"PERSUASION BOMBING"

Here’s a new phrase for you “PERSUASION BOMBING” - the tendency of an AI system to respond to scrutiny by increasing the intensity of its persuasive tactics rather than simply correcting its answer. 🤯

Or in other words: when you challenge the results given to you by an LLM, the AI responds to criticism with increasingly credible, logical, and emotionally compelling arguments which are designed to convince you it is right. 

According to this fascinating new study, researchers analysed over 4,300 prompts related to real business problems where consultants were trying to validate the AI’s output in three ways:

🔎 Fact-checking (“Review your work.”)

🤷🏻‍♂️ Exposing inconsistencies (“This doesn’t make sense…”)

🙅🏽 Pushing back (“I disagree with your conclusion…”)

Instead of simply correcting itself… GPT’s often became a better advocate for its original answer. What is really interesting is that they use Aristotle’s three classic forms of persuasion: Ethos, Logos, Pathos, outlined in his paper Rhetoric from around 335BC!

⚖️ “Ethos” – The Credible Argument

The AI strengthened its own credibility. Apologised. Explained its reasoning. Talked about how carefully it had analysed the data. And then projected competence.

🧠 “Logos” – The Logical Argument

It produced more comparisons, statistics, more structured arguments, and more evidence, making the response felt increasingly rational.

♥️ “Pathos” – The Emotional Argument

The AI became warmer. It mirrored the user’s language, reassuring and affirming, creating a feeling of collaboration.

This paper is fascinating from a STORYTELLING and comms perspective because it argues that GPT's use all three forms of persuasion.

It doesn’t just make logical arguments (logos). When challenged, it also tries to increase trust in itself (ethos) by apologising, explaining its process, and emphasising its careful analysis, and it builds rapport through affirming and mirroring the user (pathos).

Rhetoric, like AI, is a tool which can be used for good (to give a great talk) or bad (political manipulation). And like AI, rhetoric is subject to the ethics and morals of the person using it.

If you’ve got a moment, the paper is well worth a read, as it might completely change the way that you interpret whatever your favourite LLM is telling you.