GPT-4O
GPT-4O has been studied in 2 papers since 2024: 1 trial in people; 1 in the past year. The newest paper, Nov 26, 2025: “Preference-Aligned Options from Generative AI Compensates for Age-Related Cognitive Decline in Decision Making”.
2 papers study it3 findingsUpdated
“AI use significantly reduced perceived choice difficulty across age groups”
None of its findings points the other way.
- 1 lab in 1 country
- Funder: Japan Society for the Promotion of Science
- Human trial1
- In silico1
Studies in people
1 study gave GPT-4O to people and measured what changed.
down: Decreases perceived choice difficulty
no change: Has no effect on choice satisfaction
Trial130 people
Older adults
arXiv (Cornell University), Nov 26, 2025
What did studies of GPT-4O find?
2Studies of GPT-4O measured 2 outcomes; the most studied are choice satisfaction (1 study) and perceived choice difficulty (1).
Choice satisfaction
no changein humans1
1 study
Choice satisfaction
no changein humans1
GPT-4O has no effect on choice satisfaction in humans.
“choice satisfaction remained unchanged”
Human trial
arXiv (Cornell University)Preprint
Perceived choice difficulty
downin humans1
1 study
Perceived choice difficulty
downin humans1
GPT-4O decreases perceived choice difficulty in humans.
“AI use significantly reduced perceived choice difficulty across age groups”
Human trial
arXiv (Cornell University)Preprint
What is GPT-4O linked with?
1Studies link GPT-4O with 1 measure, most often mild cognitive impairment (1 study).
Mild cognitive impairment
upin silico1
1 study
Mild cognitive impairment
upin silico1
GPT-4O predicts mild cognitive impairment in silico (specificity 100.0%, NPV 50%).
“Using a cutoff of >=5 points for mild cognitive impairment, specificity was 100.0%, and negative predictive value was 50%.”
In silicocutoff >= 5 pointsn = 57
GSA 2024 abstract