How to Use AI in Your Small Business: A Practical Playbook for Owners Who Don't Code

Last spring a shop owner sat down in one of our workshops with a laptop and a confession.

She'd been "using AI" for a little over a year. ChatGPT, mostly. She opened a folder to show us: a hundred and forty documents. Instagram captions. Three versions of an About page. A newsletter series. Every one of them a draft. Almost none of them published.

"It writes fast," she said. "I just never know if it's right."

That folder is the most common thing we see when a small business tries AI on its own. Output without a system. A very fast way to generate things you don't trust enough to ship.

This is the playbook we teach instead — what to build first, the five jobs worth handing over, the three you shouldn't, and how to know whether any of it is working. You don't need to code. You need to know what you're doing before the tool does it for you.

Start With the Process, Not the Tool

Here's the rule everything else in this post depends on.

AI amplifies what you give it. Give it vague direction and you get a polished version of vague. Give it your actual voice, your actual offer, your actual customer, and it starts producing things you'd put your name on. The model isn't the variable. The input is.

We wrote about this in You Don't Own It Until You Can Teach It, and the short version is: if you can't teach a task to a new hire, you can't teach it to a model either. Same gap. Same fix.

So before you open a single chat window, write three documents. Plain text. An hour each, maybe less. They live in a folder you own, and every workflow below starts by handing them over.

The voice doc. How you actually talk. Five or six real sentences you've said to customers, verbatim. Words you use. Words you'd never use — for us, that list includes "elevate" and "unlock" and anything that ends in an exclamation point. A short paragraph on what you sound like on a good day.

The offer doc. What you sell, what it costs, what's included, what isn't. Who it's for and who it's wrong for. The three questions people always ask before they buy. The thing you wish everyone understood about your work before they called.

The customer doc. Two or three real customers described honestly — not a "persona," a person. What they were worried about when they found you. What made them say yes. The words they used, not the words you'd use.

It sounds too simple to matter. Then you watch an owner paste those three files into a conversation and get back a caption that sounds like her for the first time in a year.

The model isn't the variable. The input is.

The First Five Jobs

These are in order. The payoff-to-risk ratio gets worse as you go down the list, and most businesses should stop around number three for the first few months.

1. Turn one real thing into a month of content

What it is. A real thing happened this week — a job you finished, a shoot, a review, a question someone asked at the counter. That one thing becomes fifteen or twenty pieces of content across the platforms you actually use, each written for the platform instead of copied across it.

Set it up in an afternoon. Start a conversation with your three docs. Paste in the raw material: the review, the photos with a sentence about each, the transcript of you talking through the job for four minutes into your phone. Ask for a plan first, not the posts — what angles are in this material, which platform each belongs on, what order to publish. Then generate the pieces one platform at a time.

The prompt shape. Context first, then the material, then the ask, then the constraints. Something like: here is how we talk, here is what we sell, here is what happened this week. Give me a two-week plan for Instagram and email that uses only what's in this material, and don't invent details that aren't here. That last clause matters more than anything else in the prompt.

The guardrail. Nothing invented. Every claim has to trace back to the raw material. The model will happily write "our customers love the new spring line" if you let it. You didn't give it a spring line. Read for that.

This is the same shape as the Pinto Ranch work — one day on location, a month of content, 300,000-plus organic views in three weeks. The raw material was real, so everything built on it was real. That's the whole trick, and no tool changes it.

2. Answer the same questions once

What it is. Every business answers the same eight questions forty times a month. Hours, pricing, what's included, how to book, can you do rush. AI turns the answers you've already given into an FAQ page, an intake form that routes people correctly, and reply drafts you can send in thirty seconds instead of ten minutes.

Set it up in an afternoon. Export a month of inbound messages — email, DMs, contact form — with names removed. Paste them in with your offer doc. Ask the model to cluster the questions, count them, and draft a plain-English answer to each in your voice. Fix the ones that are wrong. Put the result on your website as an FAQ and in a doc you reply from.

