Claude Projects for business files: keep client context without re-pasting every chat

Step 1: Create a Claude Project for one matter type

Per Anthropic’s create-and-manage guide:

1. Open claude.ai/projects (or Projects in the left sidebar).
2. Click + New Project.
3. Name it for the job, not a client — e.g. EP — Questionnaire follow-ups. Description is for humans; Claude does not use the name/description as instructions.
4. On Team or Enterprise, pick visibility (private vs org) as your admin allows.

Free accounts can use Projects (Anthropic: max five on Free). Paid plans unlock larger project knowledge via RAG when you approach context limits — confirm on your plan.

Step 2: Paste project instructions (copy this)

Click Set project instructions, paste, then Save instructions:

You draft for a US estate-planning law firm.
Voice: clear, calm, respectful; no jokes; no legal advice beyond what our templates already say.
Always:
- Leave [CLIENT NAME], [DEADLINE], [DOC LIST], [ATTORNEY] as placeholders if missing.
- Say when something is not in the project files (“Not in project knowledge”).
- Output a draft email or checklist only — I will send from Outlook/Gmail and update Clio myself.
Never:
- Invent deadlines, fees, or tax results.
- Ask me to paste SSNs, account numbers, or full wills into the chat.
- Claim you filed anything in Clio, Lawmatics, or WealthCounsel.
Matter type for this Project: missing estate-planning questionnaire follow-up.

Tune the last line if this Project is for engagement-letter variants or funding checklists instead.

Step 3: Upload scrubbed templates to project knowledge

On the Project page, use + to add files Claude should always see:

• Blank questionnaire checklist (no client data)
• Follow-up email skeleton
• Soft reminder vs firm reminder examples you already approve
• Short “firm voice” note (one page)

Rules:

• Strip every real client name before upload.
• Prefer PDF/DOCX/Markdown you maintain in Google Drive or Dropbox as the master; re-upload when the master changes.
• Do not upload executed wills, trusts, or ID scans.

Anthropic notes: context is not shared across Project chats unless it is in the knowledge base (or instructions). Paid plans can auto-enable RAG when knowledge gets large.

Step 4: Run the follow-up from inside the Project (not a blank chat)

Open the Project → new chat. Paste only what is unique to this matter (still no SSN):

Draft a questionnaire follow-up.
Facts I confirm:
- Soft or firm tone: [soft]
- Days since we sent the questionnaire: [10]
- Docs still missing: [asset list, beneficiary worksheet]
- Send from: [attorney email signature I will add]
Use project instructions + knowledge. If a detail is missing, keep a [PLACEHOLDER].
Max 140 words. Subject line + body.

Copy the draft into Outlook or Gmail. Log the send and next date in Clio (or Lawmatics). Claude drafts; your practice system stays the record.

Step 5: Compare once to ChatGPT Custom GPTs — then stop shopping

If someone asks “Claude Projects vs ChatGPT custom GPTs,” use this shop rule:

• Claude Project: instructions + files that stay with a workspace of chats for a matter type (per Anthropic Projects docs).
• Custom GPT: a packaged GPT with its own instructions/files on ChatGPT (see OpenAI’s GPT docs if you use that stack).

Pick the tool your firm already pays for. Do not run two parallel knowledge piles for the same questionnaire.

• Hour 0, create: One Project named for the matter type.
• Hour 0, instruct: Paste the firm instruction block; Save.
• Hour 1, knowledge: Upload scrubbed templates only.
• Ongoing, chat inside: New chat in the Project; send from Outlook; update Clio.

Example (illustrative): a four-attorney estate-planning firm

Example (illustrative). Partner Elena was tired of associates re-pasting the engagement excerpt. She created EP — Questionnaire follow-ups, saved project instructions, and uploaded three scrubbed templates from Google Drive. Associates open chats only inside that Project. Drafts go to Outlook; Clio tasks get the real dates. No client identifiers sit in the Project.

At a glance

• Problem: Re-pasting templates into blank Claude chats; inconsistent firm voice.
• What was set up: One Claude Project, instructions, three scrubbed files.
• Tools it lives in: Claude Projects, Clio, Outlook, Google Drive masters.
• What the owner still does: Approves wording; sends email; owns Clio dates.
• Payoff: not measured here — track “minutes to first draft” and “re-paste events per week” for two weeks.

Mistakes to avoid

• Blank chats for repeat work. Start inside the Project or you lose instructions/knowledge.
• Uploading real client files. Templates only; keep originals in Clio/Drive with normal access rules.
• Letting Claude invent legal deadlines. Placeholders; attorney confirms.
• Assuming chats share memory automatically. Anthropic: put shared material in knowledge or instructions.
• One mega-Project for every firm job. Split by matter type so instructions stay sharp.

If it still doesn't work

If drafts ignore your voice, re-open Set project instructions and tighten Never/Always lines, then start a new chat in the Project. If Claude “forgets” a template, confirm the file is in project knowledge (+), not only attached to one old chat. If Free’s five-Project cap blocks you, archive an old Project (Anthropic supports archive) or upgrade and re-check RAG capacity on paid plans. Fallback without Projects: paste one template section per chat and keep a firm prompt in a Docs snippet — slower, but safe.

Doing this inside the tools you already use

Deeplathe builds custom AI inside tools estate-planning teams already use — Clio, Lawmatics, WealthCounsel, Outlook, Gmail, Drive — so questionnaire follow-ups draft from your matter fields while you approve the send. Less than a hire, more output, for any business from a shoe store to a law firm. If you want that connected rather than maintaining Project files by hand, see /solutions. You stay in control of every client message.

FAQ

What are Claude Projects?

Workspaces with their own chats, a knowledge base, and optional project instructions, per Anthropic’s “What are projects?” help article. Available on Free (with a five-Project cap) and paid plans.

Claude Projects vs chats — what is the difference?

A normal chat starts empty. A Project chat inherits that Project’s instructions and knowledge files. Use Projects for repeat firm work; use a blank chat for one-off questions.

What should I put in Claude project instructions?

Tone, placeholders, never-rules (no SSNs, no invented fees), and the matter type. Save via Set project instructions so every chat in the Project follows them.

Can Free Claude use Projects?

Yes. Anthropic states Projects are available to all users; Free can create up to five. Paid plans add larger knowledge capacity with RAG when limits approach.

Claude Projects vs ChatGPT custom GPTs?

Both can hold instructions and files. Use the stack your firm already pays for; do not duplicate the same templates in two products.

Keep these

1. One Project per matter type — not per client.
2. Instructions saved; scrubbed templates in knowledge.
3. New chats start inside the Project.
4. You send mail and update Clio; Claude only drafts.

Sources

• What are projects? (Claude Help)
• How can I create and manage projects? (Claude Help)
• Claude.ai Projects

Related reading

Part of: The process that only lives in your head

• Build a personal AI knowledge base from your files
• RAG vs paste documents into chat
• Stable team system prompts
• Estate-planning questionnaire template