Reynolds Center: new AI tools that are genuinely useful to business journalists

The Reynolds Center for Business Journalism published a recap of a SABEW panel held on February 17, 2026, bringing together journalists from the Wall Street Journal, ICIJ, Google News Initiative, and Forbes to assess which AI tools have genuinely become part of daily reporting work — not just experimentation.

The panel was moderated by Jeff Kauflin, Senior Editor at Forbes. Panelists were Isabella Cota (International Consortium of Investigative Journalists), Mago Torres (training director at the Google News Initiative, former data editor at The Examination), and Sebastian Herrera (technology reporter at the Wall Street Journal).

The conversation organized around two categories of tools: Google NotebookLM and the deep research modes now available in ChatGPT, Claude, and Gemini.

NotebookLM emerged as particularly suited to reporting because of how it limits its answers. Unlike general-purpose language models that draw on broad training data, it only responds based on documents the user uploads directly — articles, transcripts, news clippings, video content. This matters for journalism because the model cannot introduce facts from outside those sources, which reduces the risk of hallucinated details creeping into research notes.

Sebastian Herrera described using it for a time-sensitive assignment: after Oracle’s earnings call, he was asked to write a Larry Ellison profile in a couple of days. He loaded prior Ellison media appearances into NotebookLM and queried specific topics — what Ellison had said about Bill Gates, or about running a company into his 80s. The approach significantly cut the time he would have spent manually searching and reading prior coverage. His summary: “It really sped up the reporting process for me.”

The deep research modes in ChatGPT, Claude, and Gemini were described as a meaningful development from earlier versions. The panelists framed them as the models effectively running two stages of processing: a fast-response layer for simple queries, and a slower, step-by-step reasoning layer for complex multi-part research tasks. This makes them more useful for substantive reporting questions rather than basic lookups.

The panel noted ongoing concerns about accuracy, hallucinations, and ethics, but framed these as less prohibitive than they were even a year earlier, as both the tools and newsroom policies around them have matured.

For business and investigative journalists evaluating which AI tools to actually add to their regular workflow, the article is useful as a practitioner-level assessment from working reporters — not a product demo or a vendor claim.