We have seen the ways generative AI is now firmly baked into daily office life, with an impertinent byproduct starting to gum up our workflows: workslop. The term was popularized by researchers at BetterUp Labs working with the Stanford Social Media Lab in Harvard Business Review to describe AI-generated outputs that appear polished but do not actually advance a given task.
This is not harmless busywork. In their studies, 40% of full-time U.S. employees polled admitted that they had taken home workslop in the past month. The same group of researchers points to an even broader adoption paradox: Although experimentation is widespread, a large majority of organizations report that they are not gaining returns from AI—often because shiny deliverables conceal missing logic, context or accuracy.
What Workslop Looks Like in Everyday Team Outputs
Workslop is almost always dressed for a Vinegar Strokes-esque success. It’s like imagine a strategy memo, with nice heavy heading weights and confidence and none of the situational nuance that you might encounter. It recycles generalities, fudges specifics and quietly makes things up. It can be an executive summary that refers to studies no one has ever seen or a customer email that reads well but misquotes prices and delivery times.
Engineers liken it to code recommendations that compile but promote insecure patterns. Digital marketers see persona descriptions and content calendars that read more like they could describe anyone in any industry in any quarter of the year. Policy drafts with verbose boilerplate but no jurisdictional basis are thrown over to legal and compliance teams. The binding tie: All were finished products that appear complete, or appear nearly so—but would also be steps backward for the team.
Why It Spreads Within Teams and Undermines Trust
Speed incentives and ambiguous AI norms are a recipe for disaster. Under deadline pressure, a staffer pastes in a prompt, gets back a clean document and sends it along without context. The recipient stands in for editor and fact checker, and at times co-author. BetterUp Labs calls this a downstream burden shift that compounds hidden work.
Data from McKinsey’s Global AI research indicate that generative tools are widely adopted across functions, but many companies do not have consistent quality control nor measure business impact. Into that vacuum, teams conflate volume and value. Work mounts and trust collapses as co-workers are left to second-guess everything that arrives in their inbox.
The Expense You Don’t Plan For: Rework and Risk
Workslop creates a rework tax. Savings in time on a 1st draft are more than offset during review, verify and rework. Multiply that by marketing, sales, engineering and operations, and throughput decreases even as volume goes up. It also amplifies risk: hallucinated citations, misnamed data and obligation gaps can lead to being audited and reputational damage.
Regulatory frameworks are catching up. Both the NIST AI Risk Management Framework and ISO/IEC 42001 focus on governance, documentation, and controls. Companies that neglect these fundamentals often find themselves in agony as incident numbers escalate, approval lines bottleneck and shadow IT spreads. In other words, bad AI hygiene now becomes a business continuity problem, not an etiquette problem.
How to Spot and Stop It with Practical Guardrails
Check for telltales.
- Does the document not reference facts that are local, a range of data points, or stakeholder names?
- Are quotes generic or unattributed?
- Do the numbers round out too nicely or clash with internal dashboards?
- Are references untraceable?
- Are there shifts in spelling between American and British English or domain-specific lexicon?
Institute friction where it matters. Attribution: Require it, and if AI help was used, disclose that fact too. Ask for the prompt, template, and sources. Enforce a short “human validation” note: what they checked, against what system of record, by whom. Develop a simple, lightweight checklist based on the four Cs—context, correctness, citations and consequences—and require students to pass it in order to submit.
Set measurable guardrails. Properly scope the use of each function: The right to use your AI is not “all or nothing.” It’s important to define acceptable use in different functions, to identify places where AI is off-limits (e.g., legal advice, personal customer data), where it should be used with review and where it can best create value unsupervised. Give reviewers spot-check tools, such as sampling facts against your CRM or data warehouse. Track time for rework and error rates, so leaders can see the real cost curve.
Upgrade prompts and training. Strong outputs start with strong inputs: specific task definition, audience, tone/mode, canonical sources for data and success criteria.
- Work to deliver team-specific prompt libraries, and reduce the use of “one-size-fits-all” templates.
- Undertake regular red-team sessions to uncover failure paths.
We’re shown time and again by Stanford’s AI Index that model performance is task-dependent; expect drift, verify it.
Leaders Need to Role Model High-Intent AI Usage
Executives should “show what good looks like” from explicit problem framing, transparent use of AI and human judgment visible in the loop. Release a brief AI playbook, pair it with enablement sessions and align incentives around quality output versus volume of output. Gartner has projected a boom in artificially produced enterprise content; your differentiator will be governance, discrimination and not raw generation of speed.
The message to teams is clear, AI should shorten time-to-insight not extend time-to-repair. When in doubt, add context, cite sources and test assumptions. If work cannot be trusted upon entering, it is not work—it is workslop.
