Odd Lots · Bloomberg

This Is How to Tell if Writing Was Made by AI

April 2, 2026·49 min·4 clips

Joe put AI writing through Chinese, Hebrew, and back to English to trick Pangram's detector—and it still caught it.

Bad writing has already made Tracy inert to it. The episode opens on a practical problem for journalists, readers, and anyone trying to tell ordinary weak prose from machine-made text. Joe treats it as a hunch. You see something online, cannot quite name what feels off, and still feel pretty sure a model touched it. Tracy is more worn down by the whole exercise. There is so much mediocre writing already that spending real attention on authorship starts to feel like its own tax. Then the stakes get sharper. AI-generated copy used to dodge work is one problem. AI slop used to support viewpoints or influence something by sheer volume is worse. The journalism angle matters too. If a journalist uses AI to avoid basic writing work, the conversation treats that as a labor issue and a reputational risk. People can tell, more or less. The bigger question is what happens when future training data becomes harder to trust because machine text has flooded the open internet. Joe asks the provenance question directly. Older online text, especially pre-2023 and certainly pre-2019 material, can be treated as largely human-written in a way newer text cannot. The guest says that reservoir is huge, with enough pre-2023 material to train on for a long time. But old data does not solve the need for current language. Modern topics keep moving. If models need to understand writing about LLMs and AI, old data alone can make those subjects look suspicious by default. So the answer becomes less about perfect detection and more about deciding which actors can be trusted to publish human-written material.

As heard by us

A useful look at how AI writing gets spotted, and why the line keeps moving.

Odd Lots uses the AI-writing question to reach a larger concern: how journalists and model builders tell human work from machine output, and why that judgment matters when sloppy copy can spread.

Read the full review in PlayNext →

Why you'd press play

You want a sharp read on how people are trying to spot AI writing without pretending the problem is simple.

Read the full recommendation in PlayNext →

Listen to the show on

Apple Podcasts Spotify

Play clips from this episode More from Odd Lots