Custom GPT vs Claude Project vs Gemini Gem: Which Should You Use?

Don't build a custom GPT now: OpenAI is retiring them on 11 December 2026 and moving them to plugins. For shared business context, pick the assistant your team already uses: a ChatGPT Project, a Claude Project, or a Gemini Gem on Google Workspace, which Google is turning into a skill (from November 2026 on personal accounts, March 2027 on business ones).

The title's question assumed three similar ways to build a custom assistant. That's no longer the real choice. Existing custom GPTs stop running on 11 December 2026 (approved Enterprise deferrals run to 11 February 2027), and OpenAI's migration turns each one into a plugin: its instructions become a skill, its knowledge files become reference files and its connected apps come along, but custom actions don't. Google is retiring Gems in stages too: personal Google accounts from November 2026, Workspace business accounts in March 2027, with each Gem converted automatically into a skill.

Follow me on Instagram@sagnikteaches

Claude Projects are the only one of the three not being replaced, though Anthropic is testing a new version of projects that will reach Team plans later. For most small teams the sensible answer is a shared project in whichever assistant your team already uses, with its instructions kept in a document you own, so the next platform change costs an afternoon rather than a rebuild.

Connect on LinkedInSagnik Bhattacharya

Two of the three options are on their way out

So the practical comparison is between ChatGPT Projects, Claude Projects and, for Workspace businesses, a Gem that will become a skill. All three give a team a shared place with instructions and reference files; the differences are in file limits, sharing and cost.

Subscribe on YouTube@codingliquids

What each shared workspace offers today

Two rows decide most small-business cases. Price: a business already on Google Workspace gets Gems at no extra cost, while ChatGPT Business or Claude Team adds $20-$25 a seat. And file handling: Claude's retrieval suits projects with a long handbook or many documents, ChatGPT's 25-40 file cap suits focused projects, and a Gem's ten files suit one tight job.

A bakery's three custom GPTs, re-homed one by one

A bakery with five office and counter staff, used here as an illustration, runs on Google Workspace Business Standard. The owner also has ChatGPT Plus and built three custom GPTs in 2025, all of which will stop working in December. The owner listed each one with what it does, who uses it and what it depends on, then decided its future:

The third decision matters most. The allergen GPT was answering from a spreadsheet exported months earlier, and two recipes had changed supplier since. Allergen information has to come from the current recipe records, checked by someone accountable, so the owner didn't migrate it anywhere. That's the useful side of a forced retirement: it makes you look at every assistant you've built and ask whether it should exist at all.

The cost sum settled the rest. Moving all five staff to ChatGPT Business or Claude Team would cost 5 × $20 = $100 a month on annual billing. The Gem costs nothing extra on their Workspace plan, and the owner's Plus subscription ($20 a month) already covers the caption project. Keeping the shared job in Workspace, where the order emails arrive anyway, avoided a new subscription. For a team without Workspace, the equivalent comparison is set out in Gemini or ChatGPT for a Google Workspace business.

A migration checklist, filled in for one GPT

The owner worked through the same list for each GPT that was worth keeping. The wholesale order helper's copy:

GPT: Wholesale order helper           Users: owner + 2 office staff
[x] Instructions copied into a Google Doc (master copy, dated)
[x] Knowledge files downloaded: product list, delivery routes,
    order cut-offs (each checked for being current)
[x] Custom actions? None. (Actions don't migrate; any would
    need rebuilding with a supported connector.)
[x] New home: Gem in Workspace, shared Viewer with 2 people
[x] Test: 10 real order emails from last month, old vs new
[ ] Run both for two weeks, then stop using the GPT
[ ] Delete the GPT before 11 December 2026

The master copy in a Google Doc is the step people skip and later regret. Instructions written into a GPT, a Gem or a project are awkward to move; instructions kept in a dated document you own can be pasted into whatever the next platform calls its instruction box. When the Gem becomes a skill in March 2027, the owner will check the automatic conversion against that document rather than trying to remember what the Gem was told.

Ask staff one more question while you inventory: which GPTs made by other people do you use at work? The bakery's counter staff had been using a public recipe-scaling GPT nobody in the office knew about. It retires on the same date, and its replacement had to be a short, checked conversion table rather than another bot.

One set of instructions, three homes

The instructions barely change between platforms. Here is the order helper's, written once:

You help the bakery's office staff reply to wholesale orders from
cafes. Use only the product list, delivery routes and order
cut-offs in your files.

Before agreeing to any change of delivery day, check the delivery
routes file: Route A runs Tuesday and Friday, Route B Monday and
Thursday. If the requested day isn't on the customer's route, say
so and offer the nearest route day.

