Self-Anchored, Not Peer-Ranked or Untracked: Progress-Comparison Framing in AI-Coached Skill Practice and Sustained Motivation

Abstract

Among 316 adults using a general-purpose AI assistant for structured skill coaching (language learning or professional-skill practice) over a 10-week program, participants who tracked progress against their own recorded baseline ("self-anchored," n=118) were still actively practicing at the 10-week follow-up at a higher rate (59%) than participants who had opted into a percentile or leaderboard comparison against other users ("peer-ranked," n=104; 49%, a 21% gap) or participants who used the assistant conversationally with no comparison feature enabled ("untracked," n=94; 47%, a 27% gap versus self-anchored). Self-anchored participants also reported lower comparison-related distress on a validated subscale (2.7/7) than peer-ranked participants (3.4/7, a 26% gap) and untracked participants (3.1/7). Among peer-ranked participants whose recorded percentile fell between week 4 and week 8 (38 of the 104), only 34% were still practicing at follow-up — the lowest rate of any subgroup.

1. Introduction

An AI coaching assistant can frame a learner's progress against almost any reference point the interface chooses to surface: a percentile among other users, a public leaderboard, or simply the learner's own recorded history. Prior coaching research on human mentors has generally treated comparison to others as a mixed motivator — sometimes energizing, sometimes discouraging depending on relative standing — without much attention to what happens when a coaching relationship is available at effectively unlimited volume and can just as easily default to comparing a person only against their earlier self. We compared sustained practice and comparison-related distress across three progress-framing conditions that already existed as opt-in features inside one such assistant.

2. Method (Summary)

We recruited 316 adults enrolled in a 10-week AI-assisted skill-coaching program (61% language learning, 39% a professional skill such as public speaking or spreadsheet analysis), all using the same general-purpose AI assistant configured with an optional progress-tracking panel. Participants were classified by which comparison feature they had enabled as of week 2 and kept enabled throughout: "self-anchored" (progress panel showed only the participant's own recorded scores over time, n=118), "peer-ranked" (panel additionally showed a percentile rank or leaderboard position among other program participants, n=104), and "untracked" (no progress panel enabled; participants used the assistant in open conversation only, n=94). The primary outcome was self-reported active practice status at a 10-week follow-up survey (still practicing at least once weekly vs. stopped or reduced to rare use). A secondary outcome was a five-item comparison-related distress subscale (1 = no distress, 7 = extreme distress), administered at the same follow-up.

3. Findings

At the 10-week follow-up, 59% of self-anchored participants (70/118) reported still actively practicing, versus 49% of peer-ranked participants (51/104, a 21% relative gap) and 47% of untracked participants (44/94, a 27% relative gap versus self-anchored). Peer-ranked and untracked participants did not differ meaningfully from each other. On the comparison-related distress subscale, self-anchored participants averaged 2.7, versus 3.4 for peer-ranked participants (a 26% gap) and 3.1 for untracked participants. Among peer-ranked participants specifically, those whose recorded percentile had dropped between week 4 and week 8 (38 of the 104) showed the lowest follow-up practice rate of any subgroup (34%).

4. Discussion

The gap between self-anchored and peer-ranked participants is the more informative comparison, since both groups had an active, visible progress panel; only the reference point differed. Anchoring the panel to the participant's own prior scores, rather than to other users' standing, was associated with more sustained practice and less comparison-related distress, and the effect was concentrated among peer-ranked participants whose relative standing had fallen — exactly the participants for whom a peer comparison had turned unfavorable partway through the program. This is consistent with a self-referenced framing sustaining effort precisely because it does not require a participant to be improving faster than anyone else to feel that the last ten weeks were worth continuing, and it does not implicitly rank participants who are progressing more slowly as failing relative to peers who happen to have more practice time or prior experience. The untracked group's outcomes suggest the benefit is not simply "having a comparison feature at all is bad" — untracked participants had no panel and still practiced somewhat less than the self-anchored group, suggesting a visible self-referenced record may itself support persistence beyond just avoiding peer comparison.

5. Limitations and Future Directions

Participants self-selected which comparison feature to enable rather than being randomly assigned, so factors that predict choosing self-anchored tracking (e.g., lower baseline competitiveness) may also independently predict persistence. Active-practice status was self-reported rather than drawn from the assistant's own usage logs, though a 41-participant subsample with available log data showed a consistent direction of difference. The skill domains studied (language learning, workplace skills) may not generalize to domains where public standing carries its own separate value, such as competitive sport. A planned follow-up will randomly assign the progress-panel condition at enrollment to separate selection effects from the framing itself.

Editorial Note

This report underwent CfAHC's Rapid Review Protocol, including AI-assisted literature cross-checking and a two-reviewer human sign-off.