I moved my workshop off ChatGPT. The interesting part isn't the chatbot.
Most ChatGPT vs Claude takes argue about which chatbot is smarter. The actual 2026 shift is Claude becoming an operations layer. Here's what that changed for a one-person workshop, what it means if you run a small team, and what HIPAA practices need to know before they switch.
- Most 'ChatGPT vs Claude' posts are stuck comparing chatbot UX. The 2026 shift is Claude becoming an operations layer: Projects, Skills, Memory Import.
- For a small business with repeatable processes (intake review, copy in your voice, document summaries), encoding the work once into Claude beats re-prompting ChatGPT every Monday.
- HIPAA: Anthropic signs a BAA on Claude Enterprise. Free, Pro, and Team tiers are not HIPAA-eligible. Same rule on the ChatGPT side. Do not paste PHI into a consumer-tier chatbot.
- Don't fully cancel ChatGPT. Image generation, voice mode, and a few specialty GPTs still win there. Run both. The combined cost is less than one freelancer hour a month.
I moved my workshop's AI work from ChatGPT to Claude this spring. I build Jotform automation kits for small businesses and HIPAA-regulated practices, so my work is mostly: drafting client copy, reviewing intake forms, writing receipt emails, mapping integrations, summarizing long compliance docs. The migration mattered. The reason isn't the one most 'ChatGPT vs Claude' posts are arguing about.
The shift nobody is naming clearly
Most comparison pieces in 2026 are still stuck on chatbot UX. Which one writes nicer prose. Which one handles a longer document. Whose math is less wrong this week. Useful, but it misses what actually changed.
The real shift: Claude stopped being a chatbot.
Three features did it. Projects: a persistent context window per workstream, with files, instructions, and chat history pinned together. Skills: small encoded procedures Claude can run on demand, like 'review this intake form for missing required fields' or 'draft the confirmation email in our voice.' Memory Import: a one-click way to take your ChatGPT history with you. Together they turn Claude from a chat surface into something closer to an operations layer.
That phrase gets thrown around. Here's what it means in practice for a small business: instead of re-explaining your context to a chatbot every Monday morning, you encode your context once. Then you ask the system to do the work. Different shape, different leverage.
What I actually moved
Concrete examples from the workshop, in plain language:
- A Project per active client. Pinned: brand voice notes, the kit they bought, their integration map (Stripe products, HubSpot fields, Slack channels), past form drafts. I open one Project to start work for that client. No more re-pasting context.
- A Skill that reviews a Jotform intake against a list of HIPAA red flags. I built it once, an afternoon's work. It runs in 30 seconds per form.
- A Skill that drafts the confirmation email and the internal Slack notification in the client's voice, given a finished form. Was 20 minutes of writing per kit. Now 90 seconds and a quick edit.
- A Project per document type (contracts, BAA forms, audit reports) with the template baked in. I drop a draft in, it returns a structured review against my own checklist.
Less re-explaining. Fewer rewrites. Faster turnaround. The interesting part is that these are ops moves, not chat moves. I'm not having a conversation with the model. I'm running a process.
What I kept on ChatGPT
I didn't fully cancel. ChatGPT still wins at three specific jobs:
- Image generation. DALL-E for kit illustrations and quick mockups when I need a placeholder figure for a client deck.
- Voice mode. Hands-free brainstorming on walks, which I do badly with Claude's current mobile UX.
- The long tail of GPT integrations. A few specialty GPTs I rely on do not have a Claude equivalent yet.
I downgraded my plan and use it for those three jobs. Combined cost of running both Claude Pro and a smaller ChatGPT plan: less than one freelancer hour a month. The 'pick one' framing is wrong for most small businesses. Pick the right one for each job.
For HIPAA practices specifically
The compliance answer is short. Anthropic signs a BAA on Claude for Enterprise. Free, Pro, and Team tiers are not HIPAA-eligible. The same logic applies on the ChatGPT side: BAA on Enterprise, not on the consumer plans.
What this means in practice for a small practice or clinic: do not paste PHI into any consumer-tier AI. Not Claude Pro, not ChatGPT Plus, not either free tier. If you need AI in a workflow that touches patient data, the path is Claude Enterprise (or ChatGPT Enterprise) plus a signed BAA. Nothing else satisfies the privacy rule.
For the non-PHI work (drafting marketing copy, summarizing public documents, brainstorming intake question wording), the consumer tiers are fine on either side. Two accounts, two purposes.
The 30-minute migration
If you're sitting on a paid ChatGPT plan and curious, this is the smallest version of the move:
- Export your ChatGPT data. Settings, Data Controls, Export Data. The email lands within 24 hours.
- Sign up for Claude Pro. Run Memory Import. Roughly 5 minutes once the export arrives.
- Recreate your top three GPTs as Claude Projects. Pin the same instructions and reference files. About 15 minutes.
- Pick one repeatable process you do every week and turn it into a Skill. About 10 minutes the first time.
- Keep your ChatGPT account for the jobs it still wins at. Downgrade the plan if you can.
That's it. The rest is iteration. The first Skill you write is the slowest. The fifth one takes minutes.
Bottom line
Claude as a chatbot is roughly tied with ChatGPT this month. Claude as an operations layer is ahead today, especially for a small team that already has a repeatable process worth encoding. If you're using AI like a smarter Google, you're leaving the actual leverage on the table. The leverage is in the encoding, not the chatting.
