AP

Free

AP is a global news agency that runs a research program on AI in journalism. They build tools and set rules for how newsrooms can use automation without losing trust.

Their Local News AI initiative started in 2021. They asked 192 news leaders what they actually needed, then built specific tools for five partner newsrooms.

They also train their own language model on AP content. Staff do not paste stories into ChatGPT. The work sits in research and AI governance, not a product you buy off the shelf.

Open it if you run a newsroom and need help setting AI rules or building your own tools. Skip it if you want a ready-made product for general content work.

These are the people who get real use from what AP offers.

• Small local newsrooms with specific repetitive tasks: You get help with transcription, weather alerts, or police blotters. You still have to build and customize the tool yourself.
• Newsrooms with resources to customize and oversee AI tools: You get a starting blueprint. You need staff time to adapt it to your workflow.
• Organizations prioritizing ethical guidelines in automated content: You get built-in guardrails, like removing names from police blotters to avoid shaming unconvicted people.
• News organizations prioritizing trust and transparency over speed: You get public documentation of your AI principles. Readers see how you use the tools.
• Teams needing to establish AI governance before operational deployment: You get a framework to write your own rules before you ship anything.
• Organizations with proprietary content worth protecting from third-party LLMs: You get an internal model trained on your own archives. Your stories do not leak to outside AI.
• News organizations developing their own AI policies: You get a published benchmark. You adapt it to your newsroom, not copy it whole.
• Journalists seeking clarity on appropriate AI use boundaries: You get a clear line: AI assists, humans decide. Reporting and verification stay with people.
• Publishers needing a defensible framework for human-AI collaboration: You get disclosure rules and review checkpoints. You can explain your process to readers.
• Local newsrooms with limited staff covering multiple beats: You get help with routine work like transcription and summarization. The 80/20 rule means AI does the draft, you do the final check.

These are the parts you notice if you work with their program.

• They started with real needs: They surveyed 192 news leaders before building anything. You get tools that match actual problems, not imagined ones.
• Clear rules built before deployment: They wrote their 2023 guidelines first. You get a policy you can point to when questions come up.
• Internal model for sensitive content: They trained an LLM on their own archives. You keep proprietary stories off public AI tools.
• Human-in-the-loop requirements: Every AI output needs review before it publishes. You keep final say on what goes out.
• Public transparency on principles: They publish how they use AI. You get a defensible story for your readers and your board.

Skip it when the work does not match what you need.

• Not a finished product: You get frameworks and pilots, not tools that work out of the box. You still have to build most of it yourself.
• Heavy oversight required: The tools need significant human review. You do not save time unless you invest more staff time first.
• Unproven time savings: No hard numbers show these tools actually save hours. You are betting on potential, not results.
• Only fits specific workflows: The tools were built for five specific newsrooms. You may find nothing that maps to your daily work.
• Policy still untested at scale: The disclosure rules and guardrails look good on paper. You do not yet know if they hold up under deadline pressure.