6 role-based prompts · Plan Mode review · Scheduled Tasks recipes · usage optimization
If you already know what ChatGPT Work is, the real question is: what do you do with it on Monday morning? This guide covers sales, marketing, finance, ops, product, and engineering with copy-paste prompt templates, Plan Mode checklists, Scheduled Tasks recipes, and usage optimization — following OpenAI's advice to start with a task you already know well. For the launch recap and Claude Cowork comparison, see our companion post. Last updated: 2026-07-11
Understand how ChatGPT Work differs from regular Chat before copying templates. These three principles decide whether your first task delivers a usable deliverable:
| Principle | What it means | Practical tip |
|---|---|---|
| Describe outcomes, not steps | Work mode plans its own path | ❌ "Open Salesforce, export…" → ✅ "Build a weekly pipeline PPT from @Salesforce deals in the last 30 days, flagging at-risk opportunities" |
| Connect tools first | Plugins are Work's data layer | Authorize Gmail, Slack, Drive before starting; use @AppName to pin sources |
| Plan Mode is your brake | Review the plan before execution | For high-stakes deliverables (external emails, financial reports, client docs), approve every step |
The new desktop app has three modes — using the wrong one wastes usage:
| Your need | Use | Why |
|---|---|---|
| Quick Q&A, brainstorming, single-turn copy | Chat | Lightweight, fast |
| Multi-app projects, finished deliverables, hours-long tasks | Work | Plugins + Plan Mode + Computer Use |
| Code review, PRs, multi-repo development | Codex | Developer-native workflows |
| Recurring background automation | Work + Scheduled Tasks | Triggered or scheduled execution |
| Scenario | Recommended environment |
|---|---|
| Local files, Computer Use, free-tier trial | Desktop (Mac / Windows) |
| Team collaboration, task progress monitoring | Web / mobile (Plus+) |
| Sales meeting briefs + email notifications | Web Workspace Agent + scheduling |
| Local Excel reconciliation, folder batch processing | Desktop Work mode |
Five pain points beginners hit most often:
Using Chat like Work: Single-turn Q&A cannot pull cross-app data yet you expect a full PPT — wrong mode, wasted usage.
Task before plugin auth: Work's data layer is the plugin directory; without Gmail or Salesforce connected, the agent fabricates or stalls.
Skipping Plan Mode review: External emails, financial numbers, client deliverables — auto-send mistakes are expensive to fix.
Mixing desktop/Web capabilities: Computer Use and local batch jobs are desktop-only; expecting them on Web fails.
Scheduled tasks on sleeping devices: Desktop Scheduled Tasks need the machine awake and logged in; true unattended runs need Web Workspace Agents.
Whatever your role, follow this flow:
1. Connect plugins → 2. Write goal + output format → 3. Review Plan Mode → 4. Steer mid-flight → 5. Accept deliverable & iterate
[Role] + [Data sources @plugins] + [Task] + [Output format] + [Constraints] + [Acceptance criteria] Example skeleton: You are a [role]. Pull [data type] from @Salesforce and @Gmail for [time range]. Complete [action], output as [Google Docs / Excel / PPT / Sites]. Constraints: [do not modify source data / two decimal places / no external email]. When done: [Slack notify me / save to folder].
Confirm before execution:
Download the desktop app: Go to chatgpt.com/download; update Codex App if already installed.
Switch to Work mode: Select Work in the top nav; use Codex for engineering tasks.
Connect the plugin directory: Authorize Gmail, Slack, Drive, Salesforce in Settings.
Write your prompt with the formula: Outcome + @AppName + format + constraints; enable Plan Mode.
Review plan, then execute: Trim redundant steps, confirm no high-risk actions, then start.
Accept and iterate: Check deliverable quality, note usage consumed, then convert to Scheduled Task if satisfied.
Templates below draw on OpenAI examples, early tester feedback (Zapier, Nvidia, Virgin Atlantic), and the Workspace Agent Cookbook. Replace @plugin names with your stack.
Scenario A: Daily meeting briefs (scheduled) — Pain: 1–2 hours/day prepping client context. OpenAI internal case: Discovery call to custom PoC in 24 hours (weeks traditionally).
