Custom GPTs and Claude Projects: reusable assistants with knowledge files

The single biggest productivity gain from AI in 2026 is not the model — it is what you build around the model. Anything you do with AI more than twice is a candidate for a permanent, reusable assistant: a Custom GPT in ChatGPT, a Project in Claude, a Gem in Gemini, or an Agent in Microsoft Copilot.

These reusable assistants combine three things in one artefact: a set of instructions, a set of reference files, and the conversation history that builds up inside them. Every time you start a chat inside one, you skip the "let me re-explain my context" tax.

This article is how to build them, when to use which, and the patterns that compound over months of use.

What they each are

ChatGPT Custom GPTs. Created at chat.openai.com/gpts/editor or via the sidebar "Create a GPT" option. Each Custom GPT is a saved set of:

• A name and description
• Instructions (up to a few thousand words)
• Knowledge files (uploaded PDFs, docs, spreadsheets, etc.)
• Conversation starter prompts
• Capability toggles (browsing, image generation, code interpreter, advanced data analysis)

Custom GPTs live in your "My GPTs" list. You can publish them publicly to the GPT Store or share via link. Requires ChatGPT Plus, Team, Enterprise, or Edu.

Claude Projects. Created in claude.ai. Each Project is a folder containing:

• Custom instructions (specific to that project)
• Knowledge files (documents you upload that Claude can reference)
• All conversations you have in the project

Projects live in your sidebar. They are private to you (or to your team in Claude for Work). Requires Claude Pro, Team, or Enterprise.

Gemini Gems and Microsoft Copilot Agents. Equivalent features. Gemini's are simpler and tied to your personal context; Copilot's are more enterprise-oriented and can connect to your organisation's data.

The mechanics differ slightly across platforms, but the idea is identical: a saved AI assistant with your context, your style, and your reference materials pre-loaded.

When to build one

The reliable trigger is repetition. Three signs you should build a reusable assistant:

1. You keep re-explaining the same context. If three of your last ten conversations started with "I work as a product manager at..." or "Our company sells..." or "Our brand voice is..." — that context belongs in a saved assistant.

2. You keep asking variations of the same question. Drafting emails in your voice. Reviewing contracts for the same kinds of risks. Generating reports in the same format. Translating to or from a specific language.

3. You have a corpus of reference material you keep pasting in. Your brand guidelines. Your product documentation. A long policy document. A set of past meetings. If the same files keep coming up, they want to live in a Project or Knowledge area.

If you find yourself thinking "I wish ChatGPT just knew this about me" — you can make it.

A practical first build: the email-drafting assistant

Let's walk through one concrete build. We'll use ChatGPT Custom GPT as the example, but the same instructions work in Claude Projects with minor adjustments.

Step 1: Create the GPT. Go to chat.openai.com → sidebar → Create a GPT → Configure.

Step 2: Name and describe.

• Name: "Email Coach"
• Description: "Drafts and edits emails in my voice, with my situation in mind, under 60 seconds."

Step 3: Write the instructions. This is where most of the value lives.

You are an email writing coach for [your name], a [your role] at [your company] in [your location]. Your job is to draft and edit emails fast, in their voice.

>

Their voice: direct but warm, no corporate filler, prefers shorter emails, ends with a concrete next step.

>

When asked to draft an email: 1. Ask one specific clarifying question if a key detail is missing — otherwise just draft. 2. Produce three versions: short (60 words), medium (100 words), longer (150 words). Label each with its tone. 3. Do not include "I hope this email finds you well", "I wanted to reach out", "Thank you for your patience" or "Please let me know if you have any questions" in any draft. 4. End each draft with a clear next step.

>

When asked to edit an email: 1. Do three passes in order — clarity, tone, grammar — but only the pass requested. 2. Quote each suggested change and explain in one short sentence why. 3. Never rewrite for "style" without being asked.

>

When asked to chase, decline, or apologise: - Chase: warm, not pushy, soft deadline, acknowledge the prior thread. - Decline: warm, no over-explanation, no apology for the boundary. - Apologise: take ownership briefly, propose a fix, do not grovel.

>

Default output format: three labelled versions, separated by horizontal rules. No preamble. No closing summary.

