THL ChatGPT → Claude Migration

When to use

Triggers:

• "Migrate [client] from ChatGPT" / "Move [person] to Claude"
• Pre-Discovery notes show an active ChatGPT Plus / Business subscription and dissatisfaction signals (document handling, hallucinations, context loss, data residency concerns)
• User explicitly asks to plan a switch

Does not fire for:

• Clients using ChatGPT for image generation primarily — that's a different analysis (keep it or find a DALL-E alternative like Ideogram)
• Users running niche GPTs that depend on plugins — evaluate case-by-case before recommending the switch

The honest conversation first

Before starting a migration, run the "what you'll gain / what you'll lose" conversation. Don't sell the move.

What they'll gain

• Longer effective context (especially for document-heavy work)
• Claude Projects (superior to Custom GPTs for complex workflows, as of April 2026)
• Claude Code for technical users — a genuinely different class of product, not just a better chat
• MCP ecosystem — more mature than ChatGPT's plugin/tool story
• Extended thinking mode for high-stakes decisions (see claude-4-7-extended-thinking-for-sme)
• Australian data posture — Claude offers clearer zero-retention and enterprise data commitments

What they'll lose

• DALL-E image generation inline. Claude doesn't natively generate raster images. Alternatives: Ideogram, Midjourney, Google's Imagen. If they used DALL-E weekly, plan a replacement.
• Voice mode parity. ChatGPT's advanced voice is still ahead of Claude's voice experience as of April 2026. Flag if voice matters.
• GPT store / custom GPTs discovered via search. Claude Projects are private to the account/workspace. No public discovery.
• Plug-ins from ChatGPT. Map each plug-in they use to an equivalent MCP server, or accept the gap.
• Memory across conversations in the consumer experience. Claude's memory feature is improving but ChatGPT's is deeper as of April 2026. Mitigate via Claude Projects (scoped memory per project).

If any of the losses are operationally painful, run a hybrid model — keep ChatGPT for the narrow use (voice, image, specific GPT) and move everything else to Claude. Don't force a full switch.

Memory extraction prompt

Before cancelling ChatGPT, extract the memory and Custom Instructions. Use this prompt inside ChatGPT:

I'm migrating to another AI tool and need to export everything you know about me so I can re-prime my new workspace.

Please output, in markdown, the following:

## Memories about me
Everything in your long-term memory about my preferences, habits, writing style, ongoing projects, people I mention frequently, and anything else that would help a new AI assistant understand me without me re-explaining.

## My custom instructions (both fields)
The current content of:
1. "What would you like ChatGPT to know about you to provide better responses?"
2. "How would you like ChatGPT to respond?"

## Projects I actively use
For each Project: name, purpose, system prompt if visible, and the kind of work I do in it.

## GPTs I've built or heavily used
For each: name, what it does, and the source prompt if I built it. If I used someone else's, note the creator.

## Recurring prompts
What I ask you to do most often (top 10), in the actual phrasing I use.

Output everything even if it feels redundant. I'll edit down after the paste.

Save the output to [client]_chatgpt_memory_export_[date].md. This is the seed for their Claude setup.

Mapping ChatGPT → Claude

| ChatGPT concept | Claude equivalent | Migration step | |---|---|---| | Custom Instructions | Claude.ai user preferences + per-Project system prompt | Copy Field 1 into user preferences (top of Settings). Copy Field 2 into each Project's instructions. | | Memory | Per-Project memory (scoped) + pinned Project files | Paste the "Memories about me" export into one foundational Project called "About me". | | Custom GPT | Claude Project with system prompt + pinned files | For each GPT, create a Project. Paste the GPT's source prompt as the Project's system prompt. Upload any knowledge files to the Project. | | GPT Store discovery | Not available | Accept the gap. Keep an internal registry of useful Projects. | | Plug-ins | MCP servers | For each plug-in they use regularly, identify the MCP equivalent. Common: Gmail, Calendar, Drive, GitHub, Slack (the Claude.ai connector store has most). | | Code Interpreter | Claude Code OR Claude.ai's tool use mode | For one-off analysis, Claude.ai's tool use works. For developers, install Claude Code. | | Voice conversations | Claude.ai voice mode (narrower) | If voice is critical, discuss the gap honestly. | | Canvas (ChatGPT) | Claude.ai Artifacts | Near-equivalent. Most users transition cleanly. |

Plan selection

Recommend a plan based on usage pattern:

• Claude Pro: single user, mostly Claude.ai web or desktop, some Claude Code. US$20 a month, or US$17 a month billed annually.
• Claude Max: power user with long sessions, frequent Opus use and large file uploads. From US$100 a month.
• Claude Team: small business with shared Projects and admin controls, for 2 to 150 people. Standard seats are US$20 a month billed annually or US$25 billed monthly. Premium seats are US$100 billed annually or US$125 billed monthly.
• Claude Enterprise: SSO, audit logs and custom data retention controls. Self-serve Enterprise is US$20 per seat a month billed annually, plus usage at API rates. Larger organisations can talk to sales.

