How I use an AI agent to run my business website

I run a small business. I don't have time to be my own web team too. So I handed most of that job to an AI agent, and I kept the part that matters: saying yes or no.

This isn't a pitch for any tool. I use Grok Bot, but the setup is what's worth copying, not the brand. Here's what the agent actually does on gameplanfinancials.com, what it doesn't do, and what I've seen so far.

The rule: the agent proposes, I approve

My site lives in a code repository. The agent can read it, run audits, and write changes, but it doesn't push anything live on its own. Every change shows up as a pull request: a written description of what changed and why, plus the change itself. I read it, ask questions if something looks off, and approve it or don't. Only then does it merge and deploy.

That one rule is why I'm comfortable with this. I'm not handing over the keys. I'm getting a careful assistant that does the legwork and waits for my sign-off. If I'm busy, the work waits. Nothing breaks because I didn't answer.

What it does: audits

Most of the work starts with an audit. The agent looks at the site the way a search engine and a first-time visitor would: page titles, the short descriptions that show up in search results, how fast pages load, whether the sitemap lists every page, and whether the structured data that helps search engines read the page is there and correct.

It also reads the analytics. It pulls visitor numbers from Google Analytics and checks which pages get seen and which get skipped. An audit isn't a list of best practices copied from a blog. It's a short list of specific fixes for my site, ranked by how much they're likely to matter.

What it does: fixes, shipped as pull requests

Here are three real examples from my site's history. Each one was a pull request I reviewed and approved.

• #29, an SEO batch (Oct 30). Rewrote page titles so more people who see the site in search results actually click, added FAQ and blog structured data, deferred the Google Analytics script so pages load faster, cached the site's fonts, and added internal links between related posts.

• #31, lead tracking (Oct 30). When someone submits the inquiry form successfully, the site now records a lead event in Google Analytics. Before that, I could see visitors but not which visits turned into a conversation. Now I can.

• #34, audit fixes (Nov 2). Reordered the services on the home page, switched to a sitemap generated automatically from the list of posts so new pages don't get missed, delayed Analytics loading by about three seconds so it doesn't slow the first view, shortened some post titles, and added structured data to the checklist page.

None of these are glamorous. That's the point. These are the small, tedious fixes a business owner knows they should get to and never does. The agent gets to them, writes up what it did, and I spend a few minutes reviewing instead of a weekend learning how sitemaps work.

What it does: reports back

After every change, the agent tells me in plain language what it shipped and what it checked on the live site afterward. When I ask how the site is doing, it pulls the numbers from Google Analytics instead of guessing. I want to know where things stand and what decision I need to make, not read a dashboard.

What I've seen so far

Here's the honest number. In Google Analytics, total users for the seven days pulled on October 30 came to 19 users and 25 sessions. The seven days before that had 6 users and 17 sessions.

That's one week, and it's a small site. It isn't a trend, and I wouldn't build a forecast on it. But it's the right direction, and I can trace every change behind it to a pull request I read.

What it doesn't do

• It doesn't publish or deploy without my approval.

• It doesn't make up facts. When a post makes a claim, the claim needs a source, or it doesn't run.

• It doesn't replace judgment about my business. It can tell me a page title is too long. It can't tell me what my business stands for. That part is still my job.

If you want to try this

You don't need to be technical, but you do need a few things in place:

• Your site in a repository with a review step, so changes are proposed, not pushed.

• Analytics installed, so the agent has real numbers to work from.

• A habit of actually reading what it sends you. The review is the safety net. Skip it and you've just automated your mistakes.

Start small. Ask for one audit and one fix. Read the pull request. If it's clear, careful, and right, give it the next one. That's how I started, and it's how I'd tell any owner to start.