Did people follow AI-labelled sports advice more than an expert's? Not significantly—but confidence rose
In an online experiment with 273 analysed participants, identical numerical anchors moved predictions by about 4.75 percentage points whether labelled AI or expert. The AI label slightly raised confidence, but no real AI generated the advice.
The short version
- What happened
- The peer-reviewed online experiment analysed 273 German-language participants across football, handball and field hockey matches from the 2024 Paris Olympics.
- Who it affects
- People can feel more certain after seeing an AI-labelled number even when the label does not improve the evidence behind it. In consequential settings, extra confidence without better calibration can make weak advice harder to question.
- What remains unproven
- Peer-reviewed online experiment
Keep in mind
Peer-reviewed online experiment
273 analysed participants
German-language sports sample
71.7% male
Men's Olympic events
Fabricated AI and expert labels
No real advice system
No preregistration
Low-stakes predictions
Exploratory subgroup results
Source analysis · 2 linked sources. Primary analysed source date: 30 October 2026.
Keep following the evidence
Join The Impact Brief founding list for selected AI reports, their effect on people and the evidence limits. Read a sample first.
Read the full report · 4 min The evidence, methods, implications and limitations
The experiment isolated the source label from the number itself
A peer-reviewed online experiment found that people adjusted sports predictions toward a suggested probability whether the suggestion was labelled as coming from artificial intelligence or a human expert. The researchers recruited German-language participants around the 2024 Paris Olympics. Of 384 people who accessed the study, 44 did not meet age or consent rules, 64 completed less than half the matches and three participated after the deadline, leaving 273 in the analysis.
Participants estimated winning probabilities for football, handball or field-hockey matches. Each person saw 24 matches, while two flawed items were excluded, leaving 70 analysed matches across the three sports. The within-person conditions included no anchor and high or low probabilities labelled AI or expert. Crucially, the source labels were fabricated: the number was not produced by a real model or a named specialist. High and low anchors came from tournament-winner betting odds and were separated by about 15 percentage points.[1][2]
The number anchored judgement; the AI label mainly changed confidence
Mixed-effects models accounting for participant and match found an average 4.75-percentage-point anchoring effect. The estimated anchoring under the expert label was 1.32 points lower than under AI, but that difference was not statistically significant. The main result is therefore similarity: participants followed the numerical suggestion to a comparable degree under both labels, not evidence that AI advice was reliably more persuasive.
The source label did affect how sure participants said they felt. Predictions accompanied by an AI label raised confidence by about 0.1 on a one-to-six scale compared with the expert label, and the model comparison was statistically significant. Asked which information was most helpful, 51.1% chose neither source, 17.3% chose AI and 29.0% chose the expert; 2.6% did not answer. Exploratory subgroup differences were not designed for formal between-group tests and should not be treated as confirmed demographic effects.[1]
The distinction that matters
What changed under the AI and expert labels
The study held the numerical anchor structure constant and changed the attributed source.
Anchor
High versus low suggested probability
Predictions moved by about 4.75 percentage points
Source
AI label versus expert label
Anchoring difference was not statistically significant
Confidence
AI-labelled versus expert-labelled advice
Self-reported confidence rose by about 0.1 on a six-point scale
Reality check
Labels and advice were fabricated
No real AI or expert performance was compared
A label effect in sport does not establish trust in real automated decisions
The sample averaged 32.81 years old, was 71.7% male and reported a median sports expertise of seven on a ten-point scale. It was self-selected, sports-interested and limited to German-language users and men's Olympic events. The study was not preregistered. Because the advice was fabricated, it cannot show whether people recognise, challenge or calibrate confidence to the quality of a real model. Match predictions also carry fewer consequences than medical, employment or public-service decisions.
For publishers and product teams, the useful warning is narrow: attaching an AI label to a number may raise confidence even when it does not significantly change how far people follow the number. Interfaces should reveal evidence, uncertainty and track record rather than rely on a source badge. Confidence in the assessment would rise with preregistered studies using real, performance-matched AI and expert advice, more representative samples, consequential decisions and measures of whether users learn to discount inaccurate suggestions over time.[1][2]
Watch the background, then check the evidence
Illustrative photo · not a video frame
The Royal Society · 44 min 28 sec · Background explainer
Is this the AI future we want? | The Royal Society
YouTube loads only when you choose; playback starts when you press play.
A Royal Society discussion with Madhumita Murgia and biomedical engineer Alison Noble on AI’s potential in healthcare, science and culture, recorded at the 30 June 2026 event.
Evidence check: A discussion of opportunities and risks, including future possibilities. It is not evidence that every application described is available or clinically proven.
Published 2026-07-01. Source checked 14 October 2026. Check captions on the publisher’s video page.
Check the healthcare evidencePhoto: Daniel Sone / National Cancer Institute / Unsplash · Licence
Continue the story
Read next
Continue exploring human judgement: compare the finding and its evidence limits.
Can people still tell AI from humans? They struggled with saved snippets—but got every GPT-3.5 live test right
Across six constrained tasks, people often misclassified machine output. Active judges perfectly identified GPT-3.5 in one live setup, while a later Llama follow-up again produced mistakes.
Read the next reportExplore the full source and evidence record
What this means for people
- People can feel more certain after seeing an AI-labelled number even when the label does not improve the evidence behind it. In consequential settings, extra confidence without better calibration can make weak advice harder to question.
Global context
The experiment concerns European participants judging international sport. Attitudes toward AI and expert authority vary by culture, domain and institutional trust, so the source-label effect should not be assumed to generalise to other countries or high-stakes decisions.
What the evidence does not yet show
- Peer-reviewed online experiment
- 273 analysed participants
- German-language sports sample
- 71.7% male
- Men's Olympic events
- Fabricated AI and expert labels
- No real advice system
- No preregistration
- Low-stakes predictions
- Exploratory subgroup results
What to watch next
- Preregistered replication
- Real AI and expert outputs
- Performance-matched advice
- More representative samples
- High-stakes domains
- Confidence calibration
- Learning after wrong advice
- Independent replication
Living evidence record
Impact record IAI-1X0YTIR
Evidence stage
Studied
Confidence
Supported
Reporting basis
Source analysis
Independent or research support
Present
Record status
Monitoring
Last checked
30 October 2026
Source trail
2 direct sources across 2 source types.
People impact
Documented in this record.
Uncertainty
Limits and next checks are explicit.
Stages describe the evidence available—not whether a technology is good or bad. See the public method.
Related-source reporting disclosure
This record analyses 2 linked source records around the same underlying development. The extra records add method, date or context, but they do not by themselves constitute independent replication of every performance claim or predicted outcome.
Evidence trail
Sources used for this report
Links checked 30 October 2026
Research question, topics and regions
Whether people anchor sports-outcome estimates differently when identical numerical information is labelled as coming from AI or a human expert
Topics: Trust in AI · Expert advice · Anchoring · Sports predictions · Confidence · Human judgement
Regions: Germany · Europe
This report is labelled source analysis. We summarise and analyse source material in our own words; company statements remain attributed claims until independently supported. Translated summaries preserve the meaning of the original source and link back to it. Read our editorial standards.
Reader commentary
What does this mean where you are?
Tell us what you have seen at work, in your community or in everyday life. A question is welcome too. No account is needed; we review notes before publishing them.
Explore commentary across the portal →Published reader notes
0No published reader notes yet. You can start the evidence-led discussion above.
Prefer a private correction or response? Contact the newsroom.