Skip to main content
. 2023 Dec 25;25:e51229. doi: 10.2196/51229

Table 2.

Word counts for ChatGPT-generated abstracts versus human-written abstracts.

Abstract ChatGPT (mean, SD) Human (mean, SD) Difference (95% CI) P value
Structured abstract (4 subheadings)

Background 59.70 (9.81) 65.20 (20.07) –5.50 (–20.76 to 9.76) .45

Methods 49.20 (8.63) 60.00 (30.87) –10.80 (–33.39 to 11.79) .31

Results 58.80 (15.39) 85.20 (15.02) –26.40 (–40.75 to –12.05) .001

Conclusions 71.70 (15.16) 40.90 (20.62) 30.80 (13.65 to 47.95) .002

Total words 239.40 (21.76) 251.30 (24.94) –11.90 (–33.98 to 10.18) .27
Structured abstract (3 subheadings)

Background 53.30 (16.06) 50.50 (14.06) 2.80 (–11.44 to 17.04) .68

Results 66.20 (10.42) 135.60 (33.47) –69.40 (–94.10 to –44.70) <.001

Conclusions 92.80 (12.73) 34.90 (13.03) –64.90 (–96.94 to –32.86) <.001

Total words 212.30 (21.78) 221.00 (28.43) 57.90 (45.75 to 70.05) .45
Unstructured abstract (total words) 142.90 (12.05) 207.60 (36.27) –64.70 (–91.63 to –37.77) <.001