Interaction Techniques that Encourage Longer Prompts Can Improve Psychological Ownership when Writing with AI

• Psychological ownership describes possessive feelings toward creations regardless of legal ownership, and writing often elicits it through investment of time, energy, and self.Source: §1 · p. 1
• Longer prompts previously increased ownership in story writing, but strict word minimums limited flexibility and did not test whether submission friction affects ownership.Source: §1 · p. 2
• The paper evaluates temporal and motor submission costs as alternatives that let users either write longer prompts or exert effort to submit shorter ones.Source: §1 · p. 2
• Across two experiments, friction increased prompt length and ownership, whereas AI-generated expansion suggestions increased length without further improving ownership.Source: §1 · p. 2

2.1 Psychological Ownership when Writing with AI

• Psychological ownership is important in AI-assisted writing because AI roles and suggestion-integration methods can either reduce or increase writers’ felt ownership.Source: §2.1 · p. 2
• Longer human-written prompts increased ownership in prior story-writing research, likely because adding details required more thought and effort [19].Source: §2.1 · p. 2

2.2 Friction in User Interfaces

• Friction can shift interaction toward slower, more deliberate thinking and has been used to encourage reflection, expert-mode selection, and reduced over-reliance on AI.Source: §2.2 · p. 2
• Joshi and Vogel’s press-and-hold technique increased prompt length, but its effect on psychological ownership was not evaluated [19].Source: §2.2 · p. 2
• The paper addresses the gap in evidence about whether friction-based techniques designed to encourage longer prompts also affect psychological ownership.Source: §2.2 · p. 3

3 Augmented Prompt Entry Techniques

• The study compares motor-cost and temporal-cost prompt-entry techniques with an unaugmented baseline, measuring whether friction increases prompt length and ownership.Source: §3 · p. 3

3.1 Motor Cost

• The motor technique uses a gauge that users fill either by writing more words or by repeatedly moving a vertical slider up and down.Source: §3.1 · p. 3
• The motor gauge width equals prompt words divided by a 150-word minimum, with slider movements compensating for the remaining proportion.For example, 15 words require 203 of 225 movements, while 150 words require none.Source: §3.1 · p. 3
• Motor progress can be completed intermittently, but editing resets the gauge to typed-word progress; submission becomes enabled only when the gauge is full.Source: §3.1 · p. 3

3.2 Time Cost

• The time technique adapts Joshi and Vogel’s method by requiring users below the word minimum to press and hold the submission button.Source: §3.2 · pp. 3–4
• The required hold duration scales against a 60-second maximum, and releasing the mouse resets all holding progress.Source: §3.2 · p. 4

4 Experiment 1: Submission Techniques

• Experiment 1 tested whether temporal and motor submission costs increase prompt length and psychological ownership in within-subjects short-story writing.Source: §4 · p. 4

4.1 Participants

• The analysis retained 29 valid participants after excluding six of 35 recruits for technical issues or suspected copy-pasting.Source: §4.1 · p. 4

4.2 Apparatus and Task

• The interface presented story instructions and noun prompts, a generated story, and a condition-specific submission control in a ChatGPT-like web application.Source: §4.2 · p. 4

4.3 Procedure

• Participants completed all conditions on laptops or desktops, answered experience questions after each story, and spent roughly 30 minutes in the experiment.Source: §4.3 · p. 4

4.4 Design

• Conditions were baseline, time, and motor; baseline always came first to reduce order effects, while time and motor order was randomized.Source: §4.4 · p. 4
• Measures combined composite psychological ownership with NASA-TLX workload ratings and logged word count, text similarity, and initial word count.Ownership averaged personal ownership, responsibility, personal connection, and emotional connection; its reliability was high (α = .94).Source: §4.4 · p. 5

4.5.1 Word Count.

• Time and motor produced longer prompts than baseline, with median word counts of 35 and 49 versus 15.Word Count more than doubled for time and tripled for motor.Source: §4.5.1 · p. 5

4.5.2 Text Similarity.

• Time and motor produced more prompt-output similarity than baseline, with median similarities of .37 and .36 versus .28.Text Similarity was moderately positively correlated with Word Count (rs = .48, p< .0001).Source: §4.5.2 · p. 5

4.5.1 Word Count.

• The techniques did not work for all participants: 55% showed no difference between initial and final Word Count for both time and motor.Among participants who increased Word Count, 14% increased by at least 50 words with time and 17% did so with motor.Source: §4.5.3 · pp. 5–6

4.5.4 Psychological Ownership.

• Word Count and Text Similarity were only weakly correlated with Psychological Ownership, both with rs = .28.This pattern suggests workload-related factors may have contributed more to ownership than prompt length and similarity alone.Source: §4.5.4 · p. 6

4.5.5 Task Workload.

• Time and motor increased Mental Demand, Physical Demand, Effort, and Frustration relative to baseline.Mental Demand and Effort were positively correlated with Psychological Ownership, with rs = .5 and rs = .42, respectively.Source: §4.5.5 · p. 6, Figure 5 · p. 6

