Claude Projects vs Custom GPTs: which one for a UK service business in 2026

If you have been using ChatGPT for a while and want to do something more structured with it - a dedicated assistant for proposals, a client communication tool, a content workflow that runs without starting from scratch each time - you will have run into the question fairly quickly: Custom GPT, or Claude Projects?

Both promise roughly the same thing. Give the AI a set of instructions, upload relevant documents, and it will behave like a specialist assistant for your specific business rather than a generic chatbot. In practice, they work quite differently, suit different types of tasks, and have meaningfully different costs. Getting this choice wrong does not break anything, but it does mean spending time building something that turns out to be the wrong tool for what you actually need.

This comparison is written from inside real client builds. We have set up both for UK service businesses - consultancies, letting agents, accountants, coaches - and the patterns are clear enough to give you a direct answer.

What a Custom GPT is and what it is not

Custom GPTs are built inside ChatGPT. If you have a ChatGPT Plus subscription (currently £20 per month), you can create as many as you like. You give the GPT a name, a set of instructions, some reference documents, and optionally enable tools like web browsing or image generation.

The result is a version of ChatGPT that follows your specific instructions by default, references your documents when answering, and can be shared with a link - either privately with your team or publicly.

What Custom GPTs are genuinely good at:

• Following a specific format or structure every time (proposal templates, blog post outlines, client update frameworks)
• Answering questions based on a reference document library (your pricing guide, your product catalogue, your FAQ)
• Maintaining a consistent tone across a team - everyone using the same GPT will get responses in the same voice
• Being shared with clients for self-service use cases (a client-facing FAQ bot, for example)

What Custom GPTs are not good at:

• Remembering things between conversations - each session starts fresh unless you build memory into the instructions
• Complex reasoning over long documents - ChatGPT's context window, while improved, struggles with very long uploads
• Nuanced instruction-following when the instructions are long or contain conditions ("if the client mentions X, respond with Y")

What Claude Projects is and what it is not

Claude Projects is Anthropic's equivalent feature, available on the Claude Pro plan (currently around £18 per month). You create a Project, give it a set of instructions, upload documents, and all conversations within that Project share the same context.

The architecture is meaningfully different from Custom GPTs in one important way: conversations within a Claude Project are persistent. Claude remembers what was discussed in previous conversations within the same Project. This is not a minor feature - it changes how you can use the tool entirely.

What Claude Projects are genuinely good at:

• Tasks that build over time - drafting a proposal across multiple sessions, developing a strategy document incrementally, maintaining a running log of client interactions
• Complex reasoning and analysis over long documents - Claude's context window (200,000 tokens) is one of the largest available, handling lengthy contracts, reports, or research documents without losing the thread
• Following nuanced, conditional instructions - Claude tends to hold complex instruction sets more reliably than ChatGPT, particularly when the instructions have exceptions or conditions
• Writing that needs to sound like a specific person - Claude's instruction-following for tone and voice tends to produce more consistent output

What Claude Projects are not (yet) good at:

• Being shared publicly - there is no direct "share a link to this GPT" equivalent; Projects are for your own use or your team's use within Claude
• Connecting to external tools without a developer - unlike Custom GPTs' Actions feature, Projects do not have a native way to call external APIs or connect to other software without code
• Image generation - Claude does not generate images; if image creation is part of the workflow, this is a genuine gap

The cost comparison

Both tools are available on relatively accessible monthly plans. The cost difference is not the deciding factor for most businesses, but it is worth being clear about.

FeatureCustom GPTs (ChatGPT Plus)Claude Projects (Claude Pro) Monthly cost (single user)~£20/month~£18/month Team plan (per user)~£20/user/month (GPT Team)~£25/user/month (Claude Team) Number of assistantsUnlimited Custom GPTsUnlimited Projects Context window128K tokens200K tokens Persistent memory across sessionsNo (unless using ChatGPT Memory)Yes

For a solo user, the monthly cost difference is small enough to ignore. If you are buying for a team of five, ChatGPT Team at £20 per user is modestly cheaper than Claude Team at £25. But the feature differences matter more than the price difference for most use cases.

Three setup mistakes that kill both tools

Having helped clients build both, 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.

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 dramatically better output.

Mistake 2: Uploading too many documents at once

Both tools allow you to 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 - the AI misses relevant content or blends it incorrectly with something from a different 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 is used for 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. What happens if you give it incomplete information? What happens if you ask it something the documents do not cover? What happens if the client brief contradicts the pricing guide?

The answer to each of these questions will either be reassuring (the assistant handles it gracefully) or it will reveal an instruction gap you need to fill. Better to find these gaps in testing than mid-client-call.

Which tool wins for specific use cases

Rather than a general recommendation, here are direct answers by use case - these reflect what we have built and tested for UK service businesses.

Proposal and quote drafting

Use Claude Projects. Proposals develop over multiple conversations - a brief, a follow-up question, a revision, a pricing discussion. Claude's persistent memory means the assistant builds on previous conversations without you re-explaining context each time. For a consultancy doing 8-10 proposals a month, this alone is worth the switch from a Custom GPT.

Social media and marketing content

Either works; Custom GPT slightly ahead for teams. If you want consistent output that multiple team members can access through a shared link, the Custom GPT sharing feature is a genuine advantage. For a solo operator, Claude's output quality on tone-specific content is marginally better.

Client FAQ / self-service assistant

Custom GPT. The ability to share a GPT via a public link, or to embed it in a client portal, makes Custom GPTs the only practical option here. Claude Projects cannot currently be shared externally.

Document analysis and review

Use Claude Projects. If you need to analyse long contracts, lengthy reports, or multi-document reference materials, Claude's 200K context window handles this more reliably. We have tested both with lengthy UK commercial contracts and Claude consistently maintains coherence across the full document where ChatGPT begins to lose detail.

Email drafting and communication templates

Use Claude Projects for quality; Custom GPT if team access matters. Claude's instruction-following for tone produces more natural output for email work - less "assistant-sounding", more human. But if you need five team members to use the same tool from a shared link, Custom GPT wins on accessibility.

Internal knowledge base / answering questions about your business

Custom GPT for external access; Claude Projects for internal research depth. If you want something your team can query about your processes, policies, and pricing, Custom GPT is simpler to set up and share. For deeper analytical work - "compare these two approaches from our service documents" - Claude handles the nuance better.

Can you use both?

Yes, and increasingly that is what sophisticated small business operators do. The monthly cost of both plans together is under £40. For a business doing 20+ AI-assisted tasks a week, the productivity gain from using the right tool for each task is easily worth that.

A practical setup: Claude Projects for deep work (proposals, analysis, long-document review, strategic thinking); Custom GPTs for operational consistency and team access (email templates, social content, client-facing tools).

The thing both tools need that most businesses skip

Neither Custom GPTs nor Claude Projects produce great results straight out of the box. They require investment upfront - clear instructions, the right documents, and a testing phase - and they require maintenance over time as your business processes change.

Most businesses that are disappointed with these tools invested 20 minutes setting them up and then wondered why the output was inconsistent. The businesses that get real value from them spend three to four hours on initial setup, refine the instructions over the first two weeks, and revisit them quarterly.

If you want help working out which of these tools is right for your specific workflows, and what an effective build would look like for your business, that is exactly what the Agent Build scoping call is for.

Ready to build something that actually works?

A HoursBack Agent Build scoping call walks through your current workflows, identifies where a Custom GPT or Claude Project would produce measurable time savings, and scopes a build that we can deliver for you. No guesswork, no wasted setup time.

Book an Agent Build scoping call

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.