There are three things that do most of the work of turning a generic AI into one that answers like your business: your prices, your policies and your tone. Get these three right and the AI stops guessing and starts sounding like you. Get them wrong, or hand them over carelessly, and it will confidently quote a made-up figure in a voice that is not yours. This guide is about doing it properly, with the specific traps called out, because the mistakes here are the ones that reach a customer.
If you are just getting going, our guide on building a second brain this week is the wider plan. This one zooms in on the three sections that matter most.
Prices: give rules, not just numbers
The instinct is to paste a price list and be done. That is a decent start, but a bare list is where the mistakes creep in, because the AI does not know when each number applies or when it should not quote at all.
So give it the rules as well as the figures. Instead of a lonely “bathrooms £X”, write how you actually price: “Bathrooms start around a figure. I only quote properly once I have seen two photos and know the size. Never give a firm price without that.” Now the AI knows the starting point, the condition, and the line it must not cross.
Two traps to head off here. First, the AI will fill any gap with a plausible number if you let it, so tell it explicitly what to do when it does not have enough to price: ask for the missing detail, do not invent. Second, prices go stale fast, and a wrong price given with total confidence is the worst kind of error. Date your pricing section and make updating it the first thing you do when you change what you charge. A brain quoting last year’s prices is a liability, not a help.
Policies: state them plainly and completely
Policies are the load-bearing facts of how you run: deposits, payment terms, lead times, what you take on and what you turn down, how you handle complaints and changes of mind. This is exactly the territory where a blank AI invents a policy you never had, because it does not know it does not know yours.
Write each one plainly and in full, the way you would brief a new member of staff. Do not assume the AI can infer the rest from a hint. If your deposit is a specific amount taken at a specific point, say so. If you never start work before it clears, say that too. The gaps are where it guesses.
The most important instruction in this whole section is one line: tell the AI to say when it does not know, rather than make something up. Something like “if a customer asks about a policy that is not written here, do not invent an answer, flag it for me to handle”. That single instruction is the difference between an AI that occasionally commits you to terms you never agreed and one that knows its own limits. For anything a customer sees, that safety catch matters more than any clever feature.
Tone: show, do not tell
Here is the one people get most wrong, and it is the easiest to fix. They try to describe their voice: “friendly but professional, warm but not too casual”. The AI reads that and produces exactly the bland, hedge-everything writing you were trying to avoid, because a description is not something it can copy.
Tone is taught by example. Find three or four messages you have genuinely sent to customers and paste them in with a simple instruction: “This is how I write. Match this tone in everything you draft.” A quote reply, a friendly chase for a deposit, a polite no to a job you cannot take. That is all it takes. The AI is an excellent mimic once it has real material to mimic, and it defaults to corporate mush only when you leave it guessing.
Pick examples that show range, not four versions of the same email. One warm, one firmer, one where you are delivering slightly awkward news. Now the AI can match your voice across the situations you actually face, not just the easy ones.
Test it the way a customer would
Once all three are in, do not assume it worked, check it. Open a fresh chat, paste your prices, policies and tone examples in, and give it a real task: “A customer wants a rough price for a kitchen, here is their message, draft my reply.” Read the result as if you were the customer receiving it.
Did it use your real starting price and hold off on a firm figure the way you told it to? Did it respect your terms? Did it sound like you? Where it slipped, the fix is almost always a gap in what you gave it, not a fault in the AI. Patch the source, run it again, and watch it tighten up. That loop is the actual skill, and it is why AI drafts and you check remains the rule even once the brain is good.
The habit that keeps it right
Prices, policies and tone all drift. You put your rates up, you change your deposit, you start signing off differently. Every one of those has to make it back into what the AI reads, or it will keep answering with the old version, confidently. We run a persistent context system of our own behind the tools we build, and the honest lesson from it is that keeping these three current is the real, ongoing job, not the one-off setup. The candid account is in how we run our own AI brain.
Where to take this next
Our flagship course Build an AI-ready business: give your AI a brain has hands-on lessons for handing over your prices, policies and voice safely, with examples for real trades.
To pick the first things to write down, grab the AI Readiness Checklist from our free resources shelf.
And if you would rather set this up alongside people who build these systems for a living, come to our AI Automation Masterclass in Manchester. Tickets are normally £20. This one’s free, a limited-time offer to launch the series.