ChatGPT Work Tutorial: 6 Role-Based Workflows, Prompt Templates & Automation Recipes (2026)

ChatGPT Work is not Chat with extra buttons. It plans multi-step paths across 1,400+ integrations, uses Computer Use on desktop, and can run for hours. Three habits separate productive teams from quota burners:

Pick the Right Mode: Chat / Work / Codex

The new desktop app unifies three modes. Using the wrong one wastes quota:

Desktop vs Web: Where to Run Each Workflow

Most teams do not fail because ChatGPT Work lacks features. They fail because nobody translates launch coverage into repeatable workflows:

• Prompt micromanagement: Writing step-by-step instructions defeats Work mode and inflates usage — the same workflow can cost 5× more when over-specified.
• Unconnected plugins: Tasks that reference "the CRM" instead of @Salesforce pull wrong context or stall mid-run.
• Skipped Plan Mode: High-stakes deliverables go live without reviewing delete, send, or overwrite steps.
• Desktop sleep kills Scheduled Tasks: Laptops that sleep at 6:30am never fire the morning dashboard brief you configured.
• Mode confusion: Running multi-hour cross-app projects in Chat mode, or asking Codex to build marketing decks.
• No quality baseline: Teams automate tasks they cannot verify, then blame the model when outputs need heavy rework.

The sections below fix each pain with a framework, role templates, and automation recipes you can paste today.

Every role follows the same skeleton. Run it manually three times before you schedule anything:

1. 01
   Connect plugins — authorize Gmail, Slack, Drive, CRM, and any source your task needs.
2. 02
   Write goal + output format — state the deliverable, not the clicks.
3. 03
   Review Plan Mode — use the checklist below before confirming execution.
4. 04
   Steer mid-flight — pause and correct when context drifts; attach explicit source files.
5. 05
   Accept deliverable and iterate — note what worked, trim steps, then consider Scheduled Tasks.

Work mode prompt formula:

[Role] + [Data sources @plugins] + [Task] + [Output format] + [Constraints] + [Acceptance criteria]

Example:
You are a sales operations analyst. From @Salesforce and @Gmail, pull the last 30 days of pipeline activity.
Build a weekly pipeline review as a Google Slides deck with at-risk deals flagged.
Constraints: do not modify CRM records; do not send external email.
Notify me on @Slack when complete.

Plan Mode review checklist — confirm before execution:

• Are data sources correct (right account, right month)?
• Any high-risk actions: send external email, delete, overwrite files?
• Does output match your team's template?
• Can any steps be removed to save usage?
• Do you need a human approval checkpoint mid-run?

Templates below draw on OpenAI case studies (Zapier, Nvidia, Virgin Atlantic) and the Workspace Agent Cookbook. Replace @plugin names with your stack.

Sales

Scenario A — Daily meeting briefs (scheduled). Pain: reps spend 1–2 hours per day assembling customer context. Work scans tomorrow's calendar, pulls CRM notes, searches news, and archives briefs.

Create a scheduled task running every weekday at 4pm:

1. Check tomorrow's customer meetings in @Google Calendar (exclude internal-only)
2. For each customer meeting:
   - Pull 30-day account notes and activity from @SharePoint / @Salesforce
   - Search 30-day public news and executive moves for that company
   - Write a 2–3 sentence background per external attendee
3. Generate a 2–3 page brief per meeting, save to @Google Drive
4. Email me a summary via @Gmail with links to each brief

Output: email subject "Tomorrow's Customer Meeting Briefs — [date]"
Body: table (Customer | Time | Key topics | Brief link)

OpenAI reports sales teams turning a single discovery call into a customized PoC proposal within 24 hours — a process that traditionally took weeks.

Scenario B — Account command center (Sites + daily refresh). Pain: account intel scattered across CRM, email, and Slack.

From @Salesforce data for [Account Name]:

1. Build an interactive account command center (Sites) with:
   - Pipeline overview (stage, amount, expected close)
   - Key signals from the last 7 days (email, meetings, support tickets)
   - Prioritized next actions
2. Schedule daily refresh at 8am weekdays
3. Slack me on major changes via @Slack

Constraints: do not auto-send external email; amounts must match CRM source data.

Scenario C — Lead review and pipeline repair. Pain: thousands of monthly leads with invisible follow-up gaps.

Analyze @Salesforce leads from the last 30 days cross-referenced with @Gmail outreach.

Find:
1. Leads with no follow-up in 48+ hours (grouped by source)
2. Handoff breakpoints where response rate drops
3. Estimated pipeline loss in dollars

Output:
- Excel detail (Lead ID | Source | Last touch | Break type | Suggested action)
- 1-page executive PPT highlighting seven-figure opportunity risk
- A repeatable weekly review workflow for Scheduled Tasks

Marketing

Scenario A — Research → brief → multi-market assets. Pain: research, brief, and regional assets are done by different people; context gets lost between handoffs.

