How I Actually Use AI in My Business (Beyond Asking It Questions)

Most business owners I talk to use AI the way they use Google. They type a question, read the answer, and close the tab. That’s useful, and it’s also the smallest part of what it can do. If you’re trying to figure out how to use AI in small business operations, the bigger payoff comes from the work it can do on your behalf while you’re busy with something else.

I run a lot of my own business this way now, so this post is my actual setup: what AI does for me every week, how I keep it from doing anything I haven’t approved, and the parts I still do myself. I use Claude for almost all of it, but the levels apply to any AI tool you use.

Three levels: asking, briefing, delegating

I think about AI use in three levels, and most people never get past the first one.

Asking is the search-box level. You ask a question, and the answer is only as good as the question, because the AI knows nothing about your business. Every conversation starts from zero.

Briefing is when you give it real context before you ask: your offers, your prices, how you write, what you never say. Now the answers sound like your business instead of anyone’s.

Delegating is when AI does a whole job on a schedule, using your tools and your data, and hands you something finished to review. You stop prompting it. You approve or fix what it made.

Each level depends on the one before it. You can’t delegate to something you haven’t briefed, and a vague brief produces vague work on a schedule.

Level two: I wrote my business down once

The biggest change in the quality of what AI gives me came from writing things down once instead of re-explaining them every chat. These are called skills in Claude. They’re instruction files the AI loads whenever a task calls for them.

I have one for each of my brands’ voices. Mine says how I actually sound, plus a list of things I never do: no em dashes, no emojis, no hype openers, no “it’s not this, it’s that” sentences. Another skill has every product I sell, with its price, who it’s for, and what it pairs with, so a draft never invents a price or mentions a product I retired years ago.

There’s also a checklist that runs against every finished draft to catch writing that sounds generated. Every rule on it came from a real draft I had to fix. One of my favorites: a post was well written and completely wrong, because it described something my audience isn’t actually doing. Now every draft has to state, in one sentence, what the reader is doing right now before it gets written.

This is the same thing as writing SOPs for a new hire. If your process only exists in your head, AI can’t follow it any better than a person could.

Level three: the routines that run without me

Routines are AI tasks that run on a schedule. They read from my tools, do the work, and leave the result somewhere I’ll review it. Here’s what runs in my business right now:

• Monthly SEO check. On the 1st, a routine reads my Search Console data and my SiteGuru reports, writes up what moved, and rewrites the priority list the blog routine reads next. So the blog topics follow the data without me connecting the two.
• Quarterly claims review. Software companies change prices and features constantly. This one checks the facts on my site against the vendors’ own sites and tells me what’s out of date.
• Nightly time review. It goes through my time tracker, files every entry under the right client and project, writes a real description of what I did, drafts the timesheets I owe retainer clients, and tells me where my week actually went.
• Weekly email drafts. It turns the social posts I approved that week into email drafts in Drip, then sends me the links. I send them myself.

What these took over is the work I kept putting off, which was honestly a bigger problem for my business than not having enough help.

Connecting AI to my tools, with guardrails

Routines need access to your tools, so the AI is working from real data instead of whatever you remember to paste in. Mine connect to Airtable, WordPress, Drip, Search Console and my time tracker. The rule I set for every connection is the same: read freely, write carefully.

Read access is where I start, and for some tools it’s where it stays. The SEO routine only reads Search Console and SiteGuru. It can tell me what’s happening; it can’t change anything.

Write access is scoped to exactly what the job needs:

• In WordPress, AI can create drafts and update a draft. Status stays “draft.” Publishing is a click I make.
• In Drip, it can build a broadcast. It never sends one.
• In my time tracker, it can file, split and describe entries. It can’t delete anything, create a client, or touch a running timer.
• In Airtable, it adds every draft to my content calendar base with the status set to Review, so nothing moves forward until I’ve looked at it.

Airtable is where most of this lands. It’s the one place I can see everything the routines produced that week, what’s waiting on me, and what’s live.

The guardrails are written into each routine’s instructions, and where a tool lets me limit the connection itself, I do that too. Not every tool does, so every routine ends in a review step: the result waits somewhere I check, with a status that means nothing goes out until I move it.

What I still do myself

Plenty, on purpose:

• Approvals. Every blog post, email and social post gets my review before it goes anywhere. AI drafts; I decide.
• Client conversations. Discovery calls, mapping sessions, hard news, scope changes. A client hires me for my judgment, and they should get it from me.
• Final review of facts. If a draft quotes a price or a feature, it has to cite where it found it. If it can’t verify something, it leaves it out, and I’d rather have a shorter post than a wrong one.
• Deciding what’s worth building. AI is fast at doing a job, and I decide which jobs my business needs done.

Automation makes messy things messy faster, and AI is no exception. Every routine I run started as something I did by hand, then a skill I tested while watching, and only then a scheduled job.

Where to start in your own business

Don’t start with a routine. Start by writing down one thing you explain over and over: your voice, your services and prices, or your process for one recurring task. Use it as context for a few weeks and fix it every time the output misses.

Then pick one task you do weekly and could describe to a new hire, and give AI read-only access to the data it needs. Get the output right with you watching before you put it on a schedule.

If you want help deciding where AI fits in your tools, that’s part of my tech stack consulting, where we map how work moves through your business and decide what each tool should do. If you want the skills, connections and routines built and documented for your team, that’s custom systems work. Either way, the process comes first. AI connected to a messy system gives you confident answers built on bad information, which is why every system needs ongoing support once it’s live.

Frequently asked questions

What’s the best way to use AI in a small business?

Brief it on your business once, with your voice, offers and rules, then hand it one recurring task with read-only access to the data it needs. Review everything it produces before it reaches a client.

Is it safe to connect AI to business tools like Airtable or WordPress?

It can be, with limits. Give each connection only the access its job requires, start with read access, and keep publishing, sending and deleting as actions a person takes.

Will AI replace my assistant or my team?

Not in my business. It took over work I kept putting off, like reporting, first drafts and timesheet cleanup. Judgment calls and client relationships still need a person.

Which AI tool do you use?

I use Claude for skills, connectors and scheduled routines. ChatGPT has similar features. The tool matters less than writing your business down and setting clear limits on what it can change.