ChatGPT for Sales
ChatGPT Work is for delegating a substantial assignment with a clear outcome. It can use approved files, plugins and tools to retrieve information, complete several steps and create a reviewable deliverable such as a brief, analysis, spreadsheet, document, presentation, recurring update or workflow.
Use Chat when you want an answer or short draft. Use Work when you want the work product.
Product details verified 30 August 2026. Tool access, cloud or local operation, site access, approvals, credits and plan availability vary by surface and workspace. Review Get started with ChatGPT Work and ChatGPT Work and Codex before publication.
| Brief field | What to specify | Sales example |
|---|---|---|
| Outcome | The business result and artifact | “Create a Tier 1 account plan and 10-slide executive briefing” |
| Audience | Who will read or use it | “AE and regional sales director” |
| Sources | Files, sites, plugins or apps it may use | “Approved account files, annual report, company newsroom” |
| Facts | What is already verified | “Current opportunity stage and named stakeholders” |
| Questions | What it must investigate or resolve | “Which initiatives create a credible entry point?” |
| Steps | Required sequence or method | “Research, evidence table, stakeholder map, plan, deck” |
| Constraints | Claims, tone, confidentiality, format and length | “Do not infer private priorities; no unapproved customer logos” |
| Deliverables | Exact files or outputs | “Brief .docx, stakeholder .xlsx, 10-slide .pptx” |
| Approvals | Actions that need confirmation | “Ask before signing into sites, sending or overwriting files” |
| Definition of done | Objective review criteria | “Every claim sourced; unknowns labeled; deck matches template” |
Outcome: [what must be complete] Audience and decision: [who will use it, for what] Use these sources: [files, sites, plugins/apps] Known facts: [verified context] Questions to resolve: [research or analysis questions] Required steps: [method or sequence] Constraints: [privacy, claims, tone, format, commercial boundaries] Deliverables: [files, sections, formats] Approval gates: [actions requiring confirmation] Definition of done: [evidence, completeness and quality checks]
Inputs: target account, offer, ICP, approved proof, public-source boundaries and named personas.
Work: create the evidence table, account point of view, stakeholder hypotheses, outreach angles and a reviewed message package.
Deliverables: account brief, stakeholder matrix and channel-specific drafts.
Review: verify every trigger and executive claim; send nothing automatically.
Inputs: attendee or account list, event agenda, seller capacity, ICP rules and offer.
Work: prioritize targets, research material accounts, map relevant sessions and create meeting/outreach plans.
Deliverables: ranked workbook, daily brief and approved message drafts.
Review: confirm attendee data permissions and do not imply attendance or familiarity that is not verified.
Inputs: account evidence, relationship notes, product context, current opportunity/customer state and planning template.
Work: synthesize research, initiatives, stakeholders, whitespace, risks and a sequenced action plan.
Deliverables: account-plan document and executive-review deck.
Related resource: Account Plan Chain: Research → Stakeholder Map → Quarterly Actions.
Inputs: dated CRM export, transcripts, notes, MEDDPICC standard and current forecast rules.
Work: extract evidence, identify gaps and contradictions, test the path to close and prepare questions for the review.
Deliverables: deal brief, evidence matrix, risk register and 14-day action plan.
Related resource: Deal Intelligence Chain: Transcript → MEDDPICC → Next Action.
Inputs: named competitor, target segment, current positioning, approved proof and source preferences.
Work: investigate current claims, pricing/packaging where public, product differences, customer evidence and likely traps.
Deliverables: sourced research appendix, one-page battlecard and manager review list.
Review: label observations that cannot be independently verified; do not invent competitor weaknesses.
Inputs: account list, segmentation criteria, historical performance, capacity and territory objectives.
Work: clean and analyze data, test segmentation, identify concentration or coverage gaps and create a 90-day plan.
Deliverables: workbook, written plan and review presentation.
Review: validate calculations and preserve the source-data timestamp.
Inputs: discovery notes, requirements, approved answer library, proof, commercial constraints and brand template.
Work: build a compliance matrix, draft allowed sections, identify unresolved answers and assemble the reviewable file.
Deliverables: response matrix, proposal/RFP draft and open-question register.
Review: subject-matter, security, legal and commercial owners approve their sections before use.
Deep Research is optimized for investigating a question and producing a documented report with source links. Work is optimized for completing an assignment and creating a usable output.
Use Deep Research first when evidence is uncertain or the source plan needs review. Use Work first when sources are already supplied and the main challenge is analysis, production and file creation. Combine them when the assignment needs both a defensible evidence base and a finished artifact.
A Project organizes an evolving body of context. Work completes a specified assignment. A Project may hold the account’s material; a Work task may turn that material into this quarter’s account plan. Current Projects product behavior matters: Work is unavailable inside a Project using project-only memory at the verification date.
Work answers “What complete outcome should ChatGPT produce?” A Scheduled Task answers “When or after what trigger should the instruction run?” A substantial recurring briefing may combine both, but design the output, review point and notification rule separately.
Before accepting the work:
Choose a job whose outcome is obvious and reviewable: a meeting brief, deal-review package, account plan or competitor battlecard. Supply the source material and definition of done, then inspect the files before using them.
Start with account-research prompts or build the evidence base with Deep Research.
ChatGPT Work is the mode for delegating real work with a clear outcome. It can use approved files, plugins and tools, complete several steps and create reviewable outputs.
No. Use the current product name “ChatGPT Work” for the delegated-work experience. Historical agent-mode terminology should not be used as the page’s primary label.
OpenAI’s current documentation says Work can create finished files and reviewable outputs, including documents, spreadsheets and presentations where the relevant surface supports them.
Only when the connection, permission, action and explicit approval are defined. The safer default is to produce a reviewable draft or proposed update.
Each capability owns one seller job. Start with the smallest one that produces the outcome.