I uploaded customer research: [attachment / @Google Drive link]

Phase 1 — Brief:
- Extract audience, pain points, competitive positioning
- Output Campaign Brief (Google Docs) with messaging pillars and channel plan

Phase 2 — Assets:
- From the brief: 1 acquisition email, 3 LinkedIn posts, 1 landing page outline
- Save to @Google Drive "Campaign / [product name]"

Phase 3 — Regional adaptation:
- Adapt core assets for US, EU, and APAC (language, cultural references, compliance wording)
- Flag sensitive phrases requiring human review per region

Pause after each phase for my approval before continuing.

Scenario B — Slack / Teams sync to meeting agenda (weekly scheduled).

Schedule every Monday at 7am:

1. Summarize last 7 days from @Slack #product-launch and @Microsoft Teams "Go-to-Market"
2. Extract: decisions made, open questions, blockers needing alignment
3. Update the "Weekly Agenda" Google Doc in @Google Drive (preserve version history)
4. Post ≤5 bullet summary to @Slack #leadership

Constraints: quote only public discussions; do not leak confidential messages.

Finance

Scenario A — Month-end variance analysis. OpenAI's internal teams compressed month-end close from days to hours using this pattern.

Complete [Month] budget variance analysis:

1. Pull actuals and forecast from @Google Drive "Finance / Actuals" and "Finance / Forecast"
2. Build reconciliation workbook in @Google Sheets:
   - Department-level actual vs forecast variance
   - Flag line items with >5% or >$50K variance
   - Preserve original formulas; do not overwrite source files
3. Draft narrative explanations (Google Docs) by revenue / COGS / OpEx
4. Build 5–8 slide management deck with charts matching attached template style
5. List 3 judgment calls requiring human sign-off

Constraints: do not modify source data; cite source cells for every number.

Scenario B — Invoice and payment reconciliation.

You are an AP specialist. Compare:
- Payment register: [@Google Drive link]
- Invoice list: [@Google Drive link]

Flag in a table (Issue type | Vendor | Invoice # | Amount | Suggested action):
- Amount difference >2%
- Missing tax ID
- Duplicate invoice number
- Vendor name mismatch

Do not initiate payments; output review table only.

Operations

Scenario A — Daily dashboard morning briefing (scheduled).

Every weekday at 6:30am:

1. Visit [internal dashboard URL / @SharePoint report page]
2. Compare to yesterday's snapshot; flag >10% swings or new red indicators
3. Generate 1-page morning brief (Google Docs):
   - Top 3 items needing attention today
   - Metric change table
   - Suggested owners for follow-up
4. Email ops-leads@company.com via @Gmail

If the dashboard is unreachable, stop in Plan Mode and notify me — do not fabricate data.

Scenario B — Customer feedback clustering → product priorities.

Monitor 14-day feedback from:
- @Slack #customer-feedback
- @Gmail label "NPS-Detractor"
- @Google Drive "Support Tickets Export"

1. Cluster into 5–8 themes with representative quotes
2. Rank by frequency × impact × implementation effort
3. Output prioritized product review doc (Notion / Google Docs)
4. Schedule weekly Friday refresh as a Scheduled Task

Constraints: anonymize all customer references.

Product

Scenario A — Launch readiness review (Jira + GTM cross-check). Adapted from Nvidia's cross-system launch workflow.

Launch readiness for [Product / Feature]:

1. From @Jira: Epic / Story completion status and open blockers
2. From @Google Drive "GTM Plans": milestone check against launch plan
3. From @Slack #product-launch: unresolved discussions from last 7 days
4. Output readiness report (Google Docs):
   - Red / Yellow / Green score
   - Blocker list (Owner | Due date | Risk level)
   - Go / No-Go recommendation with evidence

Do not auto-update Jira; flag high-risk items for human decision.

Engineering — Work + Codex in the Same App

Use Codex for code; switch to Work for cross-team documents. Both live in one desktop app — no tool hopping.

Scenario A — PR review → release notes → team announcement.

In Codex mode:
1. Review PR #123 in [repo/name], focus on [security / performance / test coverage]
2. Leave side-panel review comments
3. If approved, draft release notes

Switch to Work mode:
4. Format release notes for @Confluence
5. Draft @Slack #engineering announcement (do not auto-send)

Scenario B — Multi-repo weekly engineering summary.

