What Workers Actually Use AI For

Cover Memo

This report draws on the AI@Work survey (CHIP50 Wave 38.6), a survey of 23,462 U.S. adults (weighted n = 23,485) fielded June 21–July 13, 2026. All percentages are weighted to U.S. Census population targets and are population-representative; unweighted respondent counts are reported as a reliability guide only.

These are interim estimates. Wave 38.6 launched June 21, 2026 and data collection is ongoing; the figures in this report reflect responses received through July 13, 2026 and will be revised in a later release.

The report analyzes the "AI task" battery (ai_use_closed_1, nine professional tasks) among employed respondents who use AI for their work (ai_work ≠ "Never"; unweighted n ≈ 6,700). The battery's stem asks how often the respondent uses AI tools for each purpose generally, not strictly at work; because the nine items are professional and knowledge-work tasks, and because the base is restricted to workers who use AI on the job, the results are read here as those workers' task profile. Two notes on measurement: the battery is coded 1 = Not at all … 4 = A lot (reverse of its on-screen order), confirmed against the data; and "some or a lot" throughout means the top two response categories combined.

Key Takeaways

Introduction

When people picture AI "at work," they often picture a coder with a copilot. The reality among American workers is both broader and more ordinary. Asked how often they use AI tools across a range of professional tasks, workers who use AI on the job describe a general-purpose cognitive assistant — one they turn to first for figuring things out and thinking through problems, and only secondarily for producing polished text, and least of all for software.

This report maps that task profile using Wave 38.6, restricted to employed respondents who use AI for their work. It shows which tasks are common and which are niche, how intensively workers rely on AI for each, and where the biggest gaps open up by education. The headline is a useful corrective: the most common workplace uses of AI are troubleshooting and brainstorming, writing code is the rarest, and the tasks that most separate the more- from the less-educated are exactly the knowledge-production tasks — writing, summarizing, and analysis — rather than the everyday problem-solving that nearly everyone uses AI for.

AI at work is a thinking aid, not a coding tool

The most common workplace uses of AI are open-ended cognitive help — troubleshooting and brainstorming — while writing code is the least common of the nine tasks measured.

Among workers who use AI on the job, troubleshooting technical issues is the single most common application: 74% use AI for it at least sometimes, and 30% "a lot." Close behind is brainstorming or problem-solving (72% at least sometimes, 30% "a lot"). Only then come the content and analysis tasks — drafting and editing emails (66%), summarizing long documents or notes (66%), and data analysis (66%), followed by math or numeric calculation (60%) and writing reports, essays, or papers (60%). Transcribing or taking meeting notes (53%) and, at the bottom, writing or fixing computer code (50%) are the least common.

The coding result is the one most worth underscoring. Despite the popular image of workplace AI as primarily a programming aid, it is the rarest of the nine uses, and a third of AI-using workers (33%) never use it for code at all. What the pattern describes instead is a tool people reach for to understand and unblock — to fix the printer, debug a spreadsheet, think through an approach — as much as to produce. AI's foothold at work is cognitive before it is generative.

Comparative context. The ordering matches independent national surveys: Pew found that among workers who have used AI chatbots at work, researching or finding information (57%) and editing written content (52%) ranked at the top while "help with analyzing data or writing computer code" sat near the bottom at 27%; Ipsos found the same shape among AI users, with looking up information or getting recommendations first at 80% and data analysis or programming lowest at 37%; and Quinnipiac put research first at 51%, with data analysis at 27%¹²³. Note that Pew and Ipsos each bundle data analysis together with coding in a single item, whereas the battery here separates them — and they diverge, at 66% for data analysis against 50% for code. One estimate does not line up: Pew put "coming up with new ideas" at 35% among AI-using workers, well below the 72% for brainstorming or problem-solving here, a gap best read as differing question wording and differing bases rather than either survey being wrong.

The knowledge-work tasks split by education; the everyday ones don't

More-educated workers use AI much more for writing and analysis — but AI's use for troubleshooting, math, and brainstorming is broadly shared.

Where the education gap is large, it is concentrated in knowledge-production tasks. Workers with a graduate degree are far more likely than high-school-educated workers to use AI for summarizing documents (75% vs. 59%, a 16-point gap), writing reports and essays (70% vs. 54%, +16), drafting emails (74% vs. 61%, +13), and data analysis (74% vs. 62%, +12). These are the tasks of the credentialed desk job, and AI adoption for them tracks education closely.

For other tasks, the divide nearly disappears. Troubleshooting technical issues is used almost identically across the education spectrum (77% of graduate-degree holders vs. 75% of high-school graduates, a 3-point gap); math or numeric calculation is if anything slightly more common among the less-educated (62% vs. 63%); and brainstorming shows only a modest gap (77% vs. 69%). In other words, AI's role as an everyday problem-solving and figuring-things-out aid is broadly shared across the workforce, while its role as a writing-and-analysis productivity tool is concentrated among the more educated — the same knowledge-work divide that runs through AI adoption overall (the socioeconomic divide is an education gap in behavior), but visible here at the level of specific tasks.

