Metacognitive Scaffolds Create Productive Friction for AI-Assisted Knowledge Work

Authors: Viktor Kewenig, Lev Tankelevitch, Sean Rintel, Advait Sarkar, Abigail Sellen
Venue: arXiv, October 2026

AThe study

Randomised online experiment with 297 US adults. Each wrote two professional briefs with AI help, in random order, and answered a survey after each.

Writing task 1Writing task 23 days later Scaffold group153 people AI chat with the scaffold Standard AI chat 3 days later: recall test159 of 297 returned Comparison group144 people Standard AI chat Standard AI chat

BThe tool

  1. Type a requestThe writer asks the AI chat to write a professional brief.
  2. Set goalsThe AI generates fields to fill in first.message to conveytarget readerreader outcomewriting objectivescope and depth
  3. Break the task downControls specific to this task.analytical approachesstructurecritical considerations
  4. AI draftsThe draft appears in an editable document, shaped by the writer's choices.
  5. Revise the controlsThe writer can ask for changes to the controls in plain language.

Steps 2, 3 and 5 are the scaffold. The comparison group had the same chat and document without them.

CWhat changed

Writing task 1only the scaffold group had the scaffold

  • ↑Scaffold group: briefs were rated better.Scored by an AI judge on a seven-item rubric, checked against human raters. A small effect, g = 0.29.
  • ↕Scaffold group: writing felt like more work.Self-reported workload, 4.71 against 3.96 out of 10. The paper sees this extra effort as productive friction.
  • ↓Scaffold group: the AI felt less helpful, and they felt less in control.Self-reported, on 1 to 5 scales.

Writing task 2nobody had the scaffold

  • ↑Scaffold group: briefs were still rated better.Same AI judge and rubric. A small effect, g = 0.21.
  • ↑Scaffold group: prompts stated goals and broke the task into steps more often.Prompts coded without knowing the group, 14.43 against 11.07 out of 39.
  • –Scaffold group: the work felt about as hard as for the comparison group.Self-reported workload, 4.08 against 3.93 out of 10.

3 days laterrecall test

  • ↑Scaffold group: more could recall what each brief was about.64.6% against 45.6% for the first brief, and 54.4% against 38.0% for the second. Only 159 of 297 people returned, so the estimate is imprecise.

↑ better than the comparison group   ↓ worse than the comparison group   – no reliable difference   ↕ mixed, a cost and a benefit

DKey takeaway

A metacognitive scaffold that prompts people to set goals and plan before AI drafts can improve their work and later AI use. Productive friction can make work feel harder yet build better thinking habits.

Made for this site with AI assistance from the verified research brief. It is not the paper's own figure.

Abstract

Generative AI imposes new metacognitive demands on knowledge workers as their work shifts from production to planning, evaluation, and orchestration. In a randomized, between-subjects experiment on AI-assisted writing (297 workers, 594 essays), we test dynamically-generated metacognitive scaffolds supporting goal specification and task decomposition. We find that scaffolds improve writing quality, with this boost persisting in short-lag transfer when scaffolds are removed. We also find evidence of improved topic recall three days later. These effects are consistent with observed changes in how people learn to prompt AI systems after exposure to scaffolding. However, scaffolded participants rate the AI as less helpful and more cognitively demanding even as they perform better, suggesting that subjective ease can understate the value of support. Our findings make the case for metacognitive scaffolds in AI-assisted knowledge work, revealing a productive friction worth designing for as workers increasingly delegate work to agentic AI.

Research brief

By Viktor Kewenig, Lev Tankelevitch, Sean Rintel, Advait Sarkar, and Abigail Sellen.

Takeaway: Metacognitive scaffolds, AI-generated controls for setting goals and breaking down a task before AI drafts, can improve people's work and their later prompting. The authors call the trade-off productive friction, because the same support also feels less helpful and more demanding.

Why this matters

What was known
Interfaces that help people state goals and break tasks down for AI already exist. But they have mostly been judged on how usable they are and how well they steer the model, not on the quality of the work people produce. Yet despite this progress, prior work has primarily examined proximal interaction outcomes, such as usability and model steerability. (§1, p. 2)
What this paper adds
We test whether such scaffolds improve the work itself. We look at output quality while they are available, whether gains carry into a later task without them, and whether people better recall the topics of their work. Methodologically, we complement the largely small-scale, within-subjects prior work with a large-scale, between-subjects evaluation designed to isolate scaffolds' causal effects with greater generalizability. (§2.5, p. 6)
Who it matters to
Designers of AI tools for knowledge work, and researchers who evaluate AI's effects on people's thinking, since how a tool feels and how well people perform with it can point in opposite directions. This suggests that studies evaluating AI's impact on cognition, including system evaluations, should include objective measures of performance and cognitive processes (§5.2, p. 21)