Orders close at 2pm two working days before delivery. Never accept
a change after the cut-off; draft a polite refusal with the next
available date instead.

Never quote prices; wholesale price bands are handled by the owner.
Replies under 100 words, friendly, signed "The bakery office team".

In ChatGPT it goes into the project's instructions; in Claude, under Set project instructions; in a Gem, in the Instructions field when you create it. When the Gem becomes a skill, the same text becomes the skill's instructions, and Google's advice for skills is to say when the skill should be used and to include examples of the output you expect, which is worth adding to the master copy now. The fuller walkthroughs for each platform are in using ChatGPT Projects to keep client work separate, setting up Claude Projects as a shared team assistant and building Gemini Gems for repeat business tasks.

Testing the new home against the old GPT

For two weeks the office ran each real order email through both the old GPT and the new Gem. One example shows why the route rule is in the instructions. An email from a café read: "Can we switch this week's 4 dozen croissants and 2 sourdough to Thursday instead of Friday?" The first illustrative Gem draft, before the route rule was added, replied: "Of course! We've moved your order to Thursday. See you then." The café was on Route A, which doesn't run on Thursdays, so the bakery would either have made a special trip or let the café down.

With the route rule added, the draft became (illustrative): "Thanks for letting us know. Your delivery route runs on Tuesdays and Fridays, so Thursday isn't possible this week. We can bring the order forward to Tuesday if that helps, or keep it on Friday as planned. Just reply to confirm." Nine of the ten test emails matched or beat the old GPT; the tenth, an order with a typo in the product name, was flagged for a person, which was the right result.

How the replacements behave day to day

Both retirements replace a "custom bot" you open with reusable instructions you call from ordinary chats. That changes habits, so it's worth knowing before you migrate.

Plugins and skills in ChatGPT

OpenAI's migration guide describes a migrated GPT as a plugin: the instructions and reference files become a skill, and any connected apps stay part of the plugin. Migrated plugins can be used in ordinary chats when plugins are enabled for the user, and in ChatGPT Work, the agent mode that replaced ChatGPT agent. Sharing a plugin with individuals or groups needs a "Share plugins" permission, so on a Business workspace an admin decides who can pass them round. Three limits catch people out: only a published GPT can be migrated, the GPT's chosen model doesn't carry over, and custom actions have to be rebuilt with a supported connector.

For a three-person team with four GPTs, a realistic plan looks like this: list the GPTs and their owners this week; decide for each whether it becomes a shared project, a plugin, or nothing; rebuild and test one per week through October and November, allowing about an hour to rebuild and half an hour to test each; and stop using the old versions by early December, leaving a margin before the 11 December cut-off. Four GPTs is roughly six hours of work spread over six weeks, far less painful than discovering on 12 December that the quoting assistant has gone.

Skills in Gemini

Google's skills work differently from Gems. Instead of opening a Gem, you type / (Google says this will change to @) followed by the skill's name in any chat, and Gemini can also apply a skill automatically when it recognises that one is relevant. Skills can be stacked in a single chat, you can have up to 100 active at once, and their names use lowercase words joined by hyphens, so the bakery's Gem will become something like wholesale-order-helper. Two practical differences: skills don't have a page listing past chats the way Gems do, and some tools that work in Gems, such as Canvas and Deep Research, don't yet work with skills. Skills are already available on personal accounts for adults, and Google says it will soon expand them to work and school accounts, so the bakery plans to try skills as soon as they appear on its Workspace accounts rather than on the day the Gem disappears.

Sharing settings that expose more than intended

Each platform shows shared users more than people expect. Google is explicit: anyone with access to a Gem can view its instructions and the files uploaded to it, and editors can change or delete it. In the bakery's first week, the Gem was shared with the whole organisation and its knowledge still included the owner's costings sheet, left over from the old GPT's files. Any member of staff could have opened it. The fix was to remove the costings file, keep wholesale prices out of the Gem entirely (the instructions already said never to quote them), and share with the two office staff by name as Viewers.

The same caution applies elsewhere. Claude's Can view members can see the project's knowledge and instructions. ChatGPT's shared projects give collaborators the project's files and use project-only memory, so nothing from people's other chats leaks in, but everything in the project is visible to them. The rule that works everywhere: a shared assistant's files should contain nothing you wouldn't hand every member on paper. Workspace users can also control Gem sharing centrally, because it follows the organisation's Drive sharing settings, which an administrator can tighten for everyone at once.

