ChatGPT for learning ops (Oct 2026): course outlines, quiz drafts, and training-request triage

Who this guide is for

Learning ops people — the L&D coordinator who turns policies into training, the HR generalist who writes the annual compliance quiz, the enablement manager fielding “can we get a course on…?” requests — whose company already has ChatGPT Business (formerly ChatGPT Team; OpenAI renamed it on August 29, 2025) or Enterprise. If your LMS already drafts courses and quizzes, see LearnUpon AI, TalentLMS AI, or Easygenerator AI; if you buy ready-made courses, see Go1 AI. For policy questions from employees, see ChatGPT for HR. Product deep-dive: ChatGPT.

Use Projects, not new custom GPTs. OpenAI’s help center (read 11 October 2026) says custom GPTs are scheduled to retire on December 11, 2026 (February 11, 2027 for Enterprise workspaces with an approved deferral), that the change affects all ChatGPT plans, and that it plans to help creators migrate GPTs to plugins. For affected Enterprise workspaces it lists October 26, 2026 (planned) as the end of new custom GPT creation. Building a new GPT now means migrating it within weeks, so this guide keeps instructions and source documents in a project, which OpenAI says keeps related chats, files, and instructions together and costs nothing extra on qualifying plans.

Ground rules: ChatGPT never invents policy. Quizzes and outlines come only from an approved document you upload, and anything it can’t source is marked, not filled in. The policy owner approves every quiz and outline before it reaches the LMS, and L&D decides which requests become training. Never paste employee data — names, completion records, quiz results, performance notes, or a manager’s description of a named person. A passed quiz shows someone read the policy, not that they can apply it.

Major features (with how someone uses them)

Examples at a glance: each feature has a concrete example from a real job plus a short Try this you can run this week.

1. Project Training drafts: a quiz from one policy (primary)

What it is: Select New project in the sidebar and name it Training drafts. Upload one approved, current policy — the version the policy owner signed off, with section numbers. In Project settings, add instructions: write questions only from the uploaded document; put the section number after each question and answer; if a question can’t be answered from the document, mark it “Not in the policy — owner to decide” and don’t write an answer; never add a rule, number, deadline, or example that isn’t in the document. OpenAI says project instructions apply only within that project and override your global custom instructions, and Business projects support up to 40 files.

The marked questions are the useful part. If ChatGPT wants to ask “How many days do you have to report an incident?” and the policy doesn’t say, that isn’t a quiz problem — it’s a gap the policy owner should know about before anyone is tested on it.

Concrete example (L&D coordinator at a 300-person distributor) An L&D coordinator uploads the approved travel and expense policy and asks for a 5-question quiz. Four questions cite sections 2.1, 3.4, 4.2, and 5.1; the fifth, about whether tips count toward the meal limit, is marked “Not in the policy.” The finance controller, who owns the policy, approves the four, decides the fifth needs a policy update first, and the quiz goes into the LMS with four questions.

Try this

Primary workflow:

Project training drafts. Upload one approved policy. Ask for a 5-question quiz and mark any question it can’t source from the document.

2. Course outlines from approved documents

What it is: Upload the approved documents a course should teach — a procedure, a product sheet, a policy — and ask for an outline: audience, three to five learning objectives, modules with the source section for each, and a list of anything a learner would need that the documents don’t cover. Ask for objectives written as things people do (“submit an expense report with the right receipt”), not things they know. The subject expert reviews the outline before anyone builds slides, videos, or LMS modules from it.

An outline is cheap to fix and a finished course isn’t. Have the subject expert cross out any module without a source section, and send the “not covered” list to whoever owns the gap — sometimes the honest answer is that the procedure needs writing first.

Concrete example (enablement manager building onboarding) An enablement manager uploads the approved returns procedure and the warranty policy and asks for a 30-minute course outline for new support agents. The outline has four modules with source sections and flags that neither document says who approves refunds above $500. The support director answers that question in the procedure first; then the course is built.

Try this

Outline before slides:

Upload the approved documents for one course and ask for an outline with a source section for each module. Have the subject expert review it before anything is built.

3. Training-request triage with names removed

What it is: Export last month’s training requests from your form or ticket queue and remove requester names, employee names, and anything about a named person’s performance before uploading. Upload your current course catalog as a list of titles and descriptions. Ask ChatGPT to group the requests by theme, suggest an existing catalog course where one plausibly fits, and flag requests that look like a process or tooling problem rather than a training need. L&D decides what to build, assign, or decline; ChatGPT only sorts.

The “already in the catalog” column often saves the most time, and the “not a training problem” column is often the most valuable: five requests for “Excel training” that all come from one broken report template need a fixed template, not a course.

Concrete example (HR generalist who owns training requests) An HR generalist exports 42 requests, removes names and team details, and uploads them with the catalog list. ChatGPT groups them into nine themes, matches 14 to existing courses, and flags six about the new expense tool as possibly a setup issue. She checks every match against the course description, sends the six to IT, and the L&D lead picks two new courses to build next quarter.

Try this

Names out first:

Export last month’s training requests with names removed and upload your course catalog list. Ask for themes, catalog matches, and requests that may not need training, then have the L&D lead decide.

5. What never goes in: employee data

What it is: OpenAI says it doesn’t train on ChatGPT Business, Enterprise, or Edu workspace data by default, and its Business pricing page lists SAML SSO and SOC 2 Type 2. That doesn’t make a project the right place for employee data. Keep out names, completion records, quiz results, performance notes, accommodation or leave details, and a manager’s description of a named person’s “training need.” The pricing page lists SCIM, role-based access controls, the Compliance API logs platform, and data residency for Enterprise, not Business. Before a connected app searches outside the project, OpenAI says ChatGPT may ask you to confirm.