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 interactions from @SharePoint / @Salesforce - Search 30-day public news and executive updates for the company - Write 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 brief links Email subject: "Tomorrow's Customer Meeting Briefs — [date]" Body: table (Account | Time | Key topics | Brief link)
Scenario B: Live account command center (Sites + daily refresh)
From all @Salesforce opportunities, contacts, and recent activity for [Account Name]: 1. Create an interactive account command center (Sites) with: - Pipeline overview (stage, amount, expected close) - 7-day key signals (email, meetings, support tickets) - Prioritized next actions 2. Schedule daily refresh weekdays at 8am 3. Slack me on major changes Constraints: no auto external email; amounts from CRM source data.
Scenario C: Lead review & pipeline repair (Zapier-style)
Analyze @Salesforce leads from the last 30 days cross-referenced with @Gmail outreach. Find: 1. Leads with 48h+ no follow-up (grouped by source) 2. Broken handoff points (where response rate drops) 3. Estimated pipeline loss Output: - Excel detail (Lead ID | Source | Last touch | Break type | Suggested action) - 1-page executive PPT highlighting seven-figure opportunity risk - Repeatable weekly review workflow for Scheduled Task
Scenario A: Research → Brief → Multi-market assets (end-to-end)
Using uploaded research / @Google Drive materials: Phase 1 — Brief: audience, pain points, positioning → Campaign Brief (Docs) with messaging pillars Phase 2 — Assets: 1 email, 3 LinkedIn posts, landing page outline → save to Drive Campaign folder Phase 3 — Localization: adapt for US, EU, APAC with sensitive-phrase flags Pause after each phase for my approval.
Scenario B: Slack/Teams → meeting agenda sync (weekly scheduled)
Every Monday 7am: summarize last 7 days from @Slack #product-launch and @Teams GTM channel. Update the "Weekly Agenda" Google Doc. Post ≤5 bullet summary to @Slack #leadership. Only cite public discussions; never leak confidential messages.
Scenario A: Month-end variance analysis (OpenAI-validated) — Close and forecast from days to hours:
Complete [Month] budget variance analysis: 1. Pull actuals and forecast from @Google Drive Finance folders 2. Build reconciliation workbook in @Google Sheets (flag >5% or >$50K variances) 3. Draft narrative explanations (Docs) by revenue / cost / opex 4. Build 5–8 slide management deck 5. List 3 judgment calls requiring human sign-off Do not modify source files. Cite all numbers.
Scenario B: Invoice vs. payment register reconciliation
Compare payment register and invoice list from @Google Drive. Flag: >2% amount differences, missing tax IDs, duplicate invoice numbers, vendor name mismatches. Return review table only. Do not initiate payments.
Scenario A: Daily dashboard morning briefing (scheduled)
Every weekday 6:30am: visit [dashboard URL], compare to yesterday's snapshot. Flag >10% swings. Generate 1-page brief. Email ops-leads@company.com. If dashboard is unreachable, stop and notify — do not fabricate data.
Scenario B: Customer feedback clustering → product priorities
Monitor 14-day feedback from @Slack #customer-feedback, @Gmail NPS-Detractor, Drive ticket exports. Cluster into 5–8 themes, rank by frequency × impact × effort. Output prioritized product review doc. Schedule weekly Friday refresh. Anonymize all customer references.
Scenario A: Launch readiness review (Jira + GTM cross-check, Nvidia-style)
Launch readiness for [Feature]: 1. @Jira: completion status and open blockers 2. @Google Drive GTM plan: milestone check 3. @Slack #product-launch: unresolved discussions Output: Red/Yellow/Green readiness report with Go/No-Go recommendation. Do not auto-update Jira.
Use Codex for code, Work for cross-team docs — switch modes in one desktop app.
Scenario A: PR review → release notes → team announcement
Codex mode: Review PR #123 in [repo], side-panel comments, draft release notes. Work mode: Format for @Confluence, draft @Slack #engineering post (do not auto-send).
