For six months, we have been using AI agents to help draft scripts and story angles for Fear & Greed — Australia’s most popular business podcast! Listen here.
We have seen some entertaining failures.
One agent proposed leading on the Super Bowl result — before kickoff.
Another was following the Coalition leadership crisis — but its source was days out of date in what was a fast-moving story.
The drafts read perfectly fine. But they were both very wrong.
Most content teams are being told to use AI more, but as Fear & Greed taught us, the results are mixed.
The tools are impressive. They can produce a draft quickly, smooth a sentence, reshape copy for different channels. But more often than not, the result is not quite right.
It can be hard to pin on one obvious failure. The draft is clearly written. It may be better than what was there before. But on closer reading, it overstates, has the wrong emphasis, or says the same thing over and over just in different words.
It often lacks ‘heart’ and doesn’t talk directly to the audience.
What we think is happening is that the model produces fluent sentences before the harder questions about audience, emphasis and tone have been settled.
Who exactly is reading this? What do they already know? What is the one thing this piece needs to do? What claims can actually be supported?
In professional writing, those questions are the main job. Getting words onto a page is the easier part.
So what’s the fix?
Changing models, switching providers or writing longer prompts sometimes helps. But a lot of the time the deeper issue is how well the task, source material, standards and review process have been set up around the model.
The next time an AI draft disappoints, try asking whether the setup gave the model enough to work with. Clearly spelling out what the piece needs to do, who it is for, what sources can be relied on and what good looks like is usually where the real improvement comes from.