Custom GPT and Gem migration plan for recruitment agencies

WIITY / RECRUITMENT OPERATIONS GUIDE

Custom GPT and Gem migration plan for recruitment agencies

By Jordan Betteridge · Originally 1 November 2026 · Reviewed 2 November 2026

A practical plan to audit, back up, migrate and test the AI tools your recruitment team already relies on. Confirm your account’s in-product notices before acting: availability, permissions and rollout dates can differ.

What’s changing

OpenAI schedules Custom GPT retirement for 11 December 2026 and is moving reusable workflows to plugins. Google begins transitioning personal-account Gems to skills on 17 November 2026. Existing workflows need review even when a conversion is automatic: instructions, files, connected systems and team access may behave differently.

Existing GPT conversations remain accessible after retirement; they are not migrated into a replacement workflow. Once a GPT is migrated, the original becomes read-only and its creator cannot delete it. Maintain the replacement plugin instead.

Why it matters for a recruitment agency

Agencies can have more custom tools than leadership realises. Some were built by one consultant and shared around the desk. A migration is not complete until the people who depend on a tool can access its replacement and get reliable results.

The biggest risk is not just losing a tool. It is a tool that still runs but gives different answers, and nobody notices.

01 / Audit what you have

Find every GPT and Gem the business relies on before migrating anything. One shared inventory prevents duplicate work and orphaned tools.

1. Ask every consultant, team lead and operations person to list the GPTs and Gems they use, including tools built by someone else.
2. Review My GPTs for each builder. In a managed workspace, coordinate with the administrator or owner to retrieve the available inventory. Permission to use a GPT is not permission to migrate it.
3. Record each tool in the table below. Mark tools that handle candidate or client data.
4. Review tools not used in the last 60 days. Agree what to keep, merge or retire before migration; follow retention requirements and do not delete required records.

02 / Back everything up

Save each tool’s configuration in an access-controlled business folder so you can recover it if conversion changes the behaviour.

• ☐ Instructions: copy the full text into a document named after the GPT or Gem.
• ☐ Knowledge files: download templates, style guides and scoring sheets you are permitted to retain.
• ☐ Actions and connections: record schemas, endpoints, authentication method and permission scopes. Do not copy passwords, tokens or API keys into the inventory.
• ☐ Conversation starters: preserve useful entry points for adoption.
• ☐ Model and settings: note the selected model and enabled tools; the chosen GPT model does not carry over.
• ☐ Sharing: record the people and groups who need access.
• ☐ Good examples: save two or three representative inputs and outputs plus a harder test case, using anonymised data.

03 / Migrate

Move the most-used tools first. Keep one owner responsible for the shared inventory and allow extra time for systems-connected workflows.

ChatGPT: Custom GPTs to plugins

1. Finish and publish essential edits before migrating. Migration uses the latest published version, not drafts. Publishing does not require public sharing.
2. When available in the builder’s account or workspace, open My GPTs and select Migrate to plugin. If the option is missing, check account notices, plugin access and administrator permissions.
3. Compare the migrated instructions and reference files against the backup. Do not assume equivalent behaviour.
4. Rebuild custom actions: they do not transfer through migration. Check whether an approved app supports the task; a custom MCP server may require technical and security review.
5. Consider separating long, do-everything workflows into focused skills, such as job ads and CV formatting. Test how each skill is selected and used.
6. Review and save the intended sharing audience. Confirm users can install and use the replacement before switching the team.

Gemini: Gems to skills

1. Confirm the account type and applicable transition date. Personal and Workspace schedules differ.
2. Download knowledge files before the relevant removal date. Google lists GitHub files, audio, video, legacy Office files (.doc, .xls, .ppt) and archives as unsupported during transition. Check the current list and the 100 MB limit.
3. After automatic conversion, compare each skill’s instructions and files with the backup. Recreate the skill manually where appropriate if you want to migrate earlier.
4. Use the skill invocation shown in the current interface. Google documents a forward slash, with a move to @ planned. Try combining skills only where useful.
5. Review account eligibility, activity settings, sharing and tool availability. Some Gem tools are not available in skills. Confirm your actual workflow works rather than assuming feature parity.

04 / Test and roll out

A tool is not migrated until someone other than the builder has tested it and confirmed the required quality and access.

1. Run the same saved inputs through the replacement, initially with anonymised or synthetic data.
2. Compare tone, format, facts, reference-file use and required output files side by side. Include one harder case.
3. Test each connected system separately in a safe environment, especially write actions. Confirm approval steps and least-privilege permissions.
4. Ask a daily user to trial the replacement. Use real personal data only after the privacy and security checks are complete and your organisation has approved that use.
5. Tell the wider team what it does, how to call it and where to report problems. Confirm their access.
6. Mark the inventory complete and maintain the plugin or skill. Do not plan to delete the original migrated GPT: OpenAI says it becomes read-only and cannot be deleted by its creator.

Data and privacy checks

• ☐ Remove unnecessary CVs, candidate details and client contracts from knowledge files, subject to retention obligations.
• ☐ Identify which account owns each tool. Do not place candidate data in unapproved personal accounts.
• ☐ Verify plan-specific data-use settings, retention, training controls and contracts; do not assume all business plans or configurations are identical.
• ☐ Review what each ATS or CRM connection can read and write. Limit access to what the workflow needs.
• ☐ Review sharing after migration so no plugin or skill is available more widely than intended.
• ☐ Record tools handling personal data in the privacy register and follow applicable privacy requirements.

Timeline and one-page checklist

Start with the audit and backups together, then migrate and test in batches. Aim to finish standard ChatGPT migrations by 10 December 2026, before the scheduled retirement. Allow for unavailable migration options and technical work; do not wait for the deadline. Approved Enterprise or FedRAMP deferrals and later Workspace Gem dates are not reasons to delay preparation.

• ☐ Create a shared list of every GPT and Gem, with its owner and users.
• ☐ Review unused tools and agree which to keep, merge or retire.
• ☐ Back up instructions, files, settings, starters and sharing for every tool you are keeping.
• ☐ Finish and publish essential GPT edits before migration; drafts do not transfer.
• ☐ Migrate the most-used tools first, when the account migration option is available.
• ☐ Rebuild and test custom actions using approved apps or integrations.
• ☐ Compare each replacement with saved examples, including a harder case.
• ☐ Ask someone other than the builder to test quality and access.
• ☐ Review sharing and install or share replacements with the right people.
• ☐ Remove unnecessary candidate and client data and review permissions.
• ☐ Tell the team what changed and how to call each replacement.
• ☐ Aim to complete standard ChatGPT migrations by 10 December 2026; confirm account-specific deadlines.

Beyond the migration

Migration keeps reusable prompts and workflows available. It does not fix the slow work between your ATS, payroll and finance systems. If reporting and manual handoffs are taking your team’s time, explore how Ella and Wiity on Assignment can help connect the wider picture.

Talk to Wiity about your migration or AI roadmap

Sources and scope

• OpenAI: Custom GPT retirement and migration FAQ
• Google: About the transition from Gems to skills
• Virtualization Review, 28 September 2026
• Notebookcheck: Gems to skills report

Dates and details were checked against provider guidance on 2 November 2026 and can change. Follow your account’s notices and current provider documentation. This guide is operational information, not legal advice or a guarantee of availability, equivalence or compliance.