Running the first AI for Journalists training session solo, before there was a trainer network or a repeatable program built around it, forced a kind of honesty in the material that easier audiences never would have demanded. Journalists do not sit through a training session politely. They interrogate it, in real time, out loud, which turned out to be the best possible stress test for figuring out what actually needed to be true in the content, versus what had just sounded true on paper.
Designing For A Room That Assumes You Are Selling Something
Most training audiences walk in neutral. They want to learn the tool, they will ask clarifying questions, and they generally extend some benefit of the doubt to the material in front of them. A room full of journalists walks in with the opposite default. Their entire professional instinct is built around assuming a claim is incomplete until proven otherwise, and a session about AI capability triggers that instinct immediately, because AI capability claims have a well earned reputation for being oversold.
That meant the training could not open the way a standard capability overview usually opens, with a confident list of what the tool can do. It had to open by being explicit about what it could not do, and where it was likely to fail, before it earned the right to talk about what it was actually good for. Leading with limitations felt counterintuitive when I was building the material, but it was the only framing that survived contact with that room, because a journalist who catches you underselling a limitation in the first five minutes stops trusting anything you say for the rest of the session.
Every Claim Had To Survive A Follow Up Question
Standard training material can get away with a certain amount of generality, statements like this tool helps you draft faster or this can support your research process. In front of journalists, a generality like that gets an immediate specific follow up. Faster how. Draft what kind of piece. What happens when the source material is contradictory. What happens when it invents a quote.
That pressure reshaped how the actual instructional content got written, not just how it got delivered live. Every claim in the material needed a concrete example sitting behind it, something specific enough that if someone asked for the mechanism, there was a real answer rather than a reworded version of the original claim. Vague competence claims got replaced with worked examples showing exactly where the tool helped and exactly where a journalist still needed to independently verify something before trusting it.
The hardest and most useful thread running through the whole session was hallucination, because for a journalist, a confidently wrong AI output is not a minor inconvenience, it is a potential professional catastrophe if it ends up in published work. The training could not treat that as a footnote. It had to be a structural part of the session, with a clear, repeatable method for how to treat AI output as a draft requiring verification rather than a finished, trustworthy answer.
Why Solo, First Time, No Safety Net Changed The Material Permanently
Because this was the first session of its kind, run without an established program behind it and without another trainer to fall back on if something in the room went sideways, every weak point in the material became visible immediately and personally. There was no version of deflecting a hard question to someone else, and no later session to quietly fix an explanation that had not landed. Whatever gap showed up in that room had to get patched in the material right there, in real time, based on the actual pushback received.
That experience ended up shaping the training content that came afterward far more than any planning session could have. Later training modules for other trainers inherited that same standard almost by default, be specific enough to survive a skeptical follow up, lead with limitations before capability, and never let a claim rest on general language when a concrete example was available instead. A gentler audience would never have forced that standard into existence. It took a room that assumed you were probably wrong until you proved otherwise.
The Actual Lesson
The best test for training material is not whether a friendly audience nods along. It is whether a skeptical, professionally trained question asker can poke at it without finding a soft spot. Designing for journalists first meant the version of the material that later reached calmer audiences was already stronger than it would have been if it had been built the easy way around.
Specific session details, dates, and participant information remain confidential given the nature of this work. Happy to discuss the general training design approach with anyone building AI literacy programs for skeptical professional audiences through the proper channel.
Written by Mohammad Farhan Habib Faraz
Senior Prompt Engineer and Prompt Team Lead at PowerinAI
www.powerinai.com
