Quick one before we start: AI Build Night is on tonight, 6pm London. Come along.
Last Saturday I did something most consultants avoid. I pointed a full audit at my own website and told it not to be polite.
Not a quick "run it through a free SEO checker" job. A proper twelve stage audit. Technical SEO, how AI search engines see the site, the hero, the messaging, the design, the offer, the analytics, the security, the lot. Each stage got its own specialist AI working through it, pulling real numbers out of my analytics, Search Console, Lighthouse and a headless browser crawling every page. Then a final panel reviewed the whole thing and wrote a build plan.
And it scored my site four out of ten.
I build systems for a living. I tell business owners their websites are leaking leads. And my own shopfront came back with a mark I'd be embarrassed to hand a client. So let's talk about what it found, because I reckon half of it is sitting on your site too.
How the thing actually works
Here's the setup, because the setup is the useful bit.
Viktor, my AI operator, ran it. One subagent per stage, each with a written brief and its own output file. Stage one did technical SEO. Stage two checked how ChatGPT and friends read the site. Stage four took apart the positioning. Stage eight went through the analytics. And so on, up to a stage that pulled it all into a ranked list of fixes with a build order.
But every finding had to come with evidence. A screenshot, a file, a query result, a line of code. No "your site feels a bit slow". If a subagent couldn't prove it, it didn't go on the list. Viktor then spot-checked the evidence before anything got published to me.
So the whole run took an afternoon. A human agency would quote you weeks for the same scope, and you'd get a PDF full of opinions.
What came out the other end: 151 separate items, each tagged by priority, effort and which goal it moves. Plus a twelve step build plan to take the site from four to a projected eight.
The bits that made me wince
My analytics were measuring nothing that matters
Google Analytics was installed. Page views, scroll, sessions, all ticking away. Lovely.
Zero key events configured. None. No booking completions, no checkout clicks, no signups. I had a dashboard telling me how many people turned up and absolutely nothing about how many did the thing I built the site for.
How many businesses do you know running exactly like this? Plenty, in my experience. You can't fix a funnel you can't see.
The AI search engines were looking at a blank page
This one surprised me. My site runs on a modern JavaScript setup. Looks great in a browser. But when the audit fetched the marketing pages the way a crawler without JavaScript does, it got an empty box. Literally an empty container where the page should be.
Blog posts were fine, they're pre-rendered as well. But the homepage and the offer pages? Blank to anything that doesn't run the full page.
A growing chunk of people now ask ChatGPT or Perplexity who to hire instead of typing into Google. If those tools can't read your page, you don't exist in the answer. The audit ran ten real buyer questions through search and I didn't come up for any of them.
Even better, I already had a bit of code that serves proper previews to Facebook and LinkedIn crawlers. But it didn't include the AI bots on its list. The fix was sitting right there.
My sitemap forgot my offers
The sitemap had over two hundred entries. Almost all blog posts. It left out the pages that actually make money: the audit, the community, the consulting page. The ones I most want found were the ones I never told Google about.
I was calling myself the thing I'm not
I'm not a coach. I've said it a hundred times. The audit found "coach" or "coaching" used about me on 27 of 51 crawled pages. Including the homepage. Including the button pinned to the bottom of every phone screen.
And that's the thing about a website that grows for three years. Every page made sense the day I wrote it. Nobody reads them all side by side. A machine does.
A dead page that told Google it was alive
So there was an old link to a capitalised booking URL that doesn't exist anymore. It showed a 404 page to humans but sent back a "200, all good" code to search engines. Google thinks it's a real page. Humans hit a wall. Small thing, but it's the sort of small thing that eats bookings for months, mate.
The homepage was the slowest page on the site
Look, mobile Lighthouse gave the homepage a performance score of 0.38. The main content took five and a half seconds to paint on a throttled phone. Main thread blocked for nearly three seconds. And the homepage hero was a fancy image slider that morphed between two photos of me, one of them in a suit. I don't wear suits.
What got fixed straight away
So this is where it stops being a report and starts being a system.
Before the audit even finished, three approved changes went live the same day. A founding member offer on the community page that turns itself off on 31 October without anyone remembering to take it down. An exit-intent prompt that points loyal readers at the community instead of a sales call. And the Build Night page now pushes people to the Monday session instead of a dead end.
But then the audit caught a near miss on those very changes. The promo code test in a headless browser came back "invalid" on the checkout. Screenshot looked like a spinner, which plenty of people would call a pass. The rule we run now: never trust a screenshot of a spinner. Load the real checkout link, read the actual total, and if the robot can't tell, a human tests it on their phone before it goes out.
So the other 151 items sit in a ledger. Each one has the URL, the change, the evidence and a risk note. The build plan groups them into twelve pull requests in a sensible order, so the scary server-side stuff doesn't land on the same day as a copy change.
And there's a re-audit booked for Boxing Day. Same prompt, same stages, so I can see whether four actually became eight or whether I just felt busy.
What you can steal from this
Look, you don't need my setup to get most of the value. Here's the short version.
Check your analytics for key events. Open GA4, go to key events, and see if booking, purchase and signup are in there. If the list is empty, your reports are decoration.
Fetch your homepage without JavaScript. Ask whoever built your site, or run curl on your own URL and look at what comes back. If it's an empty shell, AI search can't read you.
Open your sitemap. Is every page that makes you money in it? Takes two minutes.
Read five of your pages in one sitting. Out loud if you're brave. Do they describe the same business, with the same numbers? Mine didn't. My homepage and my pillar page quoted different proof numbers, which is exactly what a sceptical buyer spots.
Make every finding prove itself. If you get an audit done, human or AI, demand evidence per item. An opinion without a screenshot is a guess.
Here's the thing. I didn't run this because I thought the site was bad. I ran it because I'd stopped looking at it properly, and that's when things rot. The Autonomous Business idea isn't "set it and forget it". It's systems that check themselves so you don't have to remember to.
Four out of ten stung. But I'd rather hear it from my own machine on a Saturday than from a prospect who never books.
Tonight
Tonight's AI Build Night. 6pm London, live on Zoom, me building one of these systems from scratch while you watch. No slides, no rehearsed demo, and when something falls over on camera you get to see how I sort it out. That's usually the useful bit. Grab a seat.
Brewed by Steven, poured by Viktor
About Steven Tann: Steven helps business owners build systems that run themselves using AI. After 10+ years helping 7,000+ businesses and building his own autonomous operations, he's the bloke who actually does it, not just talks about it. Find out more at steventann.com.