An AI tool can produce a draft in seconds while leaving you with another hour of decisions. Which version should you use? Are the references real? Does the spreadsheet formula work?
Sometimes the total job still becomes much easier. Sometimes the fast first step hides extra work. The way to tell is to look at the finished result and the effort needed to trust it.
This is a guide to checking your workflow. It is not evidence that AI inevitably damages attention, memory, or your brain.
Measure the whole task
Define what “done” means before you start. For an email, it might mean an accurate message ready to send. For a spreadsheet, it might mean checked calculations and a result you can explain.
Count the time spent preparing the request, reading the answer, checking facts, correcting errors, and putting the result where it belongs. A quick draft is only one part of that process.
As a hypothetical example, a ten-minute draft followed by forty minutes of corrections is a fifty-minute task. Whether that is a gain depends on how long your usual method takes and the quality of both results.
Notice when a tool creates more open loops
A suggestion for another version, a new feature, or an additional research question can be useful. It can also take you away from the job you meant to finish.
Look for specific patterns: several unfinished chats, repeated requests for almost identical drafts, or switching tools without resolving the original question. These are workflow observations, not a diagnosis of an attention problem.
What the research does—and does not—show
A 2025 study by Microsoft Research and collaborators surveyed 319 knowledge workers about their use of generative AI. Greater confidence in AI was associated with less reported critical thinking. Participants also described work shifting toward checking information and integrating responses.
This was a survey of reported behavior. It does not establish that using AI causes permanent loss of reasoning ability, or tell you whether a particular tool improves your own work. It gives you a reason to examine how you check results.
Give AI one clear job
Instead of asking a tool to “handle this project,” start with an output you can evaluate. For example: “Turn these notes into an agenda. Preserve the decisions already made and list unresolved questions separately.”
Before requesting alternatives, decide what is wrong with the current result. If you cannot identify a problem that matters, another version may add comparison work without improving the outcome.
Keep a place for ideas to revisit later. You can finish the agenda without immediately testing a new scheduling app or redesigning your entire meeting process.
Keep the checking step deliberate
Read important source material, test calculations against known examples, and confirm that the answer addresses the task. Ask the tool to identify assumptions and missing information, but verify those statements too.
AI can help with creative work and planning as well as routine administration. The useful boundary is your ability to evaluate the output and remain responsible for the decision, not a blanket ban on using it for demanding tasks.
Try a small workflow experiment
Over a week, choose a few comparable tasks and record:
- What you intended to finish.
- Which parts AI helped with.
- Total time, including checking and corrections.
- Errors, omissions, or rework you found.
- Whether you finished and could explain the result.
When practical, compare a similar task using your usual method. This is an informal check, not a controlled study; differences in task difficulty, familiarity, and interruptions can affect the result.
Use your notes to decide what to keep. A workflow may be worthwhile because it improves quality or makes a difficult task manageable, even when the time saving is small.
Set a stopping point and protect time to think
Decide what needs checking and what level of polish the task deserves. You might stop after the facts are verified, the intended reader can understand the result, and no important question remains unresolved.
Where your responsibilities allow it, silence nonessential notifications and group routine requests into a few work periods. Keep necessary channels available. The point is to reduce interruptions you can control, not to make yourself unreachable.
If learning is the goal, try outlining your own answer before asking for help, request a hint, or explain the result without looking at it. Choose a method that lets you notice what you understand rather than merely collecting a finished answer.
A prompt for a more focused session
Help me complete this one task: [task]. A finished result must meet these checks: [checks]. Use the material I provide, identify missing information, and flag claims that need verification. Do not suggest extra tools or side projects unless they are necessary to complete it.
Review the result against those checks and stop when the task is finished. For deciding whether a new feature is worth trying at all, use our guide to evaluating AI updates.
AI disclosure: This article was drafted and revised using AI.