The prompt shape. Here are sixty real messages we received last month. Group them by what they're really asking. Tell me the eight most common. Draft an answer to each one that sounds like us and uses only what's in the offer document.

The guardrail. You approve every answer before it goes near a customer, and you re-read the set every quarter. When we consolidated Mortenson Ranch onto one site with a real booking system, calls about basic questions dropped by more than 70%. That wasn't a chatbot. It was the answers, written once, put where people could find them. AI just makes the writing-once part take an afternoon.

3. Write down your SOPs by talking them out

What it is. The most valuable use of AI for a small business, and the least glamorous. You have processes in your head that nobody else can run. Talking them out and having the model turn the recording into a step-by-step document gets them out of your head — the only way to ever hire, delegate, or take a week off.

Set it up in an afternoon. Pick one process. Open a voice memo. Talk through it like you're training a new hire standing next to you — every step, every "here's the thing people get wrong." Transcribe it; most phones do this now. Ask for a numbered procedure with a "why" line under each step and a list of the questions you didn't answer.

The prompt shape. This is a transcript of me explaining how we do this. Turn it into a procedure someone could follow without me. Keep my reasoning. Flag every place where I said "usually" or "it depends" so I can fill in the rule.

The guardrail. The flags are the point. Where the model finds a gap, there's a decision you've been making by feel. Write the rule. That's the work.

4. Research and competitive reading

What it is. Reading what your competitors publish and what your market says about itself, and getting a summary you can act on instead of forty open tabs.

Set it up in an afternoon. Give the model the pages you want read: competitor service pages, the local chamber, industry press. Ask for what each one claims, in their words, and what's missing. We did exactly this reading for our comparison of Durango marketing agencies — it's what made it possible to describe every firm only by what they publish about themselves.

The prompt shape. Read these six pages. For each, tell me what they say they offer, who they say they serve, and whether they publish pricing. Quote, don't paraphrase, when it matters. Then tell me what none of them say.

The guardrail. Verify every number. Models still produce confident, specific figures that don't exist — a founding year, a percentage, a population count. Anything with a digit in it gets checked against the source before it goes into a decision, a proposal, or a post. It's the rule our own editorial system runs on, and the one owners skip most often.

5. Edit your own website with AI coding tools

What it is. Tools like Claude Code, v0, and Lovable now change real website code from plain English. For an owner who was handed a site they own, that's the difference between calling someone to change a price and doing it yourself in five minutes.

Set it up in an afternoon. Only if you own the site — the domain, the code, the hosting, in your accounts. If you don't, start with Own Your Website and come back. Then start small: the hours on the contact page, a photo swap, the copy on one section. Describe what you want, read what it did, publish to a preview first, and only then to the live site.

The prompt shape. On the About page, replace the second paragraph with this text. Change nothing else. Tell me exactly which file you touched. Narrow scope, an explicit "change nothing else," and a demand for a report of what changed.

The guardrail. This is where the risk climbs. Copy edits and image swaps are safe. Anything touching checkout, forms, payments, or customer data is not a five-minute job, no matter how confident the tool sounds. When we rebuilt LePlatt's Pond, part of the handoff was sitting down with Krii and walking through exactly this — how to make her own changes with AI tools, and where the line is. She runs the site herself now, and she knows which changes to ask about first. If you've built something with these tools and aren't sure it's shippable, that's what the Vibe Coding Consultation is for: a senior review before it goes live.

The Three Things Not to Hand to AI

We use these tools every day. There are three places we keep them out.

Customer trust moments. The first reply to a new inquiry. The apology when something goes wrong. The follow-up after a big job. These are the moments people decide whether you're real. A draft is fine. A sent message that sounds like a template costs you something you can't measure until it's gone.

Anything with money or legal weight. Contracts, quotes on complex work, tax questions. The model will produce a confident document. Confidence isn't the same as correct, and you're the one who signs it.