4.6 Summary

• Temporal and motor submission costs encouraged longer prompts and increased psychological ownership.The techniques required additional mental and physical demand, effort, and frustration.Source: §4.6 · p. 6

5 Receiving Suggestions from AI

• AI-generated suggestions were designed to help users expand or improve prompts with more details.Each suggestion was phrased as a one-sentence question and generated with GPT 3.5 Turbo.Source: §5 · p. 6
• Suggestions appeared after one second of inactivity, remained fully visible for four seconds, and then faded over three seconds.Suggestions disappeared when users resumed typing and were not automatically incorporated as edits.Source: §5 · pp. 6–7
• The design preserved user choice by displaying suggestions without automatically modifying the prompt.This was intended to avoid reducing psychological ownership.Source: §5 · p. 6

6.1.1 Word Count.

• Time and motor produced median prompt lengths of 155 and 150.5 words, versus 19 for baseline.Both techniques exceeded baseline by more than 100 words.Source: §6.1.1 · p. 7, Figure 6 · p. 7
• Time and motor produced higher text similarity than baseline, with median values of .41 and .42 versus .32.Text Similarity was moderately positively correlated with Word Count (rs = .55, p< .0001).Source: §6.1.2 · p. 7

6.1.2 Text Similarity.

• Most participants increased their Word Count with time, whereas most motor participants were consistent and often already wrote at least 150 words.For time, 60% wrote more; for motor, 67% were consistent and 43% wrote at least 150 words.Source: §6.1.3 · pp. 7–8

6.1.3 Mechanics.

• Time produced the highest Psychological Ownership, followed by motor and baseline, with medians of 59.12, 49.38, and 39.5.Psychological Ownership was weakly correlated with Word Count (rs = .37) and Text Similarity (rs = .32).Source: Figure 7 · p. 7, §6.1.4 · p. 8

6.1.4 Psychological Ownership.

• Both interaction techniques increased Mental Demand, Physical Demand, Effort, and Frustration relative to baseline.Mental Demand and Effort were positively correlated with Psychological Ownership, with rs = .62 and rs = .5, respectively.Source: §6.1.5 · p. 8, Figure 8 · p. 8

6.1.5 Task Workload.

• Participants received more suggestions with time and motor than baseline, with median counts of 22 and 23 versus 4.5.Suggestion count correlated with Word Count (rs = .55), Text Similarity (rs = .45), Mental Demand (rs = .49), and Effort (rs = .54).Source: §6.1.6 · p. 8

6.1.6 Number of Suggestions.

• Most participants appreciated the suggestions, while 13% found them distracting or annoying.Suggestions helped participants who felt stuck and encouraged more thought about the prompt.Source: §6.1.6 · p. 8

6.2 The Effects of Receiving Suggestions

• Although suggestions increased Word Count and Text Similarity by over 100 words for time and motor, they did not further improve Psychological Ownership.Compared with no suggestions, no Psychological Ownership comparisons were significant.Source: §6.2 · p. 9, §6.3 · p. 9

7 Discussion

• Longer prompts increased psychological ownership, but AI suggestions added text without producing further ownership gains.The authors suggest ownership may plateau, or that workload and other factors beyond prompt length and text similarity may matter more.Source: §7 · p. 9
• Mental demand was the most important contributor to increased psychological ownership in the variable-importance analysis.Mental demand and effort were moderately positively correlated with ownership in both experiments, whereas word count and text similarity were only weakly correlated.Source: §7 · p. 9
• Increasing prompt length may encourage more thought and improve psychological ownership, motivating exploration of additional submission costs such as perceptual friction.The authors frame longer prompts as a valuable outcome while proposing further work on mechanisms that increase prompt length and mental demand.Source: §7 · p. 9
• These techniques may extend beyond story writing to other human-AI collaborations and non-AI contexts such as detailed image prompts or open-ended survey responses.The discussion proposes requiring detailed sketches for image generation and applying similar techniques to survey responses.Source: §7 · p. 9

7.1 Limitations

• Always presenting the baseline first may have primed participants through greater task familiarity or inspiration in later conditions.This design choice was intended to reduce order effects but could itself have influenced subsequent performance.Source: §7.1 · p. 9
• The second experiment’s findings may depend on displaying suggestions after one second of inactivity rather than using a longer delay or on-demand activation.The authors note that on-demand suggestions might increase ownership because receiving them requires an explicit user action.Source: §7.1 · pp. 9–10
• Real-world adoption may be constrained because users may resist added friction during urgent writing tasks, suggesting invocation at the user’s request.Feedback and explanations could help users understand the techniques, while user-invoked lockouts may better fit practice.Source: §7.1 · p. 10

8 Conclusion

• Friction-based techniques increased prompt length and psychological ownership, but AI-generated suggestions further increased prompt length without improving ownership.The techniques varied temporal and motor submission costs and were compared with baseline conditions, with and without suggestions.Source: §8 · p. 10

A Pilot Experiment

• Approximately 225 slider movements were estimated to match the temporal technique’s 60-second extreme-case duration.Five researchers produced a median single-movement duration of 267 ms, which was used to approximate 60,000 ms.Source: A Pilot Experiment · p. 11