Research questions

Method and data

Design
Randomized between-subjects experiment. Each participant wrote two AI-assisted professional briefs in counterbalanced order. The treatment group had the scaffold for Task 1 only, and neither group had it for Task 2. Not preregistered.
Prototype
A browser-based writing tool with an AI chat panel and an editable document canvas, connected to GPT-5.3. In the scaffold version, after the user types a high-level request, the system generates a goal panel with editable fields (the message to convey, the target reader, the desired reader outcome, the primary writing objective, and scope and depth). Once these are set, it generates a task-decomposition panel with controls specific to the task (analytical approaches, structure, and critical considerations to address). Users can ask for changes to the controls in plain language, and their selections shape the AI's draft. The controls are generated from the user's own request and earlier choices, not from a fixed template. The control group used the same chat and canvas without the two panels.
Three panels. The writing interface, with prompt controls for goals and planning beside a chat and document canvas. The study design, with two writing tasks and a recall test 3 days later. A summary of key findings.
A, The AI-assisted writing interface with the metacognitive scaffold: after an initial request (1), dynamically generated controls prompt goal specification (2) and task decomposition (3) before a draft is generated (4); controls could be revised in natural language (5). From the paper, Fig. 1, p. 1.
Setting
Online and self-paced. Participants wrote two professional briefs: a city proposal on a 4-day work week for municipal employees, and a consulting brief on launching hybrid work, with a bonus for the top five.
Participants / data
297 US-based Prolific participants with English as their first language (153 scaffold, 144 control), who wrote 594 essays. 159 completed the recall follow-up.
Duration
One self-paced session of up to 80 minutes, plus a follow-up survey three days later.
Analysis
Essays scored on a seven-item rubric by an LLM judge (GPT-5.4) on de-formatted text, checked against two human raters on 107 essays. Human raters scored 117 essays, and 10 of these calibrated the judge. Mixed-effects models with baseline covariates. Task 2 prompts were coded blind by an LLM coder, checked against three human coders on 60 sequences, for goal clarity and task decomposition (ordinal regression). Topic recall by logistic mixed models with multiple imputation for attrition. Workload by NASA-TLX.

Findings

Questions this paper answers

Does making people plan before they use AI improve their work?
It can. The authors argue goals and task structure set early shape all later work. In a randomised experiment, knowledge workers prompted to state goals and break down a writing task wrote better briefs than with standard chat, and kept the advantage on an unprompted later task. (See §4.1.1, p. 14.)
Can people learn to prompt AI better, or does the tool have to keep helping?
Some of the skill transfers. After one task with a planning scaffold, people's own prompts stated goals more clearly and broke work into steps more often. Whether this lasts for weeks or months has not been tested. (See §4.2, p. 16.)
Why might a tool that improves your work feel less helpful?
Because the effort is visible and the benefit is not. Planning support asks you to analyse the task and make decisions, which raises felt workload, while the better quality shows only in the result. We call this productive friction. (See §4.4.3, p. 18; §5.2, p. 21.)
Does using AI harm people's memory of their work?
Compared with a standard chat interface, a planning scaffold was associated with better recall of writing topics three days later, among the half of participants who returned. That does not show AI is harmless to memory compared with working unaided, since everyone compared had used AI. (See §5.1.3, p. 20.)

Quantitative results

MeasureResultWhere
Essay quality (1–4 rubric composite), Task 1 with scaffold vs controlb = 0.089, 95% CI [0.023, 0.155], P = .008§4.1.1, p. 13
Essay quality, Task 2 (neither group had the scaffold)b = 0.070, 95% CI [0.004, 0.136], P = .037§4.1.1, p. 13
Standardized effect size, Task 1 and Task 2Hedges' g = 0.29 and g = 0.21§4.1.1, p. 13
Time on Task 1, scaffold vs control (minutes)27.84 versus 16.35§4.1.1, p. 14
Human raters' composite advantage on the 107-essay benchmark, Task 10.440 [0.157, 0.724], P = .002Table 2, p. 15
LLM judge agreement with human ratersICC(A,1) = 0.456, Pearson r = 0.662§4.1.1, p. 14
Task 2 prompt score, goal clarity + task decomposition (of 39), previously scaffolded vs control14.43 versus 11.07§4.2, p. 16
Odds of a higher combined prompt score after prior scaffoldingOR = 1.779 [1.173, 2.698], P = .0067Table 3, p. 16
Participants whose Task 2 prompts copied the task text119 of the 297 participants (40%)§5.3, p. 21
Topic recall three days later, Task 1, scaffold vs control (complete cases)64.6% versus 45.6%; OR = 5.07, 95% CI [1.21, 21.25], P = .026§4.3.1, p. 17
Topic recall, Task 2, previously scaffolded vs control (complete cases)54.4% versus 38.0%; OR = 4.24, 95% CI [1.02, 17.57], P = .046§4.3.1, p. 17
Topic recall after multiple imputation, Task 1 and Task 2OR = 3.24, P = .032; OR = 3.06, P = .042§4.3.2, p. 17
Perceived AI helpfulness (1–5), scaffold vs controlb = -0.342, 95% CI [-0.510, -0.174], P < .001§4.4.2, p. 18
Perceived control (1–5), scaffold vs controlb = -0.251, 95% CI [-0.486, -0.017], P = .036§4.4.2, p. 18
NASA-TLX workload (1–10), Task 1, scaffold vs controlM = 4.71 versus 3.96; b = 0.745, P < .001§4.4.3, p. 18
NASA-TLX workload, Task 2 (no scaffold), not reliableM = 4.08 versus 3.93; P = .429§4.4.3, p. 18
Task 2 prompts that copied the task text, by condition56 of 153 previously scaffolded (36.6%) vs 63 of 144 previously unscaffolded (43.8%)§4.2, p. 16