Your point of view. What you think about your industry. What you'd tell a customer not to buy. The reason you started. AI can help you say it more clearly. It cannot supply it, and the businesses that outsource this part end up sounding like every other business that did. In a market as small as Durango — or Santa Fe, or any town where people know who you are — sounding like everyone else is the expensive mistake. We wrote about why in What Makes a Good Marketer. The point of view is the product.

How to Tell If It's Working

"We use AI now" is not a result. Measure four things, monthly, on one page.

Hours back. Track a normal week before you start — how long content takes, how long inbound replies take. Track it again after thirty days. If the number didn't move, the workflow isn't real yet.

Response time. How fast a new inquiry gets a real answer from a real person. This moves revenue more directly than almost anything else you can change.

Published output. Not drafts. Things that went live. The folder of a hundred and forty documents counts as zero.

Revenue-linked actions. Bookings, quote requests, calls, direction requests on your Google Business Profile. If content is going out and none of these move in a quarter, the content is the problem, not the tool.

Keep the page. It's the same discipline we'd ask of any marketing channel, and the fastest way to find out which of the five jobs earns its keep.

Own the System

Every build we ship ends with the same handoff: the domain, the code, the hosting, and the training, in the client's name. AI workflows deserve the same treatment. Most businesses aren't getting it.

If an agency "sets up AI" for you and the prompts live in their account, the documents on their drive, and the system only works while you're paying them, you haven't adopted AI. You've rented a black box.

The standard we hold our own AI work to, and the one to hold anyone else to:

Your docs, your accounts. The voice, offer, and customer documents are yours. The chat accounts are yours. Any automation runs on credentials you control.

A written runbook. Every workflow gets one page: what it does, what goes in, what comes out, what to check before anything ships, what to do when it breaks. If it isn't written down, it isn't a system.

You can run it without us. Hand the runbook to someone new and see if they can do it. If they can't, the runbook isn't done.

The systems that run our own studio — the content pipeline, the agents, the retrieval that keeps everything in our voice — were built in-house and documented well enough to hand to a new person tomorrow. That's the bar. A system nobody else can run isn't an asset. It's a dependency with better branding.

A system nobody else can run isn't an asset. It's a dependency with better branding.

The Five Jobs, Side by Side

Frequently Asked Questions

How can a small business use AI?

Start with three documents — how you talk, what you sell, who buys — then use AI for the jobs where you already know what good looks like: turning real material into content, answering common questions once, writing down your processes, reading your market, and making small edits to a site you own. Start with the process, not the tool.

What is the best AI tool for a small business?

The one you'll actually use, with your own documents loaded in. As of this writing, Claude, ChatGPT, and Gemini all handle the writing and reading jobs well; Claude Code, v0, and Lovable handle site changes from plain English. The difference between them matters far less than the difference between an owner with a voice doc and one without.

Is AI safe to use with customer data?

Carefully. Don't paste customer names, payment details, or anything you'd be uncomfortable seeing leaked into a consumer chat tool. Strip names from inbound messages before you cluster them. For anything touching real customer records, use business accounts with clear data terms, and keep a human approving every message that goes out.

Can AI write my website?

It can write a draft, and it can now change real code on a site you own. It can't supply your point of view, your photography, or your judgment about what belongs on the page. The sites that work are built from real material, and AI is the fastest way to shape that material — not a substitute for having it.

Do I need to hire someone to set up AI for my business?

Not for the first three jobs on this list; those are an afternoon each with the documents in hand. It helps to bring someone in when you want the workflows connected — intake that routes itself, agents that read your own files — or when you've built something and want to know if it's safe to ship. Either way, insist on the handoff: your accounts, a written runbook, and a system you can run without them.

Not ready to reach out yet? Start with the Brand World Blueprint — a $4, 14-page guide that scores your brand across five pillars. The clearest picture you'll get of where you actually stand.