Scenario B: Multi-repo weekly engineering summary
Codex mode: Cross [frontend] + [backend] repos — merged PRs, open P0/P1 issues → Markdown weekly report. Work mode: Convert to Google Docs, pull burndown from @Jira. Schedule Fridays 5pm.
| Recipe | Trigger | Action | Best for |
|---|---|---|---|
| Monday agenda refresh | Mon 7am | Slack digest → update agenda doc | Marketing / Ops |
| Daily metrics brief | Weekdays 6:30am | Dashboard diff → email report | Ops / Finance |
| Feedback clustering | Fri 4pm | Multi-channel → priority list | Product |
| Account daily refresh | Weekdays 8am | CRM changes → update Sites dashboard | Sales |
Set up Scheduled Task: - Frequency: [daily / every Monday / 1st of month / when @Slack keyword appears] - Time: [timezone + exact time] - Action: [workflow description] - Notify: [Slack channel / email / none] - Human approval: [which steps need my sign-off first]
ChatGPT Work shares a metered usage pool with Codex. The same workflow can cost 5× more depending on design.
| Factor | Impact on usage |
|---|---|
| Task step count | More steps = more consumption |
| Context size | More docs/emails pulled = higher cost |
| Output length | Output tokens cost ~6× input |
| Cache hits | Repeated reads: cached input ~1/10 of fresh |
| Model choice | GPT-5.6 heavy reasoning costs more than needed for light tasks |
Seven cost-saving tactics:
Draft in Chat first, then hand a tight brief to Work.
Trim Plan Mode steps, especially duplicate data pulls.
Reuse template docs in Scheduled Tasks for cache discounts.
Request concise outputs (table + 3 bullets > narrative report).
Split large projects into phases to avoid expensive re-runs.
Free users: test small desktop tasks before scaling.
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 2. Run once in Work with Plan Mode, note steps 3. Check consumption against your plan's included usage 4. Extrapolate daily/weekly/monthly cost 5. Optimize and re-run to compare
| Issue | Cause | Fix |
|---|---|---|
| Codex projects missing | Incomplete app migration | Update Codex app → becomes ChatGPT desktop; if broken, clean reinstall from chatgpt.com/download |
| Plugin connected but no data | Insufficient scope or wrong @name | Re-check plugin permissions; use explicit @Salesforce not "the CRM" |
| Good plan, wrong output | Stale context or AI inference | Pause and steer; attach explicit source files |
| Scheduled task didn't fire | Device asleep / logged out | Use web Workspace Agents for true background; desktop tasks need device online |
| Usage higher than expected | Verbose output, redundant pulls | See optimization section; Enterprise: Admin Console limits |
| Work vs Cowork confusion | Different workflow types | Cloud SaaS → Work; local folder batch → Cowork (companion comparison) |
| Week | Goal | Action |
|---|---|---|
| 1 | Single-task fluency | Run 3 manual Work tasks you can quality-check; practice Plan Mode review |
| 2 | Plugin depth | Connect 3 core tools; complete 1 cross-app deliverable |
| 3 | Automation | Convert Week 1 task to Scheduled Task; verify 3 triggers |
| 4 | Team rollout | Document role-specific prompt library; set admin limits (Enterprise) |
ChatGPT Work isn't valuable because it exists — it's valuable when it removes a workflow you already resent doing manually. Fastest ROI: pick one task you know intimately, run it three times, tune the prompt, then automate.
ChatGPT Work excels at Slack, Gmail, and Drive orchestration, but engineering teams needing Xcode signing, Metal local inference, and 7×24 iOS CI/CD cannot rely on chat agents alone — VM macOS has performance and EULA risks; personal laptops are poor 24/7 hosts. For stable production environments suited to iOS CI/CD and AI Agent automation, VpsMesh Mac Mini cloud rental is usually the better fit: bare-metal Apple Silicon, root access, predictable monthly cost — Work orchestrates context while cloud Mac runs builds and signing.
The task you know best and can verify — month-end variance, campaign brief, or sales meeting prep. OpenAI recommends these because you can quickly judge quality.
150–400 words focusing on data sources, output format, and constraints. Do not micromanage steps — that is what Work mode automates.
Desktop tasks need the device online. For true background automation, use web Workspace Agents (Plus+). Pair with Mac Mini M4 cloud rental for 7×24 CI builds.
Work is personal agent mode inside ChatGPT. Workspace Agents are team-built, admin-governed automations in Business/Enterprise. Same technical base, different entry points.
Treat them as 80% drafts. Always human-review numbers, names, and external statements.
Desktop Work with limits. Start with lightweight tasks like invoice reconciliation before scheduling automation. See our help center for deployment details.