Contributions and implications

Limitations

›

Concepts

Dynamic metacognitive scaffold
Interface controls, generated by AI from the user's own request, that prompt goal specification and then task decomposition before a draft is written. Unlike a template: Importantly, the scaffold was not a fixed prompt template: the options shown were generated and updated from the participant's current request, prior selections and current phase of work. (§3.2.1, p. 8)
Productive friction
Our name for the trade-off we found: support that increases felt effort and dampens perceived helpfulness while improving outcomes (§1, p. 3)
Short-lag transfer
Whether a benefit persists into a later task in the same session once the support is taken away. Our test: in a subsequent writing task in the same session where we remove the scaffolds, we evaluate how scaffold exposure affects written output quality and prompting behavior (§1, pp. 2–3)
Goal clarity and task decomposition
The two dimensions we coded in participants' own prompts, blind to condition: goal clarity (GC), which captures how explicitly participants stated the intended outcome, audience, constraints and success criteria (seven indicators), and task decomposition (TD), which captures whether they divided the work into concrete steps or components (six indicators) (§3.4.2, p. 10)
Cognitive offloading
Delegating production to AI can entail offloading, in the standard definition: recruiting external resources to reduce a task's cognitive demands (§2.2, p. 4)

Figures

Three panels. The writing interface, with prompt controls for goals and planning beside a chat and document canvas. The study design, with two writing tasks and a recall test 3 days later. A summary of key findings.
A, The AI-assisted writing interface with the metacognitive scaffold: after an initial request (1), dynamically generated controls prompt goal specification (2) and task decomposition (3) before a draft is generated (4); controls could be revised in natural language (5). From the paper, Fig. 1, p. 1.
Dot plot of essay quality on a 1 to 4 scale. Scaffolded participants score slightly higher than unscaffolded ones in Task 1, and previously scaffolded participants remain slightly higher in Task 2.
Fig. 3. Essay quality by condition and task. Scores are means across the seven 1-4 rubric ratings. Faint points show observed essay-level scores; larger markers show condition means with 95% confidence intervals. Neither group had the scaffold in Task 2. From the paper, Fig. 3, p. 14.
Three dot plots. Scaffolded participants rate AI helpfulness and perceived control lower after Task 1, and report higher NASA-TLX load in Task 1 but not in Task 2.
Fig. 6. Subjective experience by condition. A, AI helpfulness and B, perceived control after Task 1; C, unweighted NASA-TLX load after Tasks 1 and 2 (higher, more load). From the paper, Fig. 6, p. 19.

How to cite

Viktor Kewenig, Lev Tankelevitch, Sean Rintel, Advait Sarkar, and Abigail Sellen. 2026. Metacognitive Scaffolds Create Productive Friction for AI-Assisted Knowledge Work. arXiv preprint arXiv:2610.30945. https://doi.org/10.48550/arXiv.2610.30945

This brief was prepared with AI assistance and published by Sean Rintel. It is not the authors' published text. Page numbers refer to this PDF.

BibTeX

@article{kewenig2026metacognitive,
author = {Kewenig, Viktor and Tankelevitch, Lev and Rintel, Sean and Sarkar, Advait and Sellen, Abigail},
title = {Metacognitive Scaffolds Create Productive Friction for AI-Assisted Knowledge Work},
year = {2026},
month = {October},
url = {https://doi.org/10.48550/arXiv.2610.30945},
journal = {arXiv preprint arXiv:2610.30945},
doi = {10.48550/arXiv.2610.30945},
}