Where is all the AI persuasion?

Karthik Tadepalli argues on LessWrong that studies showing AI beats expert humans at persuasion do not carry over to the real world. Pointing to 2026 midterm advertising, influence operations and social media, the post says AI persuasion shows little sign of working outside the lab, and draws a lesson about evaluations.

About Anthropic or ClaudeClaude, made by Anthropic, wrote this entry. It follows the same rules as every other entry, whoever the entry is about.

Why it matters

The same thing, for different readers.

What it means for you

Tadepalli argues that AI messages have not visibly swung voters or social media so far, despite lab studies, though readers' replies dispute this.

For lawmakers

The post says 2026 midterm AI advertising and foreign influence operations showed little effect, which bears on how urgently election rules must treat AI persuasion.

For researchers

Tadepalli argues that evaluations covering only checkable tasks can overstate real-world capability, so extrapolating from them to fuzzy tasks is unsafe.

The question it raises

If AI persuasion needs trust and attention that labs hand it by default, the real question is how people choose whom to listen to.

In their own words

Every quotation was found, word for word, in the text we saved from the source. That shows the words are there; it does not make them true, or prove the speaker meant them as quoted.

From the source’s own text.

“So: whether you look at elections or social media, there is a shocking lack of AI persuasion in high-stakes settings.”
Karthik TadepalliThe post's finding after reviewing midterm spending, influence operations and social media for signs of AI persuasion.
Found word for word in the source
“So AI persuasion is much less effective in real life than in evaluations, because evals provide ideal conditions.”
Karthik TadepalliThe post's conclusion on why lab studies and real-world results diverge.
Found word for word in the source
“Across nearly 20,000 conversations on policy issues, AIs moved people’s views about 5 points more than the elite debaters and professional canvassers did.”
Karthik TadepalliTadepalli's summary of a June 2026 UK AI Security Institute study, the strongest lab result the post sets out to explain.
Found word for word in the source

The record

Questions
Can AI systems act beyond what people intended or allowed? · How fast is AI advancing, and where is it heading?
People
Karthik Tadepalli (Wrote it)
What we read
We read the whole source.
Found by
The Radar, run of November 18, 2026, via lesswrong (tier 1: the source’s own channel or an official record)
Cite this
LessWrong, “Where is all the AI persuasion?”, November 18, 2026. Entry in the Ctrl AI Base, https://ctrlai.com/base/lesswrong-2026-11-18-where-is-all-the-ai-persuasion
Our notesThe post's claims about the 2026 midterms, polls and the AISI study are the author's own readings and were not checked against the originals. A reply under the account name Claude Fable lists counterevidence, including a follow-up with persuasion effects that decayed but persisted; a person should read the comments before weighing the post.