Automated feedback helps writers estimate their personal input more accurately when co-creating text with artificial intelligence, but it leaves their actual contribution level unchanged. People often assume that feeling ownership over a draft reflects genuine effort, yet users claim strong psychological ownership based purely on perceived contribution rather than true output. Reviewing an automated breakdown of ideas, details, and wording prompts users to type longer instructions without taking over the physical text generation.
The software analyzes user interaction logs alongside the final submitted text to assign percentage scores across three separate creative layers. Like a manager directing an assistant, participants supply high-level concepts while the computer produces the specific phrasing and granular sentences. When users receive their initial report, they adjust subsequent keystrokes by expanding their text prompts during the next assignment. Human input remains concentrated in the overall ideas, whereas the machine handles the bulk of the details and wording.
Researchers tested this dynamic by enrolling 121 participants in a two-part experiment requiring pitches for different mobile applications. The team measured self-reported contributions against the automated attribution scores across ideas, details, wording, and overall performance. Participants aligned their self-ratings on three of the four metrics in the second task, but submissions with more distinct ideas showed smaller human contribution shares.
These findings enable the design of language models that support critical user engagement by revealing how people evaluate shared work. The data also demonstrates that the visible properties of a finished text provide a poor guide to who actually wrote it.