Here is a number that should change how you think about your AI setup.
Of the 6,000 full-time digital workers surveyed for the Glean Work AI Index 2026 - a research report co-authored with academics from Stanford, UC Berkeley, Harvard, and five other major institutions - 87% say they use AI at work. Seventy-five percent say it makes them more productive, saving them roughly 11 hours a week.
And yet only 13% say their organization is performing significantly better as a result.
Eleven hours a week saved. And it's not showing up in the business. Where is it going?
Researchers call it botsitting: the work required to make AI actually usable. Feeding it context it doesn't have. Checking outputs before you send them. Re-running prompts that went sideways. Cleaning up the confident-sounding errors that slip through. Workers in this survey spend an average of 6.4 hours a week doing this - close to a full working day, every week.
But here's what I want to focus on, because the problem story has been told. The interesting question is: what are the 13% doing differently?
They're Not Buying More Tools
The Glean researchers were direct about this.
"There's a reflex to solve every problem by buying more AI, adding more tools, or pushing people to use AI whether or not it helps," the report's authors write.
The 13% - the organizations actually performing significantly better - are not doing that. They're not burning more tokens or building adoption dashboards. They're "doing the harder work of treating AI as a work-design problem, not a procurement one."
That phrase is worth sitting with: a work-design problem, not a procurement one.
Procurement asks: which tools should I buy?
Work design asks: how does work actually get done around here, and where does AI fit into that - if it fits at all?
Most businesses are asking the first question. The 13% are asking the second.
What "Work Design" Looks Like in Practice
The Glean data breaks this down in a way that's useful.
High-performing AI organizations start with the work. They identify a specific task or workflow, understand what it actually requires, and then figure out whether a tool can improve it - rather than buying a tool and finding work for it to do. This sounds obvious. It is also apparently not what most businesses are doing.
They give AI the context it needs. This one is underappreciated. The report found that 53% of workers say the critical information they need to do their jobs is not accessible through their AI systems. That gap - between what AI needs to perform well and what it actually has access to - is the root cause of most botsitting. You spend time babysitting AI outputs because the AI didn't know enough about your business, your customers, or your standards to get it right the first time.
Workers in "context-rich" AI organizations - where AI has access to the right information - are 64% less likely to feel worn out by their AI tools. That's not a small difference. That's the difference between a tool that helps and a tool that creates work.
They know when to skip AI entirely. High AI achievers in this study are 18% more likely to deliberately choose NOT to use AI on certain tasks. This is the opposite of "use AI for everything." It's a judgment call - this task benefits from AI, this one doesn't - and making that call correctly is a skill that most AI advice doesn't teach.
The Training Gap Is Real
90% of workers in high-performing AI organizations say their employer provides enough AI training and support.
At organizations where AI is not moving the needle, that number drops to 52%.
For small businesses, "training" probably doesn't mean a corporate learning module. But the principle holds: do you and your team know how to use these tools effectively, or did you subscribe and figure the rest would follow? Do you know the specific prompting patterns that get good results from your AI writing tool? Do you know which tasks your AI agent reliably handles and which ones it tends to get wrong?
Most small business owners are learning this on the fly, through trial and error, often without time to document what's working. That's not a character flaw. It's a setup problem - and it's exactly the kind of thing that separates a tool that helps from one that sits unused.
The Honest 30-Minute Audit
If you want to know whether your AI setup falls into the 13% or the 87%, here are three questions worth asking about each tool in your stack.
1. What specific task does this tool handle, and how did work get done before it?
If you can't answer this clearly, the tool probably landed in your stack because it seemed useful, not because it was solving a defined problem. That's not always wrong - but it's a good signal that you may be the one doing most of the work.
2. Does this tool have access to the information it needs to do its job?
Your AI email responder: does it know your tone, your pricing, your policies, your common objections? Your AI content tool: does it understand your brand voice and your audience? If the answer is "sort of" or "I explain it in the prompt each time," you have a context gap - and you're paying for it in time every week.
3. When was the last time this tool saved you time you can actually name?
Not a "general feeling" that things are running smoother. A specific instance: this task used to take me 45 minutes, now it takes 10. The 2026 QuickBooks AI Impact Report found that more than half of small businesses measure AI ROI by a general feeling rather than a measurable outcome. A general feeling is not a data point. If you can't name the time saved, you may be in the 87%.
One Thing Worth Doing This Week
Pick one AI tool in your current stack. One.
Map the workflow it's supposed to improve - write it out, even roughly. Identify the three most important pieces of context that tool needs about your business that it might not have. Give it that context deliberately, explicitly, in a system prompt or saved instructions.
Then run the workflow for two weeks and compare.
You're not building a technology strategy. You're running an experiment. And a well-run experiment on one tool teaches you more than having six tools running in the background while you half-manage all of them.
The 13% of businesses seeing real results from AI aren't necessarily the ones who started first or spent the most. They're the ones who stopped treating AI as a product to buy and started treating it as a question to answer: how does this change the way the work actually happens?
That question is free. And it's the one most people aren't asking.
Sources: Glean Work AI Index 2026, co-authored with researchers from Stanford, UC Berkeley, Harvard, Notre Dame, Emory, UC Santa Barbara, and UNC Charlotte. Forbes coverage by Joe McKendrick (June 12, 2026). SBE Council 2026 Small Business Tech Use Survey (April 2026). QuickBooks AI Impact Report (Intuit/University of Chicago, 2026).