Consultants use AI most on the work around the client, not the client's secrets: turning a kickoff call into a recap and proposal outline, scanning a prospect's market with cited sources, drafting decks and status reports, and pulling themes out of interview notes. They keep client data safe by using a business plan that does not train on their content, checking what each client's contract allows, and taking out names and figures the task does not need, a routine you can try in Kickoff Call to Proposal Outline.
The confidentiality question is the first one a client will ask. Many master services agreements and NDAs limit sharing client information with third parties, and some clients now add AI clauses of their own. Read the contract before you paste anything, and when in doubt, ask the client in writing.
Which client data can go where?
A simple three-level rule keeps a small firm consistent:
- Public: any approved tool. Competitors' websites, published prices, industry reports, the client's own public pages. Research tools can search these freely.
- Internal but not sensitive: business plan only. Your proposal template, meeting recaps with names removed, your own analysis frameworks. Use a Team or Enterprise plan of ChatGPT or Claude, or Copilot under your firm's Microsoft 365, not a personal account.
- Confidential: only with client permission. Financials, customer lists, employee data, deal terms, anything marked confidential. Either the client approves the tool in writing, or it stays out. Replace it with placeholders ("Client A", "a regional retailer") when you need help with structure.
Write this on one page and give it to every consultant and subcontractor.
What are consultants using AI for day to day?
- Kickoff and discovery calls. A transcript becomes the scope, budget and timeline the client actually stated, an action list and a recap email.
- Meeting notes. A bot-free notepad like Granola transcribes on your computer and writes notes around your own jottings. You still tell the client you are taking AI notes.
- Prospect and market research. Claude's Research or ChatGPT's Deep Research reads public sources and writes a report with a citation on every fact.
- Deliverable drafts. Outlines, slide structure, executive summaries, from your own analysis.
- Interview synthesis. Themes and quotes from stakeholder interviews, with names removed.
See the consultant start path for a lesson sequence, and AI automation for consultants for workflows across a practice.
What does this look like in a real lesson?
In Kickoff Call to Proposal Outline a small marketing agency has just finished a kickoff call with a physical therapy clinic. You copy the transcript from Google Meet into ChatGPT under a request for a summary, the decisions on scope, timeline and budget, an action items table and open questions, with "Not stated" instead of guesses. You fill in two missing dates (an audit due Thursday, the full proposal by October 30), then ask for a recap email with a proposal outline. The monthly ad spend stays an open question, because nobody decided it.
Before the pitch, Research: a Cited Scan of a Prospect's Market has Claude's Research read competing clinics' public pages and write a cited report. You open one source and catch a rate card from 2024 that the report did not flag. For the call notes themselves, Granola: client call notes has you check the budget line against the transcript before sharing. And Use AI at Work Without Leaking Client Data covers the settings and the redaction habit.
What mistakes cost consultants the most?
- Pasting a client's financials into a personal account. Even with training off, it may break the contract.
- Sending AI research unchecked. One outdated price in front of a client undoes the whole deck.
- Recording without telling the client. Bot-free tools are invisible, so say it out loud.
- Letting AI set the price. It can structure a proposal. You write the fees.
- Forgetting subcontractors. Your rules apply to everyone touching the client's work.
Key takeaways
- Use AI on recaps, research, drafts and synthesis.
- Sort data into public, internal and confidential, and treat each differently.
- Use business plans, check each client contract, and take out names.
- Check every source and number before a client sees it.
Frequently asked questions
- Do I have to tell clients I use AI?
Check your contract and the client's policies. Many firms now say so in the engagement letter, which avoids awkward questions later.
- Is a business plan of ChatGPT or Claude enough to protect client data?
It removes training on your content by default, which helps, but it does not override the client's contract. Permission still comes first.
- Can I build a project for each client?
Yes, and it keeps instructions and files separate. Put only the documents the client has cleared for AI use into it.