A slower start helped people write better with AI—and the benefit carried into the next task
A Microsoft Research preprint randomised 297 US workers to an AI writing interface with or without planning prompts. Quality gains were small, and the study did not test long-term workplace performance.
The short version
- What happened
- In a randomised experiment, 297 US-based Prolific participants wrote two professional briefs, producing 594 essays; only the treatment group saw planning scaffolds during the first task.
- Who it affects
- Workers often experience AI tools as pressure to accept a fast first draft. A planning scaffold could preserve more of their judgement, but only if organisations allow the extra effort and measure whether it reduces later rework.
- What remains unproven
- The preprint was not preregistered and all authors were affiliated with Microsoft Research, which has a commercial interest in AI workplace tools.
Keep in mind
The preprint was not preregistered and all authors were affiliated with Microsoft Research, which has a commercial interest in AI workplace tools.
Primary essay scoring used an LLM judge whose absolute agreement with the human benchmark was moderate, ICC 0.456.
The sample was US-based, English-first and recruited through Prolific; the tasks were short online briefs rather than embedded workplace projects.
Only 159 of 297 participants returned for the three-day recall survey, and the recall confidence intervals were wide.
Source analysis · 1 linked source. Primary analysed source date: 22 October 2026.
Keep following the evidence
Join The Impact Brief founding list for selected AI reports, their effect on people and the evidence limits. Read a sample first.
Read the full report · 3 min The evidence, methods, implications and limitations
The interface asked for a plan before a draft
Participants used the same GPT-5.3 chat-and-canvas system, but one group first received dynamic controls that asked them to state the goal and break the job into parts. They spent longer on task one—27.84 minutes versus 16.35—but produced slightly higher-scoring briefs. In task two, both groups lost the scaffold; the previously scaffolded group still held a smaller quality advantage and used more deliberate prompts.[1]
Better work felt less helpful
The extra planning raised cognitive load and reduced perceived helpfulness even as measured output improved. That tension matters for workplace rollouts: ease and satisfaction can reward an interface that gets to a draft quickly, while a short pause to specify purpose, audience and checks may leave the worker with a stronger result. The experiment cannot tell whether the effect survives routine deadlines or repeated use.[1]
The distinction that matters
What the scaffold changed
The intervention inserted two planning steps before AI drafting.
Goal
State purpose, audience and constraints
Makes success criteria explicit
Decompose
Split the assignment into checkable parts
Creates places for review and revision
Transfer
Remove the scaffold on a second task
Tests whether the prompting habit carries over briefly
The evidence supports a design test, not a universal rule
Teams can test lightweight planning prompts against their own work: compare quality, rework, factual checking and worker control, not just completion time. The study’s primary judge was another language model; human ratings of 107 essays pointed in the same direction but were imprecise. A three-day topic-recall signal also had wide confidence intervals and high follow-up loss.[1]
Watch the background, then check the evidence
Illustrative photo · not a video frame
UNESCO · 11 min 27 sec · Background explainer
UNESCO Talks: Interview with 2024 Nobel laureate Professor Daron Acemoglu
YouTube loads only when you choose; playback starts when you press play.
Economist Daron Acemoglu discusses how choices about AI affect workers, inequality and society. Pair the interview with our comparison of task exposure, hiring and skills evidence.
Evidence check: An expert interview from 2024, offering a perspective on policy and work rather than a measured forecast of job losses.
Published 2024-10-14. Source checked 14 October 2026. Check captions on the publisher’s video page.
Read the AI and jobs guidePhoto: Dylan Gillis / Unsplash · Licence
Continue the story
Read next
Continue exploring knowledge work: compare the finding and its evidence limits.
Should AI wait until people think first?
In an unreviewed study of 398 adults across five countries, a writing assistant withheld direct generation until users stated a position and supporting argument. The design shifted effort earlier and improved one conditional error-classification measure, but did not significantly improve overall judgment accuracy.
Read the next reportExplore the full source and evidence record
What this means for people
- Workers often experience AI tools as pressure to accept a fast first draft. A planning scaffold could preserve more of their judgement, but only if organisations allow the extra effort and measure whether it reduces later rework.
Global context
Writing conventions, job autonomy and AI access differ across workplaces and countries. Employers should evaluate scaffolds with the people doing the work and avoid treating a US online writing task as a universal productivity estimate.
What the evidence does not yet show
- The preprint was not preregistered and all authors were affiliated with Microsoft Research, which has a commercial interest in AI workplace tools.
- Primary essay scoring used an LLM judge whose absolute agreement with the human benchmark was moderate, ICC 0.456.
- The sample was US-based, English-first and recruited through Prolific; the tasks were short online briefs rather than embedded workplace projects.
- Only 159 of 297 participants returned for the three-day recall survey, and the recall confidence intervals were wide.
What to watch next
- Longer preregistered workplace trials
- Human-rated quality, factual accuracy and rework
- Whether benefits persist for weeks rather than one subsequent task
- Effects across languages, occupations and levels of AI experience
Living evidence record
Impact record IAI-1ASZUPF
Evidence stage
Studied
Confidence
Supported
Reporting basis
Source analysis
Independent or research support
Not yet
Record status
Monitoring
Last checked
23 October 2026
Source trail
1 direct source across 1 source type.
People impact
Documented in this record.
Uncertainty
Limits and next checks are explicit.
Stages describe the evidence available—not whether a technology is good or bad. See the public method.
Single-source reporting disclosure
This record analyses one direct source. It can establish what arXiv published or reported, but it is not independent corroboration of every performance claim or predicted outcome. The confidence label will change only when broader evidence is added.
Evidence trail
Sources used for this report
Links checked 23 October 2026
Research question, topics and regions
Whether prompts for goal specification and task decomposition improve AI-assisted writing and transfer after the prompts disappear
Topics: Knowledge work · Writing · Metacognition · Interface design · Workplace skills
Regions: United States · United Kingdom · Global
This report is labelled source analysis. We summarise and analyse source material in our own words; company statements remain attributed claims until independently supported. Translated summaries preserve the meaning of the original source and link back to it. Read our editorial standards.
Reader commentary
What does this mean where you are?
Tell us what you have seen at work, in your community or in everyday life. A question is welcome too. No account is needed; we review notes before publishing them.
Explore commentary across the portal →Published reader notes
0No published reader notes yet. You can start the evidence-led discussion above.
Prefer a private correction or response? Contact the newsroom.