ChatGPT for Sales
A ChatGPT Project is a persistent workspace for one evolving body of sales work. It keeps related chats, files and project instructions together so the seller can return to the same account, opportunity, territory or motion without reconstructing the context in every conversation.
Projects are most useful when the subject stays the same while the questions change. A strategic account may need research, stakeholder mapping, meeting preparation, transcript analysis and deal strategy over several months. The Project holds that body of work; the CRM still owns the official customer and opportunity record.
Product details verified 30 August 2026. OpenAI currently makes Projects available across free and paid subscription types. File limits, memory behavior, sharing and tool access vary by plan and workspace. See Projects in ChatGPT.
Use four context layers. Fewer, better files are usually more useful than a document dump.
| Layer | Include | Do not treat as authoritative |
|---|---|---|
| Project instructions | Seller role, objective, approved methodology, output conventions, evidence rules, review boundaries | Live deal facts that change every week |
| Reference files | Approved product material, customer proof, account documents, pricing guidance, meeting notes, transcripts | Unapproved personal data, obsolete collateral or uncontrolled exports |
| Project chats | Research, call plans, analysis, drafts and decisions tied to the same body of work | The official CRM record |
| Current-state brief | A dated summary of stage, stakeholders, evidence, gaps, next step and owner | A permanent truth after the deal changes |
Use this as a setup template, then adapt it to your organization’s policies:
You are supporting a B2B seller working on [account / opportunity / territory]. Use only the files, chats and facts provided in this Project or sources explicitly approved for the current task. Separate verified fact, buyer statement, seller inference and unknown. Do not invent contacts, priorities, metrics, commitments or MEDDPICC evidence. When information conflicts, name the conflict and ask which source is current. End substantive outputs with: evidence used, open questions, recommended next action, owner and review required. Do not write to external systems or create customer commitments without explicit approval.
This is a small setup asset created for this page, not a substitute for a full Promptifi Project product record.
OpenAI currently documents unlimited Project creation, with per-Project file caps of 5 for Free, 25 for Go and Plus, and 40 for Edu, Pro, Business and Enterprise. A maximum of 10 files can be uploaded at one time. These limits are operational details, not reasons to fill a Project to capacity; use only current files the sales job requires.
Use it for: a named account that warrants research, executive mapping, multithreading and a long-term point of view.
Recommended files: account brief, annual report or approved research, org information, relevant customer proof, product overview, prior correspondence and meeting notes.
Maintain: current initiatives, stakeholder hypotheses, verified triggers, open questions, relationship map and the next account-development action.
Run inside it: Account Research to Personalized Outreach in 3 Steps and the Account Research & Intelligence Briefings use case.
Use it for: one complex opportunity from discovery through decision.
Recommended files: discovery notes, call transcripts, mutual action plan, approved commercial material, technical requirements and the latest CRM summary.
Maintain: evidence by MEDDPICC element, stakeholders, risks, decision process, compelling event, agreed next step and source dates.
Run inside it: Deal Intelligence Chain: Transcript → MEDDPICC → Next Action, Discovery Call to MEDDIC-Scored Deal Summary and Diagnose MEDDPICC gaps.
Use it for: a segment, geography or book of accounts that needs prioritization and a 30-, 60- or 90-day plan.
Recommended files: account list, segmentation rules, closed-won/lost data, coverage model, relevant market evidence and activity constraints.
Maintain: tiering logic, target-account movements, assumptions, pipeline coverage and weekly focus.
Use it for: a recurring discovery motion, executive meeting series, event or buying-committee program.
Recommended files: agenda templates, approved questions, persona context, research, transcripts and follow-up standards.
Maintain: hypotheses tested, buyer language, unanswered questions, commitments and next-meeting objective.
Use it for: thoughtful multichannel work across several stakeholders at one account or a tightly defined ABM cohort.
Recommended files: account research, trigger evidence, persona angles, approved proof and message constraints.
Maintain: who received what, response evidence, message learning and the next channel decision. Do not use the Project as a substitute for sequence or CRM activity history.
Run inside it: Cold Outreach Chain: Research → Email → Follow-Up.
Use it for: a document-intensive evaluation where requirements, answers, evidence and reviewers must stay aligned.
Recommended files: requirements, discovery notes, approved response library, proof points, security material, commercial boundaries and current draft.
Maintain: requirement owner, answer status, evidence source, unresolved question, risk and approval state. Never let an AI-generated answer bypass subject-matter review.
Use it for: a manager or seller who reviews a stable set of opportunities on a regular cadence.
Recommended files: dated CRM export, stage definitions, forecast rules, deal notes and prior review actions.
Maintain: data timestamp, stale opportunities, evidence gaps, next-step quality and actions from the prior review. Replace exports rather than allowing multiple undated versions to compete.
Related resource: Pipeline Review & Deal Risk Assessment.
Use it for: one strategic customer with adoption, value, renewal-risk and growth work over time.
Recommended files: goals, success plan, QBR material, approved usage summaries, stakeholder notes and renewal timeline.
Maintain: realized value, evidence, risk, relationship health, dates and next customer action.
Projects can use default memory or project-only memory. OpenAI currently allows the setting to be changed later for eligible personal projects; shared projects automatically use project-only memory and cannot switch to default memory. Project-only memory keeps chats from referencing conversations outside the Project and prevents outside chats from referencing Project conversations. It also means ChatGPT Work is not available in that Project at the time of verification.
Choose project-only memory when isolation matters more than convenience. Choose default behavior only if your account, workspace and data policy allow context beyond the Project. Review the current Projects memory documentation before publication because this behavior is highly changeable.
| Question | Project | Custom GPT | Skill | Work |
|---|---|---|---|---|
| What stays stable? | The subject: one account, deal or territory | The assistant’s configuration | The procedure | The requested outcome and brief |
| What usually changes? | The questions and evolving evidence | The user and case input | The account or task input | The steps required to finish the assignment |
| What is the core value? | Persistent body of work | Reusable specialist | Reusable method | Finished multi-step deliverable |
| Best sales example | Strategic-account workspace | Team-approved discovery coach | MEDDPICC evidence-extraction method | Completed account plan and executive deck |
Start with one strategic account or live deal. Add only the files required for the next decision, set the evidence rules, then run a canonical Promptifi prompt or chain inside the Project.
Browse account and deal prompts or see the MEDDPICC use case.
OpenAI’s current Projects documentation says Projects are available across free and paid subscription types. File limits and other capabilities vary by plan and workspace.
No. A Project is a working context environment. The CRM remains the authoritative record for contacts, opportunity fields, stages, activity and forecast commitments.
No. Create one when the account warrants repeated work and persistent context. A prompt or a lightweight research brief is more appropriate for low-priority accounts.
OpenAI currently sets shared Projects to project-only memory. They do not access an individual member’s external context, Custom Instructions or memories outside the Project.
Each capability owns one seller job. Start with the smallest one that produces the outcome.