Step 4: Conversation starters. Three or four prompts your future self might type:

• Draft an email declining this meeting politely
• Rewrite this email shorter
• Edit this email for tone — should sound calmer
• Help me write a hard message to my manager

Step 5: Knowledge (optional). If you have a brand style guide, a company tone-of-voice document, or example emails you have written that capture your voice, upload them here. The GPT will reference them.

Step 6: Save and use. From now on, every time you need to draft an email, click on the "Email Coach" GPT in your sidebar instead of starting a generic ChatGPT conversation. You skip the introduction every time.

After a week, edit the instructions to fix anything that did not work the way you wanted. After a month, you have a tool calibrated to your voice that produces nearly-ready drafts on the first try.

A few more builds that work consistently well

To give you a flavor of the range:

The patient tutor. Loaded with the four-step learning loop (explain, examples, quiz, revision). Use it whenever you want to learn something new.

The decision sparring partner. Forces you to answer five questions first, then lists arguments for and against, then gives a calibrated recommendation. Use it for any real decision.

The contract reviewer. Instructions for the three-pass document workflow (first impressions, risks, decisions). Upload your relevant law or policy documents as knowledge. Use it on every contract you receive.

The technical writer. Instructions for your team's documentation style. Knowledge files: your existing docs. Use it for every new doc you draft.

The customer interview synthesiser. Loaded with instructions for extracting themes, quotes, pain points, and feature requests from raw interview transcripts. Use it after every customer call.

The brand-voice rewriter. Instructions for matching your company's voice. Knowledge files: 5-10 examples of writing that hit the voice perfectly. Use it on any external content you draft.

Each of these takes 10-30 minutes to build. Each one saves 10-30 minutes per use, every use, for as long as you keep using it.

Custom GPT vs Claude Project: when to choose which

A short decision guide:

Choose Custom GPT when:

• You want to share with others or publish to the GPT Store.
• You need the capability toggles (browsing, image generation, code interpreter).
• Your work lives mostly in the ChatGPT ecosystem.
• You want a public-facing assistant.

Choose Claude Project when:

• You want better writing quality (Claude is widely considered the best writer).
• You are working with very long documents (Claude handles long context better in 2026).
• You want your conversations grouped by project context.
• You prefer Claude's interface and tone.
• You want stronger defaults around data privacy.

For most users with both subscriptions, the pattern that works is: build the writing-heavy ones in Claude (Email Coach, Tutor, Sparring Partner), and the integration-heavy ones in ChatGPT (anything that needs image generation, browsing, code interpreter, or sharing).

Common mistakes to avoid

Instructions that are too long. Three pages of instructions is worse than three paragraphs. The model handles 200-500 words of instructions reliably; beyond that, it starts ignoring some.

Knowledge files that are too large or low-quality. Uploading every document you ever wrote into one Project does not make the assistant smarter; it makes it slower and more confused. Curate the references.

Not iterating. Your first version of any assistant will be 70% right. Use it for a week, notice what is missing or wrong, and edit. The second version is much better. The fifth version is sharp.

Building too many. A common over-engineering trap is to build 20 Custom GPTs because you can. Resist. Five excellent assistants that you actually use beats twenty that you forget about. Build only what you would reach for at least once a week.

Not sharing within the team. If you and your team use the same brand voice, the same processes, the same reference material — build the assistant once, share it. Both ChatGPT (Team/Enterprise) and Claude (Team/Enterprise) support this.

A few habits that compound

Once a week, review what you built. Look at the conversations you had inside each assistant. Note any places where the response was wrong or where you had to correct it. Update the instructions accordingly.

When you keep saying the same thing in conversations, push it up to the instructions. Anything you repeat to the assistant in chats wants to live in the assistant's setup.

Keep a "version notes" comment at the bottom of each assistant's instructions. "v3 — added the 'no I hope this finds you well' rule." Helps you remember why you changed things.

Maintain a small library, not a sprawling one. 5-10 well-tuned assistants you actually use, not 50 that you built once and forgot.

The takeaway

Custom GPTs and Claude Projects are the difference between using AI as a chatbot and using AI as a personal tool. They take 10-30 minutes to build, save real time on every use, and accumulate value as you tune them.

Build one this week. Use it for a month. Then build the second. Within a quarter you will have a small set of permanent assistants that handle a meaningful share of your weekly work, and you will wonder why you ever started conversations from scratch.