Poynter: AI advice from journalists who stopped talking and started building

Alex Mahadevan, director of MediaWise at Poynter, wrote this March 2026 piece specifically for newsrooms that have spent years debating AI but haven’t yet committed to a concrete workflow. The article collects five strategies from journalists and editors who have gotten past that stage, drawing on real implementations rather than hypothetical best practices.

The first strategy is to start with a specific problem that already eats up a disproportionate amount of time. Mahadevan cites Claire Dannenbaum at the Pew Research Center, whose audience team was spending 95 percent of their time drafting formulaic social media posts. She built a WordPress plugin that generates those drafts automatically, which freed the team to focus on reader engagement — the part of the job that actually required editorial judgment.

The second strategy is to distinguish between using AI for thinking and using AI for writing. Chad Davis at Nebraska Public Media made a deliberate choice: he no longer uses AI to write editorial content but treats it as what he calls a “curiosity partner” for research and brainstorming. Corporate and administrative writing is acceptable territory for AI; journalism that speaks in the organization’s editorial voice is not.

The third strategy is to be explicit about where humans remain in the loop and not let that slip over time. KQED’s Ethan Toven-Lindsey described a workflow where AI suggests notable moments from a radio program, but a human producer decides which ones actually make the cut. The point is not that AI can’t make reasonable suggestions — it can — but that the decision remains with someone who has editorial accountability.

The fourth strategy is to use AI to uncover what audiences actually want to know. The Texas Tribune trained a chatbot on its school voucher coverage and found it surfaced questions from readers that the newsroom had not addressed. Those questions became story ideas. Rather than replacing journalism, the tool revealed gaps in the reporting.

The fifth strategy is to learn enough to prototype ideas independently. Mahadevan points to the current generation of AI coding tools, including Claude, as having removed the technical gatekeeping that previously required engineering resources for any custom newsroom tool. Journalists who can describe what they need and iterate with AI assistance can now test ideas without waiting for development cycles.

The piece is most useful for editorial managers and journalists who are ready to move beyond the debate about whether to use AI and want concrete examples of how colleagues have approached specific workflow problems. It does not cover investigative or data-heavy applications in depth — the examples are mostly production and audience-side.