The occupation divide: data jobs drive code and analysis; troubleshooting is universal

Whether a worker uses AI for the more technical tasks depends heavily on the kind of job they do — but troubleshooting is used across the board.

Comparing workers whose jobs involve data processing or analysis with those who do manual or physical labor sharpens the picture. Data and analytical workers use AI far more for the technical tasks: 75% use it for data analysis (versus 62% of manual-labor workers) and 59% for writing or fixing code (versus 47%) — the two tasks where occupation matters most. They also lead on drafting emails (73% vs. 62%). This is the mechanism behind the report's code and data-analysis numbers: those uses are concentrated in analytical desk jobs rather than spread evenly across the workforce. (These two job types are the poles of a non-exclusive work_context variable, so a large middle — managers, client-facing, and other roles — sits between them; the contrast marks the ends of a spectrum, not a clean binary.)

Troubleshooting is the exception that proves the rule. Manual-labor workers use AI to troubleshoot technical issues almost exactly as often as everyone else — 74%, essentially the workforce average and only a few points below data workers' 78%. AI's role as a universal fix-it and figure-it-out aid reaches blue-collar and white-collar workers alike; it is the writing, analysis, and coding uses that concentrate among knowledge workers.

Conclusion

The everyday reality of AI at work is less dramatic, and more interesting, than the coding-copilot image suggests. Workers who use AI reach for it first to troubleshoot and to brainstorm — to think, unblock, and understand — and only then to write and analyze; coding is the rarest use, not the defining one. That makes workplace AI look less like a specialist tool bolted onto technical jobs and more like a general-purpose assistant spreading across the ordinary cognitive work of many jobs at once.

The education pattern sharpens the point. The tasks that separate more- and less-educated workers are precisely the ones tied to formal knowledge production — writing reports, summarizing, analysis — while the more universal uses, like troubleshooting and calculation, are shared across the workforce. As AI use deepens, the tasks to watch are these knowledge-work applications, where adoption is both highest and most unequal.

Appendix A — Methods

Data source. AI@Work survey — CHIP50 Wave 38.6 (Civic Health and Institutions Project / COVID States Project), the AI/employment supplement. Total unweighted n = 23,462; weighted n = 23,485. Fielded June 21–July 13, 2026. Data collection was ongoing at the time of this extract — figures reflect responses through July 13, 2026 and are preliminary. Estimates are weighted to U.S. Census population targets.

Measure and base. The task battery ai_use_closed_1 (nine items) asks "How often do you use AI tools for the following purposes?" It is fielded to respondents who have used AI tools at all. This report restricts the base to employed respondents who use AI for their work (ai_work ∈ {tried once/twice … several times a day}, i.e., ≠ "Never"), applied as a data filter; unweighted n ≈ 6,700 (item bases vary slightly with non-response). The battery's stem is not work-specific, but the base restriction and the professional nature of the items support reading the results as these workers' at-work task profile.

Coding. The battery is coded 1 = Not at all, 2 = Not much, 3 = Some, 4 = A lot — the reverse of the on-screen display order — confirmed against the data (the "not at all" share is highest for the rarest task, code-writing, which is only coherent under this direction). "Some or a lot" denotes categories 3 + 4; "a lot" is category 4.

Estimation. Weighted shares within each task, overall and by education. Overall worker figures are computed by aggregating the weighted response counts across education groups (the education breakdown was the vehicle for applying the ai_work filter, because the derived work-use indicator could not be used directly as a grouping variable in this tool). No regression is reported. Cells with unweighted n < 10 are suppressed; the "Some High School or Less" group is thin on several items (unweighted n ≈ 80–100) and is used only in the aggregate, not reported as a standalone column.

Reproducibility. Figures map to Appendix B tables; values are copied verbatim from the underlying tabulations. Wording and coding sourced from the Qualtrics instrument via the project's Wave 38.6 verified-updates crosswalk.

Appendix B — Data tables

Appendix C — Question wording (verbatim)

• ai_use_closed_1 — "How often do you use AI tools for the following purposes?" (A lot / Some / Not much / Not at all — stored 1 = Not at all … 4 = A lot). Items: Writing or editing emails and letters; Writing or editing professional reports, essays, academic papers; Summarizing long documents, articles, or meeting notes; Brainstorming ideas or problem-solving; Data analysis; Math or any numeric calculation; Writing or fixing computer code; Transcribing or taking notes during meetings; Troubleshooting technical issues (e.g., "how do I fix my phone, printer, TV etc.").
• ai_work (base restriction) — "How often do you use AI for your work?" (Never; I tried once or twice; About once a month; About once a week; Multiple times a week; Once or twice a day; Several times a day). This report includes all responses except "Never."