Picking by the tools already on your team's screens

• Your business runs on Google Workspace: build a Gem now for shared jobs, share it by name, keep the instructions in a Doc, and expect it to become a skill in March 2027.
• Your team already uses ChatGPT Business: use shared Projects for context and files, and OpenAI's plugins, with their skills, for repeatable procedures that used to be GPTs. Check that plugins are enabled for your workspace before relying on them.
• You work with long documents or a large handbook: Claude Team's projects handle big knowledge sets through retrieval and aren't being replaced; see RAG vs custom GPT vs fine-tuning for when a bigger system is warranted.
• You were going to let customers talk to it: none of these is built for that. Customer-facing assistants need handover, logs and guardrails; whether a custom GPT should answer your customers covers safer options.
• You're a sole trader: ChatGPT Business and Claude Team both need two seats, so one person pays for two (about $40-$50 a month) or uses an individual plan. A personal Plus or Pro plan with a project, or a Gem on Workspace, is usually enough.

Questions about moving off custom GPTs and Gems

What happens to custom GPTs we use but didn't build?

They stop working at retirement along with every other custom GPT. If a GPT from someone else is part of a daily routine, recreate the useful part yourself now: write down what you use it for, the instructions you'd want, and any reference material, then set it up as a project in the assistant your team uses. Test it alongside the old GPT before December.

Will a Gem's knowledge files move across when it becomes a skill?

Google says it will automatically transition Gems, along with supported files, to skills when Gems go away. If you want to move sooner, you can recreate a Gem as a skill by hand, but Google's instructions involve downloading each knowledge file individually and uploading them with the skill. Keep a master copy of instructions and files either way.

Do custom GPT actions carry over to the plugin?

No. OpenAI's migration guide says custom actions don't transfer, the GPT's selected model doesn't carry over, and only a published GPT can be migrated; drafts don't move. Anything your GPT did through an action, such as updating a spreadsheet or CRM, has to be rebuilt with a supported connector before the retirement date.

Further reads

• ChatGPT vs Claude vs Gemini vs Copilot for Small Business (2026) — The wider comparison of the four main assistants for small businesses.
• What Gemini in Google Workspace Includes on Each Business Plan — What Gemini features each Workspace plan includes.
• Is Gemini Safe for Confidential Business Data in Workspace? — Check Gemini's data handling before uploading files to a Gem.
• How to Train ChatGPT on Your Business Information Safely — Setting up ChatGPT Projects with your business context safely.
• How to Build a Company Knowledge Base AI Can Answer From — Build the documents any of these assistants will answer from.
• AI Change Management for Small Teams: A Practical Plan — Move a team from one tool to another without losing them.
• How to Give AI Your Business Context So Answers Fit — Build a one-page business profile and task instructions, store them where each AI tool reads them, and test that answers fit your business.
• Why AI Gives Different Answers Each Time, and How to Fix It — Why the same prompt gets a different answer, why 'set temperature to zero' no longer works, and the fixes and consistency test that do.
• Does ChatGPT Remember Conversations? Memory Settings for Business — The three ways ChatGPT remembers, where the memory switches are in September 2026, and how to stop one client's details turning up in another's work.
• AI Glossary for Business Owners: 50 Terms in Plain English — Fifty AI terms in plain English, grouped by where you'll meet them, each with what it means for your decisions, plus five words vendors like to stretch.
• AI Prompts for Bakers: Pricing, Captions and Customer Replies — Twelve copy-ready prompts for bakers covering costing and prices, social captions, and customer replies, each with what to check before you use the answer.
• AI Engagement Letters for Accountants: Faster Drafts, Clear Scope — Which parts of an engagement letter AI should draft, five prompts with sample outputs, and the partner checks that keep scope clear and terms untouched.
• How Tutors Can Plan a Week of Lessons With AI in an Hour — A timed one-hour routine: a student roster, one batch prompt for every lesson skeleton, materials only where needed, and a check on every answer key.
• AI Employee Onboarding: Plan a New Starter's First Two Weeks — Use AI to draft a realistic day-by-day plan for a new starter's first fortnight, a reading path through your documents, and an assistant they can ask.
• How Much Does a Custom AI Assistant Cost a Small Business? — Three ways to get an AI assistant that knows your business, what each costs up front and every month, and the upkeep that most quotes leave out.
• AI Vendor Lock-In: How to Keep Your Data and Prompts Portable — Where AI lock-in really comes from, which plans can export what in 2026, and a costed exit drill that keeps your prompts and data movable.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: OpenAI's custom GPT retirement notice (verified fact sheet) and ChatGPT migration guide for moving GPTs to plugins; OpenAI help pages on projects (via search results); Anthropic help centre on projects and sharing; Google Gemini Apps Help on Gems, sharing a Gem, uploading files and the transition from Gems to skills. All checked September 2026.