If you’re using AI to write, research, or automate any part of your online business, you’ve probably hit the same wall: the tool forgets everything between sessions. You end up re-explaining your audience, your voice, your product, and your preferences every single time.
Both Anthropic and OpenAI offer solutions. Claude has Projects. ChatGPT has custom instructions (and now GPTs, though those are a different beast). They solve the same problem—context persistence—but they work in fundamentally different ways, and picking the wrong one costs you time.
How Claude Projects work
Claude Projects let you create a named workspace with its own knowledge base. You upload documents, paste style guides, add background about your business, and then every conversation inside that Project has access to that material.
Each Project can hold up to 200,000 tokens of reference material (roughly 150,000 words). That’s enough for a full brand guide, product documentation, past article archives, customer research notes, and your editorial calendar all in one place.
The big advantage: you can have multiple Projects. One for customer support replies, one for content writing, one for course development. Each stays isolated. The AI doesn’t mix context from your SaaS docs into your newsletter drafts.
Projects are available on Claude Pro ($20/month) and the API. The free tier doesn’t get them.
How ChatGPT custom instructions work
Custom instructions are a single set of persistent guidelines that apply to every conversation in your account. You get two text fields: one for context about you, one for how you want ChatGPT to respond.
The context field is where you explain your business, audience, products, and role. The response field is where you set tone, format preferences, and constraints (e.g., “always use American English,” “never use exclamation marks,” “keep answers under 300 words”).
The limitation: it’s global. You can’t scope instructions to specific use cases unless you manually turn them off, edit them, or work around them in your prompts. If you want formal tone for client emails and casual tone for social posts, you’re stuck toggling or overriding.
Custom instructions are free on all ChatGPT tiers, including the free plan. GPT-4 and GPT-4o respect them. GPTs (the custom chatbot builder) bypass custom instructions entirely and use their own system prompts instead.
When to use Claude Projects
Use Projects when you have distinct workflows with different context needs. If you’re writing a newsletter, managing a membership community, and building a course, those are three separate Projects. Each gets its own voice guide, audience research, and example library.
Projects also shine when you need to reference large documents repeatedly. Upload your entire content archive, and Claude can pull from it without you pasting excerpts into every prompt. It’s particularly useful for research-heavy work—competitor analysis, SEO audits, or synthesizing reader feedback.
The major downside: you have to remember to switch Projects. If you’re in your “newsletter” Project and ask a question about your course, Claude won’t know what you’re talking about. It’s not smart enough to auto-switch.
When to use ChatGPT custom instructions
Use custom instructions when your needs are consistent across everything you do. If you’re a solo operator with one brand, one voice, and one audience, global context works fine. Set it once, forget it.
Custom instructions also work well for constraints and formatting preferences that apply universally. Things like “never use jargon,” “default to bullet points,” or “cite sources inline” belong here. You don’t want to re-specify those in every Project or GPT.
The catch: if your work varies in tone or scope, custom instructions become a lowest-common-denominator compromise. You’ll end up writing longer prompts to override the defaults, which defeats the purpose.
The hybrid approach
Most operators I’ve talked to don’t pick one. They use both tools for different parts of the business.
Claude Projects handle the heavy, structured work: content production, documentation, research synthesis. ChatGPT with custom instructions handles the quick, ad-hoc stuff: drafting emails, brainstorming headlines, troubleshooting code snippets.
If you’re paying for both—Claude Pro and ChatGPT Plus both run $20/month—this split makes sense. If you’re only paying for one, pick based on where you spend the most time. Content-heavy businesses lean Claude. Generalists lean ChatGPT.
Reply to this email if you’re using Projects or custom instructions in a way that’s working (or not). I’m collecting notes for a follow-up on how operators are actually structuring their AI workflows.