AI ‘workslop’ may be destroying your productivity, suggests new Harvard study

Introduction

In recent months, a provocative idea has gained traction in the circles of technology, management, and organizational psychology: that AI is not only falling short of productivity promises, but may actively erode meaningful work. The term “workslop” has been coined to capture this phenomenon: superficially polished, AI-generated content that lacks real substance and ends up creating more work rather than less.

This essay explores the concept of workslop: its definition, prevalence, mechanisms by which it damages productivity, psychological and organizational consequences, and possible strategies to mitigate it. It also examines limitations of the current analysis and points to open questions.

Definition and Origins of “Workslop”

The term “workslop” arises in a recent article in the Harvard Business Review (in collaboration with researchers at BetterUp Labs and Stanford’s Social Media Lab). The authors define workslop as:

“AI generated work content that masquerades as good work, but lacks the substance to meaningfully advance a given task.” (Harvard Business Review)

In other words, it is when a colleague or subordinate hands you a memo, report, email, slide deck, or other deliverable that looks clean and plausible (i.e. properly formatted, coherent prose), but on inspection is missing critical context, coherence, insight, or requires reconstruction.

This concept builds on the broader notion of “AI slop” (i.e. low‐quality AI output in public domains), but applies it into the workplace. (Axios)

According to the study, workslop often hides deeper deficiencies—errors, lack of domain understanding, ambiguity or missing assumptions—that force the receiver to spend time deciphering, correcting, or redoing parts of the work.

The authors suggest that while many organizations have pushed for widespread adoption of generative AI, the anticipated gains in efficiency and creativity have often failed to materialize. One possible culprit is precisely this phenomenon: the proliferation of low-value AI output. (Harvard Business Review)

Prevalence and Empirical Findings

The study that introduced the concept of workslop draws on a survey of 1,150 full-time U.S. desk workers, carried out in August and September of the relevant year. (Axios)

Some of the key quantitative findings:

• 40% of respondents reported having received workslop in the past month. (Axios)
• On average, respondents estimated that 15.4% of the work content they receive qualifies as workslop. (Harvard Business Review)
• In terms of time costs, respondents said each incident of workslop required 1 hour and 56 minutes (≈ 116 minutes) of effort to deal with (i.e. reading, clarifying, rewriting). (Gizmodo)
• Translating the time cost into money, the authors estimate that workslop yields a hidden cost of about US$186 per employee per month (based on average U.S. salaries). (Gizmodo)
• Beyond pure time, there are reputational and relational costs: about half of the respondents said they viewed colleagues who sent workslop as less creative, capable, or reliable. (Axios)

Moreover, workslop arises in different directional flows within organizations:

• Peer-to-peer (40% of the time) (Harvard Business Review)
• From direct reports to managers (18%) (Harvard Business Review)
• From managers downward (16%) (Harvard Business Review)

The prevalence across functions seems broad, but the authors note that professional services and technology sectors may be disproportionately affected. (Workplace Insight)

These results suggest that workslop is not a marginal issue but a systemic friction point in organizations attempting to scale AI adoption.

Mechanisms: How Workslop Destroys Productivity

Why does workslop become so destructive rather than neutral? The study authors and related commentaries point to several interlocking mechanisms:

1. Burden shifting / cognitive burden transfer
   Because the AI output is imperfect, the receiver must spend effort interpreting, filling gaps, querying unclear assumptions, or reworking the output. In effect, the “work” is shifted downstream from the creator to the recipient. “It transfers the effort from creator to receiver.” (Gizmodo)
2. Illusion of progress / activity over impact
   The polished form of workslop can create a false impression that something meaningful has been produced, encouraging more superficial output. This in turn encourages volume over depth, and conflates doing with accomplishing. (Workplace Insight)
3. Quality control hidden costs
   Revising, editing, or double-checking AI output often consumes time that doesn’t appear in official metrics. In many organizations, the extra overhead is invisible to management but felt acutely by workers. (morningbrew.com)
4. Interpersonal and reputational friction
   Recipients of workslop may come to trust the submitter less, generate friction in follow-up, or require more meetings for clarification. Over time, these social costs can degrade collaboration and morale. (Harvard Business Review)
5. Encouragement of superficial AI use
   When organizations mandate AI usage without guidance, employees may default to “prompt, generate, submit” to satisfy quotas, rather than thoughtful use. The “blanket AI adoption” mentality risks turning AI into a checkbox rather than a tool for leverage. (Workplace Insight)
6. Amplification of errors
   AI models sometimes produce hallucinations, inconsistencies, or subtle inaccuracies. If not caught early, these errors propagate downstream, causing retracing of steps or rework. The more layers of dependency, the higher the risk. (TechRadar)

Taken together, these processes can invert the promise of AI: instead of “faster, better output,” organizations may find themselves mired in low-leverage edit loops.

Impacts: Individual, Team, and Organizational

On Individuals
Workers receiving workslop may feel frustration, ambiguity, or uncertainty about quality. The extra time spent cleaning up may crowd out higher‐value tasks (ideation, strategy, relationship building). Some respondents reported feeling annoyed (53%), confused (38%), or even offended (22%) upon receiving workslop. (Gizmodo)

Furthermore, repeated exposure could erode trust in colleagues or lead to micromanagement as people second-guess each other’s output.

