Does Claude have a Custom GPT equivalent? Claude Projects, now GPTs are retiring

Yes. The nearest like-for-like is Claude Projects: your own instructions and reference files, loaded into every chat inside the project, on every plan including Free. Skills cover reusable instructions that work across any chat, not just one project, and plugins bundle skills, connectors and sub-agents together on paid plans.

That question matters more than it used to. OpenAI is retiring Custom GPTs on 11 December 2026 and moving them to "plugins" - the same shape as Claude's own answer. Personal ChatGPT accounts (Free, Go, Plus and Pro) can no longer create new ones; existing GPTs keep working until the cut-off. The detail is in OpenAI's Custom GPT retirement and migration FAQ.

We have built both for consultancies, letting agents, accountants and coaches. The pattern is consistent. Give it twenty minutes and you get output nobody trusts, abandoned inside a month. Give it three or four hours, refined over a fortnight, and you get something a team actually uses. The setup time matters more than which of the two products you pick.

What a Custom GPT is, and why it is being retired

Custom GPTs are built inside ChatGPT. Until recently, anyone with a Plus subscription could build one. That has changed: only ChatGPT Business, Enterprise and Edu workspaces can create new GPTs now, and personal accounts cannot. Existing GPTs still work, but OpenAI has scheduled all of them to retire on 11 December 2026, migrating to a new "plugins" format.

A GPT has no memory. Every conversation starts fresh, whatever was discussed the day before. GPTs can still connect to external tools through Actions, but custom Actions do not carry over when a GPT migrates to a plugin, so anything built on them needs rebuilding on the other side.

Separately, ChatGPT also has its own Projects, on every plan including Free, which keep chats, files and instructions together so a task can carry on from where it left off. This is OpenAI's own answer for ongoing work, and it works differently from a GPT, which does not remember anything between chats. For the fuller picture of what ChatGPT can do day to day, our ChatGPT guide covers the basics.

What Claude Projects is

Claude Projects sits on every plan, including Free, where you can create up to five. Give a Project a set of instructions, upload reference files, and every chat inside it works from the same context.

Each Project has its own memory, separate from every other Project. On Free, Pro and Max, that memory is switched on by default. On Team and Enterprise it is switched off by default, until an owner turns it on for the organisation. Projects can be shared with colleagues on Team and Enterprise plans, but only with people inside your own organisation - there is no public link and no client-facing version.

Claude connects to outside tools through connectors, on every plan including Free (limited to one custom connector on Free). Google Drive, Gmail and Slack are ready-made examples, no developer needed. Claude does not generate photos or illustrations, though it can build charts and diagrams, currently in beta.

Skills and plugins cover the rest. Skills are reusable instructions that load automatically in any chat, not just one Project. Plugins bundle skills, connectors and sub-agents together on paid plans. Between the three, most of what a Custom GPT used to do is covered.

Three setup mistakes that kill both tools

We have helped clients build both, and there are three setup mistakes we see repeatedly. They apply equally to Custom GPTs and Claude Projects, and they are the reason most first attempts feel disappointing.

Mistake 1: Treating instructions like a mission statement

Most people write instructions that describe what the assistant is for rather than how it should behave. "You are a professional proposal writer for a consulting firm. You help create compelling proposals." This tells the AI almost nothing it does not already know.

Effective instructions describe behaviour: what format to use, what to include and exclude, what to ask when information is missing, what tone to take in different situations. The more specific the instruction, the more consistently the assistant will follow it. Our prompt library has worked examples written this way.

A better version: "When I provide a client brief, produce a proposal with these five sections in this order: [section list]. Keep the executive summary to under 150 words. Use second person (you/your) throughout. If the brief does not include a budget figure, ask for it before drafting. Never include case studies unless I specifically provide them."

These instructions take ten minutes longer to write and produce far more consistent output.

Mistake 2: Uploading too many documents at once

Both tools let you upload reference documents that the AI draws on when answering. The instinct is to upload everything: the full product catalogue, the complete pricing guide, past proposals, terms and conditions, company history.

The result is an assistant that produces confused, hedging responses because it is trying to reconcile conflicting or overlapping information from too many sources. Longer documents also reduce the reliability of the retrieval, so the assistant misses relevant content or blends it with the wrong document.

Start with one or two documents that directly support the specific task the assistant is for. A proposal writer needs the pricing guide and the service descriptions, not the company history and the staff handbook. Add documents only when you encounter a genuine gap, not as a precaution.

Mistake 3: Skipping the test phase

Both tools feel impressive when you first set them up, because you are using them with tasks you have mentally prepared for. The failure modes only appear when the assistant meets something slightly outside the instructions: an unusual client request, an edge case in your process, a question the documents do not answer.

Before using either tool in a real business workflow, run ten test prompts that include edge cases. Try it with incomplete information, a question the documents do not cover, and a client brief that contradicts the pricing guide.

The answer will either be reassuring, because the assistant handles it gracefully, or it will show you an instruction gap to fill. You want to find those gaps in testing, not mid-client-call.

If you already rely on a Custom GPT

Existing Custom GPTs keep working until 11 December 2026 (Enterprise workspaces with an approved deferral get until 11 February 2027). Nothing needs doing today, but two things are worth planning for before the date arrives.

