Hey there,
By now you’ve probably used ChatGPT or Claude for one-off tasks — a prompt here, a draft there. But you re-explain your business every single time: who you are, what you sell, your tone, your rules.
The next level is building a custom assistant once, so it remembers all of that forever. It’s the difference between renting a helper and training one. Here’s how.
WHAT A CUSTOM ASSISTANT ACTUALLY IS
Think of it as a version of ChatGPT (or Claude) with your business baked in. You set it up once with your context — what you do, your voice, your policies, your common tasks — and from then on it answers like someone who already works for you.
No more pasting the same background every time. No more generic output. Just open it and go.
One heads up on plans: building and saving a custom GPT requires a paid ChatGPT plan. Plus, at $20/month, is the entry point — the free tier can’t create them (and now shows ads). If you use AI daily, this is the upgrade that pays for itself fastest.
HOW TO BUILD ONE (NO CODE, ABOUT 20 MINUTES)
In ChatGPT Plus, open the sidebar, find “My GPTs,” and click “Create a GPT.” It walks you through setup in plain conversation. Give it:
A clear job. “You are the customer-support assistant for [business]. You write replies in a warm, plain-English tone.”
Your knowledge. Paste in your FAQs, policies, and pricing, plus 2-3 examples of replies you’re proud of. You can also upload documents — your price list, service menu, or brand guide.
Your rules. “Never promise delivery dates. Always offer a next step. If you’re unsure, tell me to follow up personally.”
Name it, and it’s ready. Now anyone on your team can use it and get consistent, on-brand output.
A USEFUL FIRST BUILD
Don’t try to build ten assistants. Build one for the task you do most. The highest-ROI first assistants for small businesses:
A reply-writer trained on your tone and FAQs. A content assistant that knows your brand voice and audience. A proposal-and-quote drafter loaded with your services and pricing.
Pick the one that eats the most of your week.
MAKING IT GENUINELY GOOD (NOT JUST OKAY)
The difference between an assistant that’s “meh” and one your team relies on is the setup. A few things that level it up:
Feed it real examples. Three of your actual best emails teach it your voice better than any description. Show, don’t tell.
Give it your edge cases. Write down the weird situations — the awkward refund, the angry client, the after-hours request — and tell it exactly how you want each one handled.
Update it monthly. As your prices, policies, or offers change, spend 10 minutes refreshing its knowledge. A stale assistant gives stale answers.
Test it before the team uses it. Throw 5 real scenarios at it. Fix what’s off. Then roll it out.
THE TEAM MULTIPLIER
Here’s where it compounds. A good custom assistant turns your best practices into something your whole team can use. Your newest hire writes support replies like your most experienced one. Everyone quotes consistently. Your voice scales without you sitting in every loop.
That’s the real win — not saving yourself 10 minutes, but making everyone around you noticeably better at the repetitive stuff.
WHAT NOT TO PUT IN IT
Never load sensitive data you wouldn’t want leaving your control — customer financials, passwords, anything private. Treat it like a capable contractor, not a vault.
Never let it run fully unsupervised on anything that leaves your business. It drafts; a human still hits send on anything that matters.
YOUR ACTION PLAN THIS WEEK
Pick the one task that eats the most of your week. Spend 20 minutes building a custom assistant for it — job, knowledge, rules, and three real examples. Use it on five real tasks and tweak as you go. You’ll wonder how you worked without it.
Talk soon,
Dave
The AI Playbook