On Teams and Collaboration
Frequent rework, clarification loops, and ambiguity create friction in team workflows. Meetings may expand to “figure out what that memo meant.” Shared assumptions become less reliable. Teams might fall back on more synchronous communication (calls, meetings) at the cost of asynchronous efficiency.

On Organizations / ROI of AI
One of the striking claims is that weak implementation and rampant workslop may help explain why many companies that have invested in AI see no measurable productivity gain. The HBR article points out that a prior MIT Media Lab study found 95% of organizations reported zero return from AI deployments. (Harvard Business Review)

If workslop is widespread, its hidden time costs could offset or wipe out the theoretical gains from automation. In monetary terms, even $186/month per employee can scale to millions of dollars in lost value across large enterprises.

Also, productivity metrics may misinterpret volume (number of memos, slides, reports) rather than quality or impact. In that sense, high activity but low effective output leads to organizational misalignment.

Strategies to Mitigate Workslop

If workslop is indeed a real threat, how might organizations respond? The authors and commentators suggest a range of strategies:

1. Model purposeful, high-quality AI use at leadership levels
   Leaders should avoid indiscriminate AI mandates. Instead, they can role-model how to use AI thoughtfully: as a drafting, ideation, or augmentation tool—not as a substitute for thinking. (Workplace Insight)
2. Set clear standards, guardrails, and review norms
   Define which types of tasks AI output may assist, and which tasks must involve human judgement. Use checklists, peer review, or “AI hygiene” protocols (e.g. “explain your prompt, show your sources”). (Workplace Insight)
3. Encourage a pilot mindset / experimentation
   Rather than massive rollouts, adopt controlled pilots. Collect data on time saved vs. time spent revising, and iterate. Use small teams to identify best practices before scaling. (Workplace Insight)
4. Train people in prompt engineering, evaluation skills, and fact-checking
   Often the problem isn’t AI per se, but poor prompting or lack of domain vetting. Helping employees become better editors and evaluators of AI output is key. (morningbrew.com)
5. Measure outcomes, not output volume
   Shift key performance indicators away from quantity (e.g. number of reports) to impact (e.g. decisions made, results delivered). This discourages superficial AI dumping. (Workplace Insight)
6. Implement human-in-the-loop oversight
   For critical outputs (e.g. client deliverables, decisions, strategy memos), require review by a domain expert. Don’t let AI output go unchecked. (morningbrew.com)
7. Audit “fixing time” invisibly absorbed
   Use time tracking or qualitative feedback to capture how much time is spent editing AI output. Bringing that hidden cost into visibility helps adjust expectations. (Workplace Insight)

Critical Reflections and Caveats

While the notion of workslop is compelling and resonates with anecdotal experience, there are several caveats:

1. Survey, not experimental validation
   The evidence is largely based on cross-sectional survey data rather than controlled field experiments. The measurement of “time spent fixing” is self-reported and may suffer from recall bias or attribution errors.
2. Causality vs. correlation
   While respondents believe they spend time correcting low-quality AI output, it’s possible that weak clarity from the sender (regardless of AI) or ambiguous requirements contribute significantly. AI may be a visible scapegoat.
3. Context dependence
   The impact of workslop likely depends heavily on domain, task complexity, and team maturity. For routine, formulaic tasks, AI output may be good enough with minimal correction. For complex, strategic tasks, human insight still dominates.
4. Selection and sampling biases
   The article mentions an “ongoing survey” and lacks detailed disclosure of sampling frame or response rates. Some critics have argued the article functions more as marketing or thought leadership rather than rigorous empirical research. (Pivot to AI)
5. Risk of overgeneralization
   Not every AI-generated output is workslop. Many organizations are already using AI meaningfully (e.g. coding assistants, summarization, data augmentation). The line between helpful drafting and harmful slop may blur.
6. Changing AI quality over time
   As models improve (better factuality, context awareness, controllability), the quality of AI output may reduce the incidence of workslop. What is slop today may be quite usable tomorrow.
7. Psychological framing / narrative appeal
   The concept of workslop has strong rhetorical force—a catchy word that easily communicates a shared frustration. But rhetorical appeal doesn’t guarantee empirical robustness.

Thus, while the study raises a red flag worth paying attention to, it should be interpreted as an invitation for further controlled research rather than a definitive verdict.

Conclusion

The concept of workslop—AI output that looks good superficially, but lacks depth and becomes a burden rather than a boon—is a useful lens for understanding why many organizations struggle to achieve productivity gains with AI. The reported prevalence and hidden costs suggest that naive or blanket AI mandates may backfire.

However, to move beyond cautionary tales, organizations must invest in discipline: clear protocols, human oversight, training in prompt and editing craft, and metrics that reward impact over volume. As AI tools mature, and as human–AI collaborative practices evolve, the hope is that we can overcome the workslop challenge and realize the productivity promise without drowning in low-value artifacts.