisted by AI transcription and data analysis tools.”

The test: would a reasonable reader want to know? If yes, disclose.

Enforcement and Evolution

Policies require enforcement mechanisms and evolution pathways.

Training is essential. A policy that isn’t understood isn’t followed. All staff should receive training on AI policy, updated as the policy evolves.

Violations have consequences. Specify what happens when policy is violated. Proportional responses—coaching for minor issues, serious consequences for deliberate deception.

Regular review cycles. AI technology evolves rapidly. Build in quarterly or semi-annual policy reviews. What was prohibited last year might be standard practice next year.

Feedback mechanisms. Staff should be able to flag policy gaps or propose modifications. Top-down policy without practitioner input becomes disconnected from reality.

Sample Policy Language

Here’s template language you can adapt:

---

[Publication Name] AI Policy

Effective [Date] | Next Review: [Date]

Principles [Publication] embraces AI tools that help journalists work more effectively while maintaining our commitment to accuracy, transparency, and editorial independence. AI assists our journalism; it does not replace human judgment and accountability.

Permitted Uses Staff may use AI tools for transcription, translation, grammar assistance, research, and other routine tasks that support reporting. Human review of AI output is expected.

Restricted Uses Uses involving content generation, data analysis for publication, or image manipulation require editor approval and will be evaluated case by case.

Prohibited Uses Staff may not: publish AI-generated content without substantial human involvement and review; use AI to fabricate information; present AI-generated images as photojournalism; or use AI in any way that deceives readers.

Disclosure When AI meaningfully contributes to published content, that contribution will be noted in the story. The format and specificity of disclosure will be determined by the editor.

Questions Staff with questions about whether a proposed AI use complies with policy should consult [designated contact].

---

Final Thoughts

No policy is perfect. The goal is creating a framework that permits innovation while protecting against genuine harms.

The newsrooms succeeding with AI aren’t those with the most permissive or most restrictive policies. They’re those with thoughtful policies, well-communicated, regularly updated, and actually followed.

If you’re drafting policy now, you’re in good company. Most newsrooms are working through these questions simultaneously. If you need help developing frameworks, the team at Team400 and similar specialists work with media organizations on exactly these issues. Don’t let perfect be the enemy of good—publish a policy, learn from experience, and iterate.

The alternative—no policy at all—leaves decisions to individual judgment without guidance. That’s worse for everyone.