OpenAI's migration path turns a GPT's instructions into a skill inside a new plugin, and any connected apps carry over as apps within it. Custom Actions do not transfer, so a GPT that calls an external API needs that connection rebuilt on the other side.

The alternative is to move the same instructions and reference files into a Claude Project instead, sidestepping the migration entirely. Either route needs the same groundwork: clear instructions, the right one or two documents, and a proper test pass. The retirement is a live example of the risk in our piece on AI vendor lock-in: the more a workflow leans on one provider's features, the more work it takes when that provider changes them.

Custom GPTs and Claude Projects compared

Feature	Custom GPTs (ChatGPT)	Claude Projects
Status	Retiring 11 December 2026	Available on every plan
Who can create one	Business, Enterprise and Edu workspaces only	Anyone, including Free accounts (up to 5)
Single-user price	ChatGPT Plus: £20/month	Claude Pro: £18/month, or £15/month billed annually
Team price, billed monthly	ChatGPT Business: £18/user/month	Claude Team: £22.50/user/month
Team price, billed annually	ChatGPT Business: £15/user/month	Claude Team: £18/user/month
Memory between chats	No, every chat starts fresh	Yes, per project (on by default on Free/Pro/Max, off by default on Team/Enterprise until switched on)
Share with people outside your business	Only via an existing public link, until retirement	Not possible - Team and Enterprise sharing stays inside your organisation
Connect to other tools	Yes, via Actions (does not transfer to the plugin migration)	Yes, via connectors on every plan, no developer needed
Make images	Yes	No (can build charts and diagrams, in beta)

Prices are as at 28 September 2026, taken from ChatGPT's pricing page and Claude's pricing page. Claude's prices include VAT; OpenAI's page does not say either way. Check both before you buy, since plan names and prices change often.

What to use each tool for

Proposal and quote drafting

Claude Projects. Proposals develop over multiple conversations - a brief, a follow-up question, a revision, a pricing discussion. Claude's per-project memory means the assistant builds on earlier conversations without you re-explaining context each time. Our breakdown of where a consultancy's proposal hours go shows the setup in practice.

A client-facing assistant

Neither, on its own. A Custom GPT cannot be embedded in your website or a client portal; OpenAI's own guidance is to use the API for that instead. Claude Projects cannot be shared outside your organisation at all. If you want an assistant clients can use on your own site, that is a build, not a setup task inside either tool. That is the kind of job our AI agent builds are for.

Document review and long contracts

Claude Projects. For lengthy contracts, reports or multi-document reference material, Claude's larger context window on its newest models handles long documents more reliably without losing the thread partway through.

Team knowledge base

Claude Projects, if the team is on Team or Enterprise. A shared Project keeps processes, pricing and policies in one place colleagues can query. A Custom GPT could do this while it lasted, but building a new one on a personal plan is no longer possible, and any existing one has a retirement date attached.

Can you use both?

Yes, and while GPTs still exist, some businesses do. ChatGPT Plus and Claude Pro together cost £38 a month for one person. For a business running twenty or more AI-assisted tasks a week, the gain from using the right tool for each task is easily worth that, and it gives you time to move anything GPT-dependent across before December. If you are not sure the subscriptions you already have are paying back, the free AI tool stack audit checks.

Why most of these get abandoned within a month

Neither tool produces great results straight out of the box. Both need clear instructions, the right documents and a proper testing phase, plus maintenance as your processes change. It is the same pattern behind most AI automation attempts that fail. The five-day implementation plan we give every client shows what finishing the setup looks like.

If you want help working out which of these tools fits your specific workflows, and what it would take to set one up properly, that is what the assessment is for.

Common questions

Does Claude have an equivalent to Custom GPTs?

Yes. Claude Projects is the closest match: your own instructions and files, loaded into every chat inside the Project. It is available on every plan, including Free, where you can create up to five.

What is the Claude equivalent of a Custom GPT?

There is no single feature with that name. Projects cover instructions and reference files, Skills cover reusable instructions that work across any chat, and Plugins bundle skills, connectors and sub-agents together on paid plans. Between the three, most of what a Custom GPT did is covered.

Can you build a custom GPT inside Claude?

Not by that name, but you can achieve the same outcome. Set up a Claude Project with your instructions and one or two supporting documents, and every chat inside it will follow them, the same job a Custom GPT was doing.

Are Custom GPTs being discontinued?

Yes. OpenAI has scheduled Custom GPTs to retire on 11 December 2026, moving them to a new "plugins" format. Personal ChatGPT accounts can no longer create new GPTs; only Business, Enterprise and Edu workspaces still can, until the retirement date.

Are Claude Projects free?

Yes, on the Free plan, up to five projects. Paid plans are not capped at five and, on Team and Enterprise, add the option to share a project with named colleagues inside your organisation.

Start with an AI workflow assessment

An AI workflow assessment is a 60-minute session that maps where your time actually goes, then tells you which of these tools would earn its setup time in your business and which would sit unused. If a build is the right answer, we scope it from there. Our walkthrough of what actually happens in a HoursBack assessment covers the process end to end.

Start with an AI workflow assessment

Or if you are earlier in the process and want to understand where AI can help your business before deciding what to build, start with the free 2-minute AI readiness quiz.