Pull to refresh
Logo
Daily Brief
Following
Why Ranks Sign Up
Study finds AI advice moves people away from their starting opinions

Study finds AI advice moves people away from their starting opinions

New Capabilities

Two economists ran 1,500 people through 30 decisions and found chatbot advice narrowed the gap between opposing views — even when the AI flattered them

July 2nd, 2026: HumanProgress amplifies the depolarization finding

Overview

Chatbots are trained to please. They agree with you, praise your questions, and echo your leanings back at you. The worry has been obvious: AI that tells people what they want to hear should push opposing groups further apart.

A new experiment finds the opposite. Two economists gave 1,500 people advice from a large language model across 30 different decisions. On average, the advice pulled people away from their starting positions and toward the middle — even though the same AI was measurably flattering them at the same time.

Why it matters

Hundreds of millions of people now ask AI for advice; whether that nudges opposing groups apart or closer together shapes how divided everyday choices become.

Questions about this story

No questions yet — be the first to ask.

Key Indicators

1,500
Participants
People who received AI advice across the experiment.
30
Decision environments
Distinct choices spanning economics and social science, from moral to strategic tasks.
Depolarized
Average direction of effect
Advice moved people away from their initial leanings, not toward them.
Weaker
Effect of more sycophancy
Turning up the AI's flattery shrank the depolarizing effect but did not reverse it.

Voices

Curated perspectives — historical figures and your fellow readers.

Ever wondered what historical figures would say about today's headlines?

Sign up to generate historical perspectives on this story.

Play

Exploring all sides of a story is often best achieved with Play.

Log in to play. Track your picks, climb the leaderboards. Log in Sign Up
Predict 3 ways this could play out. Contrarian picks score more — points lock when the scenario resolves. Log in to play
Higher or Lower Two numbers from this story. Guess which is bigger. 5 rounds to set a streak. Log in to play
Connections Sixteen names from the news. Find the four hidden groups of four. Log in to play

People Involved

Organizations Involved

Timeline

March 2026 July 2026

3 events Latest: July 2nd, 2026 · 1 week ago
  1. HumanProgress amplifies the depolarization finding

    Latest Coverage

    The outlet HumanProgress highlights the study, framing it against the common assumption that flattering AI drives people apart.

  2. Economists release 'AI Sycophancy and Decisions'

    Research

    Conlon and Schwardmann publish a working paper finding that AI advice, on average, moves people away from their starting opinions across 30 decisions.

  3. Stanford researchers flag AI over-affirmation

    Research

    A Stanford study reports that leading AI models over-affirm users asking for personal advice, sharpening worries that flattery could distort judgment.

Historical Context

2 moments from history that rhyme with this story — and how they unfolded.

2011

'Filter bubble' theory takes hold (2011)

Activist Eli Pariser argued that personalized algorithms trap users in self-reinforcing 'filter bubbles,' feeding them only agreeable content. The idea spread fast and became a standard explanation for online division.

Then

The phrase entered common use and shaped how policymakers and platforms discussed recommendation systems.

Now

Later empirical work complicated the theory; several studies found personalization's effect on polarization was smaller or messier than the tidy metaphor suggested.

Why this matters now

Like filter bubbles, 'flattering AI divides us' is an intuitive story. This study is an early data point suggesting the intuition may not survive contact with measurement.

2020

Facebook feed-ranking experiments (2020)

Researchers worked with Meta to alter what users saw in their feeds during the 2020 U.S. election, then measured effects on attitudes. The goal was to test whether feed changes shifted political polarization.

Then

Some feed changes altered what people consumed but produced small or no measurable change in their broader political attitudes.

Now

The results pushed researchers toward more careful, experiment-based claims about technology and division, rather than sweeping assumptions.

Why this matters now

Both cases test a widely believed harm with a controlled experiment. Both found the real effect smaller or different than expected, a caution against assuming AI advice must deepen division.

Sources

(5)