Skip to content
EN DE

Custom GPTs: Building Specialized AI Assistants

L3 Lesson 3 of 5 — Context as Infrastructure
1
2
3
4
5

Claude Projects are workspaces — you work inside them. Custom GPTs Specialized versions of ChatGPT configured with custom instructions, knowledge files, and optional API actions. They can be used privately, shared via link, or published in the GPT Store. take a different approach: You build a specialized assistant that’s particularly good at one thing. Less workspace, more tool.

The concept is related — persistent context through predefined rules and knowledge — but the implementation differs.

Every Custom GPT consists of three elements:

ElementWhat It DoesRequired?
InstructionsSystem prompt — role, rules, behaviorYes
Knowledge FilesUploaded documents as referenceOptional
ActionsAPI calls to external servicesOptional

The core. A free-text field (up to ~8,000 characters) that loads as the System Prompt A hidden instruction that sets the AI's behavior for the entire conversation — it's set before the actual conversation starts and is typically invisible to the user. in every conversation. This is where you define who the GPT is, what it can do, and how it should behave.

Documents the GPT uses as a knowledge base. Formats: PDF, DOCX, TXT, CSV, and more. Files are automatically indexed — the GPT searches them when a question is relevant.

The distinguishing strength of Custom GPTs: Through Actions API calls that a Custom GPT can make to external services — configured via an OpenAPI specification. Enables real tool integration, e.g., fetching data or triggering operations. , a GPT can call external APIs. This turns it from a conversation partner into a real tool — it can fetch data, trigger calculations, or write to external systems. This requires technical knowledge (OpenAPI specifications), though.

In ChatGPT: “Create a GPT” (available for Plus, Team, and Enterprise users). The GPT Builder offers two modes:

  • Create (guided): ChatGPT asks questions and builds the GPT interactively
  • Configure (manual): You write instructions, upload files, and configure actions directly

For maximum control, use Configure mode.

Example for a meeting notes GPT:

You are a meeting assistant for a product team.
Your task:
When given meeting notes or a transcript, create a structured summary.
Format:
- Attendees (who was there)
- Decisions (what was agreed)
- Action items (who does what by when)
- Open questions (what still needs clarification)
Rules:
- Keep the summary under 300 words
- Format action items as "Who + What + By when"
- For unclear points: flag as "Unclear from notes"

For the meeting GPT, for example: a list of team members with their roles, the project roadmap, the meeting protocol template. The GPT uses these files as reference — it knows the names, understands who’s responsible for what, and uses the right format.

Three options:

OptionFor Whom
Only mePersonal assistant
Anyone with a linkTeam or selected people
GPT StorePublic for all ChatGPT users

What Custom GPTs Do Well — and What They Don’t

Section titled “What Custom GPTs Do Well — and What They Don’t”
  • Visual builder: No code required, accessible even for non-technical users
  • Actions: Real tool integration via API calls — something no Claude Project can do
  • Shareability: Share via link or publish to a store with millions of users
  • Broadest user base: ChatGPT has the largest user base of any AI platform
  • Knowledge files aren’t secure: With targeted prompts, users can extract the content of uploaded files. Don’t upload confidential data you wouldn’t want shared.
  • No real memory: Every conversation with a Custom GPT starts from zero. There’s no cross-session memory.
  • Single context type: Instructions and knowledge files — no separation between “rules” and “reference knowledge” like Claude Projects.
  • GPT Store monetization discontinued: The revenue program for the GPT Store was shut down in mid-2025.

Comparison: Custom GPTs vs. Claude Projects

Section titled “Comparison: Custom GPTs vs. Claude Projects”
AspectCustom GPTsClaude Projects
MetaphorSpecialized toolPersistent workspace
Context setupInstructions + files in one fieldInstructions and knowledge base separated
Multi-chatEvery chat starts freshKnowledge base applies across all chats
ShareabilityGPT Store, link, or privateTeam sharing (Team/Enterprise)
API integrationYes (Actions)No
Knowledge securityExtractableProtected within project
AvailabilityChatGPT Plus/Team/EnterpriseClaude Free (5 projects) through Enterprise

Custom GPT when you:

  • Need a specialized assistant for a recurring task
  • Want to share the assistant with others (even outside your team)
  • Need API integration (e.g., pulling data from a CRM)

Claude Project when you:

  • Need a workspace for ongoing work (multiple chats, one context)
  • Want to use confidential documents as a knowledge base
  • Value the separation of rules and reference knowledge

Both when you use different AI tools for different tasks — which is the most sensible approach for most knowledge workers.

Create a Custom GPT for a task you do at least weekly. Write instructions with role, task, format, and rules. Test with 3 real requests.

Upload a non-confidential document as a knowledge file (e.g., a public style guide). Ask questions about it. Observe how the GPT uses the document — and where it ignores it.

Take the same task and set it up as both a Custom GPT and a Claude Project. Send identical requests to both. Note the differences in quality, convenience, and control.

Custom GPTs and Claude Projects solve the same fundamental problem — persistent context — with different philosophies. GPTs are specialized tools for sharing. Projects are workspaces for working. In the next lesson, you’ll see a fundamentally different approach: Microsoft 365 Copilot, where context isn’t uploaded but comes live from your everyday work.

Part of AI Learning — free courses from prompt to production. Jan on LinkedIn