Prices were checked at claude.com/pricing on 29 September 2026. Check the page again before you quote a client.

For THL Partner Tier clients with ≥5 people on Claude, Team is the default. Enterprise for compliance-sensitive clients.

[Huxley: worth adding the Australian-billed equivalents when you have them in front of you. The prices above are in USD from claude.com/pricing, and AU pricing may differ.]

Subscription wind-down

Order of operations matters — cancel AFTER the memory extraction, not before.

1. Run the memory extraction prompt (above)
2. Save the output to the client's shared drive
3. Check ChatGPT billing: note next renewal date. Aim to cancel with ≥5 days buffer before renewal so they don't auto-renew into the migration month.
4. In ChatGPT: export all conversations (Settings → Data controls → Export data). Download the zip.
5. Set up Claude Pro/Team/Max — new billing account
6. Run the first week's Playbook (below) on Claude to make the switch sticky BEFORE cancelling ChatGPT
7. Cancel ChatGPT after the Playbook is complete and the user is confident

Do NOT cancel ChatGPT the day the migration starts. Always have a 7–14 day overlap.

The first-week playbook

To make the switch sticky, set up these in the client's Claude workspace in week 1:

Day 1 — Foundation

• Paste their Custom Instructions into Claude.ai user preferences
• Create the "About me" Project with the memory export pinned
• Install Claude desktop app
• (Technical users) Install Claude Code + authenticate

Day 2 — First three Projects

Re-create their three most-used Custom GPTs as Claude Projects. Each Project:

• Good name
• System prompt (from the extraction)
• 1–3 knowledge files if relevant
• One test run to confirm it works

Day 3 — Workflows

Pick two daily workflows and move them to Claude:

• Writing workflow (email drafts, content drafts) → a "Writing" Project with tone examples
• Decision workflow → a "Decisions" Project that defaults to extended thinking (see claude-4-7-extended-thinking-for-sme)

Day 4 — Integrations

Install the most useful MCP connectors / Claude.ai connectors:

• Gmail (if they use email heavily)
• Calendar (if they schedule in AI)
• Drive / SharePoint (depending on file storage)
• Slack (if team comms)
• Their CRM (HubSpot / Pipedrive / Salesforce) if a connector exists

Day 5 — Measure

• Check: did the user actually open Claude 3+ times in the last 48 hours? If not, something's off — diagnose before continuing.
• Ask them which ChatGPT workflow they missed most. Design around that gap specifically.

Day 7 — Retrospective

• 15-minute call. What's better, what's worse, what's unresolved.
• Decide whether to cancel ChatGPT now or keep the overlap longer.

Billing cadence check — avoid double-charging

This is the most common migration mistake:

• ChatGPT renews on the same date each month (the day they originally signed up)
• If they sign up to Claude on the 15th and ChatGPT renews on the 20th, they pay for both in the overlap month
• Solution: time the ChatGPT cancellation so it doesn't auto-renew during the migration. But don't cancel so early that the user has no fallback.

Practical rule: "cancel ChatGPT the day after a successful Day-7 retrospective."

Compliance / data residency

For compliance-sensitive clients (legal, medical, financial, NDIS), the migration conversation has an extra layer:

• Claude's enterprise zero-retention option + Amazon Bedrock / Google Vertex deployment options give options ChatGPT doesn't.
• For clients on ChatGPT Enterprise: the switch isn't a clear upgrade on data posture — it's a lateral move. Don't push.
• For clients on ChatGPT Plus / Business with AU data concerns: Claude Team or Enterprise is a genuine upgrade if configured with zero-retention.

See au-privacy-act-compliance-audit for the audit methodology.

When to NOT migrate

Signals to leave ChatGPT in place:

• The user's primary use is image generation via DALL-E → ChatGPT Plus still wins here
• The user relies on one specific GPT with no Claude Project equivalent AND the prompt isn't easily recreatable
• Voice mode is load-bearing to their workflow
• The user has deep habits around ChatGPT that would take weeks to rebuild AND the upside is marginal

In these cases, recommend hybrid: add Claude, keep ChatGPT Plus, let the user decide in 90 days whether to cancel.

Related skills

• claude-4-7-extended-thinking-for-sme — teach them extended thinking in week 2 for stickier adoption
• thl-partner-tier-proposal-generator — if this is part of a new engagement
• au-privacy-act-compliance-audit — for compliance-sensitive clients
• thl-google-workspace-migration — often bundled with this migration