Concrete example (people ops lead writing team rules) A people ops lead writes three rules at the top of the project instructions: only approved documents and anonymized requests go in; no names, completion records, results, or performance notes; and nothing reaches the LMS until the policy owner approves it. When a manager asks for a quiz “for the two people who keep getting this wrong,” the team writes a general refresher from the policy instead.

Try this

Write the three rules:

Write three rules for your training project — what may go in, what never goes in, and who approves anything that reaches the LMS — and paste them at the top of the project instructions.

6. If you already have a training GPT

What it is: OpenAI says existing custom GPTs keep working until retirement on December 11, 2026, that migration is to plugins (My GPTs → Migrate to plugin when it reaches your account), and that the GPT’s instructions become a skill and its knowledge files become reference files. The selected model, custom actions, and sharing settings don’t carry over; migration uses the latest published version; the original becomes read-only; and the new plugin starts private. Chats created with a GPT can’t be moved into a project. Existing conversations stay accessible after retirement.

For a quiz-writing GPT, the simplest move is often a project: copy its instructions into Project settings and upload the same approved documents. Either way, test with the same requests and check that the “Not in the policy” rule survived — a replacement that fills gaps from general knowledge is worse than the GPT it replaced.

Concrete example (HR generalist with an old quiz GPT) An HR generalist has a Compliance quiz helper GPT built last year. She saves five requests it handled, copies its instructions into a Training drafts project, uploads the current policies, and reruns them. One quiz now includes a question about a reporting deadline the policy doesn’t state; she tightens the instruction and reruns before telling colleagues to switch.

Try this

Test the replacement:

Before December 11, 2026, save five requests your training GPT already handled. Move its instructions into a project, rerun them, and compare every marked question before anyone switches.

Pricing context (October 2026)

OpenAI’s Business pricing page and its ChatGPT Business help article (as listed 11 Oct 2026, USD): Standard seat $20 per user per month billed annually or $25 billed monthly; Premium seat $100 per user per month billed annually or $125 billed monthly, with 5x the usage of Standard and no 5-hour limit. A Business workspace needs at least two seats and is for teams of 2 to 200; API usage is billed separately. Enterprise is priced through sales.

OpenAI’s Projects help article says projects cost nothing extra on qualifying plans. Pricing may vary by country and currency. Full dated notes are on the ChatGPT deep-dive.

Uncertainty: OpenAI says its retirement dates are subject to change and that migration may reach accounts at different times; the October 26 creation cutoff is listed for affected Enterprise workspaces. Project limits (40 files, 100 collaborators on Business) and memory behaviour come from the help center and can change. Skills is labelled beta on the pricing page. Nothing here replaces your own policy: if a quiz question can’t be sourced from the approved document, the policy owner decides, not ChatGPT.

When it’s useful (honest take)

Useful when your company already pays for ChatGPT Business or Enterprise, you have approved, current policies and procedures with named owners, and someone owns the project’s instructions and files.

Useful when L&D wants first-draft quizzes and outlines it can check against a source, or needs to sort a pile of training requests without reading each one twice.

Less useful when your LMS already drafts courses and quizzes from the same documents, policies aren’t written down or approved, the team can’t keep employee data out, or anyone expects ChatGPT to decide what the policy says.

Related options (educational)

FAQ

How should a learning ops team set up ChatGPT?

In a Business or Enterprise workspace, create a project (for example, Training drafts), upload one approved policy, and add project instructions: write questions only from the document, cite the section, mark anything it can’t source, and never add a rule that isn’t there. Share it at chat access and have the policy owner approve every quiz.

Should we build a custom GPT for quizzes?

Not a new one. OpenAI says custom GPTs are scheduled to retire on December 11, 2026 (February 11, 2027 for Enterprise workspaces with an approved deferral) and plans migration to plugins. Use a project instead; if you already have a GPT, save test requests and compare the replacement before switching. See OpenAI’s retirement FAQ.

Can we upload completion records or quiz results?

Don’t. Keep names, completion records, quiz results, performance notes, and anything about a named person out of a general chat tool; those belong in your LMS and HR system. Training requests can go in once names and personal details are removed.

Who can see files in a shared project?

Everyone you add can view and download the uploaded files, whether they have edit or chat access, OpenAI says. Edit access also allows changing the instructions and files and inviting people; chat access only allows using the project, which keeps its own project-only memory. See Projects in ChatGPT.

What does ChatGPT Business cost in October 2026?

As listed on 11 October 2026: Standard seats $20 per user per month billed annually or $25 monthly, and Premium seats $100 annually or $125 monthly with 5x the usage, for teams of 2 to 200. Projects cost nothing extra. See openai.com/business/pricing.

Optional next step (if seats overlap)

If learning ops pays for ChatGPT, an LMS with its own AI authoring, and a separate course writer that all draft quizzes from different copies of the same policy, fix that overlap before adding another seat. Map the overlap with the $19 AI toolkit audit or the $9 AI toolkit spreadsheet. Email Audit $19 or Toolkit spreadsheet $9.

Checkout is live via Stripe. After you pay, we email delivery (spreadsheet) or intake next steps (audit).

Related: ChatGPT · ChatGPT for HR · LearnUpon AI · Go1 AI · Highspot vs Seismic · Claude for HR · Gemini for HR · Easygenerator AI · TalentLMS AI · Claude Projects · Custom GPTs · What’s new in ChatGPT (Sep 2026) · People ops stack map · Run recurring ops work in Claude Projects · Stack map for people ops · Three checks before another AI seat · ops hub