In Codex mode, across [frontend-repo] and [backend-repo]:
1. Summarize merged PRs this week and open P0/P1 issues
2. Generate engineering weekly report in Markdown

Switch to Work mode:
3. Convert to Google Docs; insert burndown chart from @Jira
4. Schedule every Friday at 5pm as a Scheduled Task

Four high-frequency patterns from OpenAI's recommended automations:

Setup prompt pattern:

Set up a Scheduled Task:
- Frequency: [daily / every Monday / 1st of month / when @Slack keyword appears]
- Time: [timezone + exact time]
- Action: [workflow description]
- Notification: [Slack channel / email / none]
- Human approval: [which steps need my sign-off first]

Safety checklist before unattended runs:

• Limit plugin scope to necessary tools only
• Disable auto-external-send unless explicitly required
• Set output archive paths to avoid overwriting shared files
• Enterprise: confirm agent network policy with admin
• Run manually 2–3 times before switching to scheduled

ChatGPT Work shares a metered usage pool with Codex. The same workflow can cost 5× more depending on how you write prompts and structure steps.

Seven cost-saving tactics:

1. Draft in Chat first; hand a tight brief to Work
2. Trim Plan Mode steps, especially duplicate pulls from the same source
3. Reuse template documents in Scheduled Tasks to benefit from cache discounts
4. Request concise outputs: "table + 3 bullets" beats a narrative report
5. Split large projects into phases to avoid expensive full re-runs
6. Free users: test small desktop tasks before scaling automation
7. Enterprise: set workspace / group / individual limits in Admin Console

Pre-launch usage test:

1. Pick a real task you know the human time cost of (e.g. month-end variance, usually ~2 hours)
2. Run once in Work with Plan Mode; note step count
3. Check consumption against your plan's included usage
4. Extrapolate daily / weekly / monthly cost
5. Apply tactics above and re-run to compare

Hard numbers for internal reviews: over 5 million weekly Codex users (1M+ already doing non-coding work), 1,400+ Work integrations at launch, and output-token pricing roughly 6× input — design outputs accordingly.

1. 01
   Install or update the desktop app: Download from chatgpt.com/download. Existing Codex installs migrate in place.
2. 02
   Switch to Work mode and connect plugins: Authorize the three tools your first workflow needs — typically email, chat, and file storage.
3. 03
   Paste a role template from Section 02: Start with a task you can verify — invoice reconciliation or a sales meeting brief.
4. 04
   Review Plan Mode line by line: Remove redundant steps, confirm data sources, block any unintended send or delete actions.
5. 05
   Execute, measure usage, iterate the prompt: Run three times; trim output format and duplicate pulls between runs.
6. 06
   Promote to Scheduled Task after manual validation: Use the setup pattern in Section 03; confirm triggers on three consecutive cycles before expanding scope.

ChatGPT Work pays off when Scheduled Tasks and Computer Use run reliably — not when a laptop sleeps through the 6:30am dashboard brief or a VPN drop kills a multi-hour agent session. Shared laptops introduce three production risks: sleep and wake interruptions, bandwidth jitter on hotel or home networks, and resource contention when compile jobs and agent sessions compete for the same machine.

Oversubscribed cloud VMs add a fourth: noisy neighbors and long-connection drops that break unattended Work runs mid-plan. For teams running daily Scheduled Tasks, multi-repo Codex reviews, and Computer Use file workflows around the clock, NUKCLOUD multi-region bare-metal Apple Silicon nodes provide dedicated compute, auditable tenant boundaries, and always-on macOS sessions purpose-built for agent workloads. Compare specs on the pricing page or provision a trial node via order — then point your desktop Work client at a machine that stays awake while your automations run.

• Which workflow should I try first?
  Start with the task you know best and can quality-check — month-end variance analysis, a campaign brief, or sales meeting prep. OpenAI recommends these because you can verify output quickly.
• How long should my prompt be?
  Aim for 150–400 words focused on data sources, output format, and constraints. Do not micromanage steps — Work mode plans its own execution path.
• Do Scheduled Tasks run when my laptop is off?
  Desktop Scheduled Tasks require the device online and logged in. For true background automation, use web Workspace Agents on Plus or higher plans.
• What is the difference between Work mode and Workspace Agent?
  Work is personal agent mode inside ChatGPT. Workspace Agents are team-built, admin-governed automations in Business and Enterprise with Admin Console controls. Same technical foundation, different entry point and governance.
• Can I use generated slides or reports externally as-is?
  Treat them as 80% drafts. Always human-review financial numbers, customer names, and external statements before publishing.
• What can Free users run from this guide?
  Desktop Work mode is available with usage limits. Start with lightweight tasks like invoice reconciliation before scheduling long-running automation.

Last updated: 2026-07-11 | Sources: OpenAI Blog, OpenAI Cookbook — Sales Meeting Prep, ChatGPT Learn Changelog