Save Your Best Instructions: Projects, GPTs and Skills · Lesson 6 of 7
Share it, keep the master copy, and give it an owner
25 minutes. One editor, everyone else a user, the original in your drive, and a date to look again.
---
A saved instruction only one person can reach is just a nicer sticky note. The payoff comes when the whole team starts from the same tested version. That means sharing it, and sharing it carefully, because a shared instruction that anyone can edit drifts as fast as the pasted text it replaced.
This lesson covers four things: who can use it and who can change it, how to share it in each tool, why the master copy is your way out of every product change, and the owner and review date that keep it current.
For each saved instruction, decide:
- Who can use it: usually everyone whose work it's for.
- Who can edit it: one person, the owner. Maybe a backup. Not the whole team.
When someone wants a change, they tell the owner. The owner changes the master copy, reruns the three tests from Lesson 5, and updates the tool. That sounds slow. In practice it takes a few minutes, and it's the only way to know what the instruction says on any given day.
Sharing usually needs a business or team plan, and an admin can turn it off. If you don't see a share option, that's the first thing to check.
Claude
- Projects (Team and Enterprise): share with specific people, or with the whole organization. Pick Can view for people who should use it (they can see the instructions and files and chat in the project) and Can edit only for the owner.
- Skills (Team and Enterprise): share with specific people, groups (where allowed), or publish to your organization's directory. Shared skills are view-only, and when you update the skill, everyone gets the new version automatically. Sharing is off until an owner turns it on in Organization settings. An owner can also add a skill for the whole organization.
ChatGPT
- Projects: Business, Enterprise and Edu accounts can share projects with teammates. In a shared project, deleting a file removes it for everyone, so keep the originals in your drive.
- Skills (business plans): share with people or groups in your workspace, or publish to the workspace library. Admins control whether members can share or publish.
- Custom GPTs you already shared: when a GPT is migrated to a plugin, its sharing does not carry over. Migrated plugins start private, and everyone who used the GPT needs a fresh invite.
Microsoft 365 Copilot
- Agents: a new agent is private until you select Share. Give the owner Can edit and everyone else Can chat. Groups can only be added as Can chat. There's an org-wide toggle that lists the agent in your Agent Store. Changes you make aren't seen by others until you select Update.
- A catch with files: people only get answers from SharePoint files they already have permission to open. If someone gets thinner answers than you do, check their file access first.
- Notebooks can be shared too, but shared notebooks are built for working together, so anyone you share with can typically edit. For a "use, don't change" instruction, an agent is the better fit.
Google Gemini
- Skills in Workspace (built in Workspace Studio): open the skill and use Share to share it with your organization. People who open the link get their own copy, including the files. That means your later edits don't reach copies people already made. Tell people where the current version lives, and when it changes.
- Skills in the Workspace apps and in the Gemini app don't sync with each other right now. If you need one in both places, you create it in both. Your master copy makes that a paste.
Here's what happens to people who only have the copy inside the tool, using two changes that are underway right now.
Gemini Gems to Skills. Skills reached Workspace domains on the Rapid Release track October 5 to 12, 2026, and Scheduled Release domains start on October 19, finishing by mid-November. The Gemini app started on October 13 ([brief](https://appliedaiinsider.com/briefings/2026-10-03), [brief](https://appliedaiinsider.com/briefings/2026-10-05)). On November 17, 2026, Gems move into the Gemini app's Settings but still work. For business and enterprise accounts, Gems stop working no sooner than March 1, 2027, and Google will then turn leftover Gems into draft skills.
Google's advice is to rebuild your important Gems as skills early rather than wait. To move one by hand:
- Open the Gem and copy its name, description and instructions into a master copy document (if you don't have one yet).
- Download its knowledge files and put them in the same folder. If any are Word files, save them as PDF for the Gemini app.
- Create a new skill, paste in the name, description and instructions, and upload the files.
- Run your three test requests through both the Gem and the skill. If the answers match, you're done. If not, fix the master copy and update the skill.
- If anyone built a Workspace Studio flow with an "Ask a Gem" step, that step needs updating too. Tell your admin.
ChatGPT custom GPTs to plugins. Custom GPTs are scheduled to stop running on December 11, 2026 (some Enterprise workspaces can get an extension to February 11, 2027). The migration copies a GPT's instructions into a skill and its knowledge files into reference files, but not its chosen model, its custom actions or its sharing. OpenAI also warns that a migrated plugin may respond differently. So: migrate, rerun your three test requests, and reshare.
A vendor switch. If your business moves from one AI tool to another next year, the same steps apply. Paste the master copy into the new tool's Project, skill or agent, upload the reference files from your drive, run the three tests. An afternoon for three instructions, not a project.
A model change. As in Lesson 5: rerun the tests, compare, fix the master copy if something slipped.
Every one of these is easy if the master copy is current, and painful if it isn't. That's why the next part matters.
Every saved instruction gets:
- An owner. Usually the person who does that work most, or your AI champion. They approve changes, keep the master copy and the tool in step, and rerun the tests.
- A backup. So it doesn't stall when the owner is on vacation.
- A review date, 90 days out. These tools change that fast.
At each review, the owner does four things, in about fifteen minutes per instruction:
- Check the reference files are still current (prices, wording, names).
- Rerun the three test requests and compare with last time.
- Ask one person who uses it: "Anything it keeps getting wrong?"
- Set the next review date and note anything that changed.
Also review early whenever: the model changes, a product renames or retires something (like Gems or custom GPTs), your prices or services change, or the owner leaves.
Put all of this in the Saved Instructions Register: one line per instruction with its owner, backup, which tools it's installed in, who can edit, the last test date and the next review date. It's one page, and it's how you know what your business has.
- Decide the owner and a backup for your first instruction. Write both in the master copy.
- Share it in your tool: everyone who needs it can use it, only the owner (and backup) can edit. If sharing is off, ask your admin, using the steps above.
- Ask one teammate to run one real request through it and tell you if anything looks off.
- If you have any Gems or custom GPTs from Lesson 1's list, copy each one's instructions and files into its own master copy today, even if you won't move it until later.
- Copy the Saved Instructions Register into your shared drive, add a line for this instruction, and put the review date on the owner's calendar.
Check it worked
At least one teammate has used the shared instruction. They can use it but can't edit it. The register has a line with an owner, a backup, where it's installed, and a review date 90 days out that's on someone's calendar. Every Gem or custom GPT you found in Lesson 1 now has its instructions saved in your own document.
---
Next: do it twice more, so your team has three saved instructions it actually uses.
→ Capstone: three instructions your team actually uses
Previous: Test it with three real requests · Next: Capstone: three instructions your team actually uses