What Employees Really Do with ChatGPT at Work: Writing Tops, Editing Dominates
OpenAI: 27% of daily ChatGPT messages are work-related (47% last year). Writing leads-mostly editing and translating; HR should enable safe workflows and decision support.

What employees really use AI for at work - and how HR can turn it into real productivity
A new OpenAI report offers a clear snapshot of how employees actually use ChatGPT on the job. In June 2025, 27% of daily ChatGPT messages were work-related, down from 47% a year earlier - a shift driven by changing usage patterns within existing users, not a wave of new non-work users.
The headline: writing is the dominant work use case. And most of that "writing" isn't starting from zero - it's editing, improving, translating, and tightening what people already wrote.
The top use cases, by share of work messages
- Writing: 40% (about two-thirds is editing, critiquing, translating)
- Practical guidance: 24.1%
- Information-seeking: 13.5%
- Technical help: 10.0%
These patterns are consistent across very different occupations. The common thread: decision support. The report argues ChatGPT boosts output by improving decision quality, especially in knowledge-heavy roles. Think "co-pilot" that advises rather than a "co-worker" that fully replaces tasks.
What this means for HR
- Focus enablement on editing and decision support. Provide templates, checklists, and prompts that help employees refine existing drafts and weigh options faster.
- Standardize safe, repeatable workflows. Because use cases look similar across jobs, HR can roll out common practices for writing, research, and task guidance that apply org-wide.
- Position AI as assistive. Set expectations: human judgment stays in the loop; AI accelerates thinking and drafting.
High-impact HR applications you can deploy now
- Writing/editing: polish job ads, offer letters, policy updates, and employee comms; convert long docs into summaries; translate messages for global teams.
- Practical guidance: draft onboarding checklists, coaching tips for managers, interview rubrics, meeting agendas, and 1:1 frameworks. Route legal or sensitive issues to the right experts.
- Information-seeking: generate FAQs from internal docs, summarize survey findings, and surface relevant policy excerpts. Always verify critical outputs.
- Technical help: get spreadsheet formulas, text classifiers for open-ended survey responses, and simple automation ideas for HRIS workflows - without exposing PII.
Guardrails HR should enforce
- Data protection: prohibit sharing PII, health data, or confidential financials. Use approved tools with enterprise controls.
- Human review: require sign-off for all external comms, policy changes, and legal-adjacent content.
- Prompt standards: provide approved prompts for common tasks (job posts, policy edits, internal memos) and a checklist for verifying output.
- Attribution and bias checks: cite sources for factual claims; scan for biased language in job ads and performance inputs.
- Logging and monitoring: track use by function and task type to spot wins, gaps, and risk patterns.
A 30-day enablement plan
- Week 1: inventory use cases by team; identify 5-7 high-frequency tasks per function.
- Week 2: publish policy, PII rules, and a prompt library; ship 10 reusable templates.
- Week 3: run a pilot with 2-3 teams; host office hours; collect examples and pitfalls.
- Week 4: measure time saved and quality gains; iterate templates; prep org-wide rollout.
Metrics that matter
- Time saved per draft, analysis, or summary
- Quality: fewer revisions, higher clarity scores, fewer tone issues
- Adoption: % of employees using approved tools and templates
- Rework rate: AI-assisted drafts needing major edits
- Risk: incidents involving sensitive data or factual errors
- Sentiment: manager and employee feedback on usefulness
Workforce impact
Adoption is growing and some firms are publicly considering future headcount changes as AI integrates further. HR's role: guide responsible use, elevate decision quality, and upskill people so they stay productive and relevant.
Upskill your teams
If you need structured, role-based enablement and prompt libraries for HR, see these resources:
Source
This article is based on a new OpenAI analysis of internal ChatGPT message data on personal and workplace use. For related research, visit OpenAI Research.