Building a Client-Prep Custom GPT

So far you've been writing the same context into prompts over and over — your voice, your compliance constraints, your meeting-prep format. A Custom GPT (in ChatGPT) — or a Claude Project, or a Gemini Gem — lets you bake all of that in once, so every conversation starts pre-configured. This lesson walks through building one for client-meeting prep, the highest-frequency advisor task.

What You'll Learn

• What a Custom GPT (and the Claude/Gemini equivalents) is and when it's worth building
• Writing clear instructions that capture your process, voice, and guardrails
• What knowledge files to attach — and what never to
• How to test, refine, and use it safely

What a Custom GPT Actually Is

A Custom GPT is a saved configuration of ChatGPT: a name, a description, a set of standing instructions, optional uploaded reference files ("knowledge"), and a few settings. Once built, you (and, with a team/enterprise plan, your colleagues) just open it and start working — no re-explaining context. Claude offers Projects (a workspace with persistent instructions and documents); Gemini offers Gems (custom assistants). The concepts are the same; pick whichever your firm has approved. Important: building a Custom GPT requires a paid ChatGPT plan, and you should only attach firm-related material if your firm's data and AI policies allow it.

When is it worth building? When you do a task often, in a consistent format, with consistent constraints. Client-meeting prep is the poster child. So are: drafting client communications in your voice, processing meeting notes, writing newsletter sections, and explaining concepts to clients. Build a small set of these, not fifty.

Designing the "Client-Prep Assistant"

Think of the instructions as the briefing you'd give a sharp new associate on day one. Cover role, process, format, voice, and hard rules. Here's a template you can adapt:

Name: Client-Prep Assistant

Description: Helps me prepare for client review meetings — briefs, agendas, likely questions, talking points.

Instructions: You are an experienced paraplanner supporting a financial advisor. Your job is to help prepare for client review meetings. When I give you a client situation (always anonymized — no names, SSNs, or account numbers) and my rough notes, produce, clearly separated:

1. A 3–4 sentence "where things stand" summary.
2. The 4 most important topics to cover, in priority order, each with one line on why it matters.
3. 4–6 questions the client is likely to ask, each with a 2-sentence starting point for my answer grounded in sound planning principles.
4. Talking points for any sensitive topic I flag.
5. A short list of things I should double-check or bring (especially any numbers I should verify against the actual statement). If I ask, also produce a client-facing meeting agenda and a one-page printable meeting sheet.

Voice: warm, plain-English, calm, professional, no jargon. When I ask for client-facing text, write the way I would: [paste 2–3 sentences of your own writing].

Hard rules — never break these: Do not predict specific future returns. Do not use the words "guarantee" or "guaranteed." Do not give individualized investment, tax, or legal advice as if it were final — frame options and note that I decide. Treat every dollar figure you produce as a placeholder for me to verify. If I appear to have included a client identifier, point it out and ask me to remove it. Remind me at the end of any client-facing draft that it needs my review (and firm review where applicable).

Start each session by asking me for: the client situation (anonymized), what's changed since last time, my rough notes, and any sensitive topics to handle.

That's it. Save it and you have a teammate that already knows how you work.

Knowledge Files: Yes and No

You can attach reference documents the GPT will draw on. Good candidates:

• Your voice/style guide (a page on tone, words you use and avoid, examples).
• Your standard meeting agenda template and review checklist.
• Generic, non-confidential explainers you've refined (your standard "what is a Roth conversion" etc.).
• A compliance do's-and-don'ts cheat sheet for client communications (the constraints, in your firm's words).

Never attach:

• Anything with client PII — client lists, account data, plans with real names/numbers. This is the privacy rule again; a knowledge file is still data on a server.
• Confidential firm material your data policy doesn't allow in the tool.
• Anything you haven't gotten the okay to use this way.

When in doubt, generalize it first or leave it out.

Test It Like You'd Test a New Hire

Before you rely on it, run a few realistic (anonymized) cases through it:

• Give it a typical retired-couple review and check the brief is genuinely useful and the format is right.
• Throw in a sensitive topic ("need to suggest reducing withdrawals") and confirm the talking points are honest and kind.
• Deliberately include a fake "client name" and confirm it catches it and asks you to remove it.
• Ask for a client email and confirm it sounds like you and ends with the review reminder.
• Try to get it to predict returns or say "guaranteed" and confirm it won't.

Adjust the instructions wherever it falls short. Treat the instruction text as living — tighten it over the first few weeks of real use.

Using It Safely

• It's still ChatGPT under the hood: the data and compliance rules from Module 1 and 3 all apply. Anonymize. Use your firm's approved/enterprise version. Don't paste PII.
• Its output is still a draft. Every brief, email, and agenda gets your review; client-facing material gets firm review per policy.
• If you share it with colleagues, make sure they're trained on the same rules — a shared GPT spreads good practice and bad habits.
• Revisit it periodically: did a constraint change? Did your firm update its AI policy? Did you find a better format? Update the instructions.

Done right, a Custom GPT turns "open a blank chat and re-explain everything" into "open the assistant and go" — and it bakes your guardrails into the tool so they're harder to forget.

Key Takeaways

• A Custom GPT (or Claude Project / Gemini Gem) saves your role, process, format, voice, and hard rules so every session starts pre-configured — worth building for high-frequency, consistent tasks like client-meeting prep.
• Write the instructions like a day-one briefing for a sharp associate: the deliverables and format, your voice (with samples), and unbreakable rules (no return predictions, no "guaranteed," treat figures as placeholders, catch client identifiers, remind you to review).
• Attach only generic, non-confidential knowledge files — voice guide, agenda template, compliance cheat sheet — and never anything with client PII or restricted firm material.
• Test it with realistic anonymized cases (including a planted identifier and a "make it guarantee returns" attempt), refine the instructions over time, and remember the output is still a draft subject to your review and firm review.