Write specific, distinct project descriptions in Claude
AIXPORT's by-project layout works best when your Claude projects have clear, concrete descriptions. Claude's export file tells us each project's name, description, and instructions — and we use those to decide where each conversation belongs.
Good vs. broad
| Broad (weak signal) | Specific (strong signal) |
|---|---|
| "Business support, content creation, everything for the business" | "Q3 product launch marketing — investor decks, campaign briefs, launch op-eds" |
| "Client stuff" | "Acme Corp partnership — term sheets, capacity-lease negotiations, legal review" |
| "Notes" | "Engineering manufacturing plan — reactor design, production module specs" |
Avoid overlap between project names
If you have "Acme Project", "Acme Rev 1", and "Acme Deployments" as separate projects, make each description tell us what's unique about it. If the descriptions are similar, the classifier will guess — and guesses tend to favor the more general name (the catch-all effect).
Understand the creation-date constraint
Projects created recently can't contain older conversations. If you made a "Q2 2026 Planning" project last week, conversations from 2024 won't route to it even if the topics overlap — the conversations predate the project's existence.
This is usually what you want. If it isn't, the workaround is to either re-organize in Claude first, or use Flat layout and sort manually afterward.
Export soon after organizing
Claude's export file reflects the state of your workspace at export time. If you've just restructured projects, wait a few minutes for the changes to persist, then export.
Go to claude.ai/settings and click Export data. You'll receive a download link by email within a few minutes.
What AIXPORT does with your project metadata
For exports that preserve conversation-to-project membership (older Claude exports), we use the direct mapping. For exports that don't (most 2025+ Claude exports), we classify each conversation against your project catalog using Claude Haiku and emit a transparent _inference_report.csv showing what landed where and why.
The Discovery screen tells you which mode applies to your export:
| Badge | What it means |
|---|---|
Linked | Direct project membership preserved. No inference needed. |
Best effort | Inference will run. Prep steps above improve quality substantially. |
No projects | Your workspace has no Claude projects. Flat layout is the only option. |
When to use Flat layout instead
Flat layout puts every conversation in one transcripts/ folder, ordered chronologically. It skips the classification step entirely — no inference, no risk of misrouting, no per-project folders.
Open your AIXPORT ZIP and find memory-seed.md
This is the most important file. It contains everything ChatGPT needs to understand your project instantly.
Open it with any text editor (Notepad, TextEdit, VS Code) and copy all the text.
Open ChatGPT and start a new conversation
Go to chatgpt.com. Click New chat.
Paste the memory seed and send
Paste the full text of memory-seed.md into the message box and send it. Then say:
For permanent context — use ChatGPT Projects
To avoid pasting every time:
1. Click Projects in the sidebar → New Project
2. Go to Project Instructions
3. Paste memory-seed.md there
4. All conversations in this project will have your context automatically
project-brief.md and decision-log.md as files in the project.Use your prompt pack to continue the work
Open prompt-pack.txt — it has 5 prompts written specifically for your project. Copy any one and use it to pick up exactly where you left off.
Create a new Claude Project
Go to claude.ai. In the left sidebar, click Projects → New Project. Name it after your work.
Paste your memory seed into Project Instructions
In the project, click Project Instructions. Open memory-seed.md from your ZIP, copy all the text, and paste it there.
This gives every conversation in this project full context automatically.
Upload project files to the knowledge base
In the project, click Add content. Upload files from the projects/ folder in your ZIP — especially any documents that were attached to your original Claude projects.
Start a conversation and verify context
Open memory-seed.md and copy the text
Open your ZIP, find memory-seed.md, open it with any text editor, and copy all the text.
Go to Gemini Advanced
Open gemini.google.com. Make sure you're using Gemini Advanced (1.5 Pro or later) for best results with large context.
Start a new conversation and paste
Start a new conversation. Paste the memory seed as your first message, then say:
Create a Gem for permanent context
In Gemini Advanced, go to Gems and create a new Gem. Paste memory-seed.md into the system instructions. Use this Gem for all work on this project.
Find your project instructions
Open the projects/ folder in your ZIP. Find project-instructions.md. This contains the system prompts that were configured for your Claude project.
Create an equivalent in your target AI
ChatGPT: Create a new Project → Project Instructions → paste the instructions.
Claude: Create a new Project → Project Instructions → paste.
Gemini: Create a new Gem → paste the instructions as system prompt.
Upload your project files
Open the projects/[project-name]/ folder. Upload any documents there to your target AI's project knowledge base.
Test the setup
Files you'll use immediately
| File | What to do with it |
|---|---|
START-HERE.md | Quick-start guide for your specific transfer (same info as this page) |
memory-seed.md | Paste this first — gives your target AI full project context instantly |
project-brief.md | Read this — AI-written summary of everything you worked on |
prompt-pack.txt | 5 ready-to-use prompts written for your specific project |
Files to reference when needed
| File | What it contains |
|---|---|
decision-log.md | Every confirmed decision, numbered and grouped by topic |
open-questions.md | Unresolved items — blocking flagged in red, non-blocking in yellow |
user-profile.md | Your profile + Claude's accumulated memory about you |
message-analytics.md | Stats: conversation count, message volume, user vs AI ratio |
Reference archive (don't need to open these)
| File/Folder | What it contains |
|---|---|
transcripts/ | Full conversation history — 400 individual conversations as markdown |
projects/ | All your Claude project files, organized by project name |
manifest.json | Machine-readable metadata about the transfer |
conversation-index.csv | Searchable index of all conversations |
memory-seed.md and prompt-pack.txt. Everything else is there if you need to dig deeper.
Transfer another project
Run as many transfers as you need. Each one is a fresh continuity pack for a different project or export.
Start a new transfer →