AI is increasingly used not just to find answers, but also to produce work: drafting reports, summarizing meetings, creating analyses, and preparing presentations. Its speed is indeed impressive. However, there is one issue that is easily overlooked: work completed faster does not necessarily lead to better decisions.
Microsoft Research uses the term workslop to describe AI-generated content that appears professional but is inaccurate, incomplete, or not sufficiently useful. In the New Future of Work report, Microsoft states that 40% of employees in a survey in the United States reported receiving workslop in the previous month. ([microsoft.com](https://www.microsoft.com/en-us/research/blog/new-future-of-work-ai-is-driving-rapid-change-uneven-benefits/?utm_source=openai))
This phenomenon is significant because the workload does not truly disappear. It simply shifts to the next person who has to read, check, correct, or redo the results.
When a Neat Appearance Masks Weak Content
Imagine someone asks AI to create a meeting summary. The result has a title, bullet points, and convincing language. At first glance, the document seems ready to be sent. However, upon review, it turns out there are decisions that were never made, project names that are mixed up, or deadlines that are incorrectly inferred.
This is the hallmark of workslop: not always obvious errors, but results that are reasonable enough to be trusted yet too weak to be used directly.
Similar issues can arise in various tasks:
- Research summaries omit important limitations from the original sources.
- Data analyses explain patterns without checking if the data is complete.
- Email drafts sound polite but fail to answer the main questions.
- Strategy documents contain a lot of jargon but lack actionable steps.
- Program code appears correct but has not been tested in real conditions.
In other words, AI can speed up text production without automatically improving the quality of the thinking behind it.
Individual Productivity Does Not Necessarily Equal Team Productivity
This also explains why the experience of using AI can feel very productive for one person but exhausting for the team. The person creating the document may save 30 minutes. However, if three colleagues each need 15 minutes to check and correct the content, the organization does not actually save time.
Microsoft Research notes that the benefits of AI in the workplace heavily depend on roles, usage methods, and workflow changes. Another study from Microsoft in 2026 also found increased productivity and communication activity among high-intensity AI users, but cautioned that increased activity does not automatically equate to increased work value. ([microsoft.com](https://www.microsoft.com/en-us/research/publication/adoption-of-generative-ai-in-the-workplace-increasing-and-shifting-the-balance-of-productivity-and-communication-activity/?lang=ko-kr&utm_source=openai))
Analysis: a more sensible measure is not just how many documents are created or how quickly a draft is finished, but how many results can be used without repeated correction work.
Why Do People Often Skip Checking?
There are several practical reasons. First, AI's language usually sounds confident, even when the information is incomplete. Second, time pressure makes people more likely to accept results that are “good enough.” Third, the responsibility for checking is often unclear. Everyone assumes someone else has already checked it.
Research from Microsoft Viva summarized in February 2026 states that 60% of employees in related research reported skipping accuracy checks. The research also emphasizes the importance of metacognition, which is the ability to be aware of how we use AI and when the results need to be questioned. ([techcommunity.microsoft.com](https://techcommunity.microsoft.com/blog/microsoftvivablog/research-drop-fighting-ai-slop-with-meta-cognition-around-ai/4493933?utm_source=openai))
So, the issue is not just whether AI can make mistakes. Humans also need to build habits to avoid passing those mistakes into the workflow.
How to Prevent Workslop in Daily Work
1. Define 'Ready to Use' Standards
Before asking AI to create something, first define what the finished result should look like. For reports, for example, the final output should include sources, conclusions, assumptions, and recommendations. For emails, the final output should clarify who does what and when.
These standards help us evaluate outputs based on their usefulness, not just their linguistic neatness.
2. Separate Drafts from Decisions
Use AI to create drafts, options, or initial structures. Do not immediately consider the results as final decisions. Simple labels like draft for review can prevent unverified documents from circulating as official information.
3. Ask AI to Show the Basis for Its Answers
For work that relies on data or documents, ask AI to include sources, short quotes, assumptions, and uncertain parts. Instructions like the following are more useful than simply asking for the “best answer”:
Use only the data available in this document. Separate facts, assumptions, and recommendations. Mark unverifiable information.This instruction does not guarantee AI is always correct, but it makes the parts that need checking easier to find.
4. Check the Most Costly Parts if Incorrect
Not every sentence requires the same level of scrutiny. Prioritize numbers, names, dates, regulations, health claims, financial decisions, and code that will run in real systems.
If an error could lead to customers receiving incorrect information or the company making wrong decisions, human checking should not be eliminated.
5. Measure the Work That Needs to Be Redone
For two or three weeks, track how many times AI outputs need to be corrected, rewritten, or explained to others. This simple data is often more honest than the feeling that AI is “very helpful.”
If the number of documents increases but the correction work also grows, the workflow may need to be changed—not just by adding access to more expensive models.
What Does This Mean for Us?
AI remains highly useful. It can help kickstart work from a blank page, simplify long documents, identify initial patterns, and speed up repetitive tasks. However, those benefits arise when humans maintain control over goals, quality standards, and final decisions.
The most important change is not making AI sound smarter, but clarifying the work process: who requests, who checks, what sources are used, and when results are considered safe enough to pass on.
If not, we are merely replacing writing tasks with checking tasks for seemingly convincing writing. That is the crux of the workslop issue: AI saves time upfront but can reclaim its costs at the next stage.
Sources & Further Reading
- New Future of Work: AI is driving rapid change, uneven benefits — Microsoft Research
- New Future of Work Report 2025 — Microsoft Research
- Research Drop: Fighting AI Slop with Meta-Cognition around AI — Microsoft Viva
- Adoption of Generative AI in the Workplace — Microsoft Research
– Rio Yotto @rioyotto