AI audit vs. strategist audit · A roofing company
We ran a roofing client's website through a ChatGPT site audit. A lot of it sounds right, and some of it is right. The longer the track runs, the further off course it gets, and the misses are the ones that matter most.
Shared ground
These three recommendations are real, and they line up with what we would do. This is why AI audits are convincing: the first few stops are right.
Rewriting the city pages is core to our plan, starting with the suburbs that bring in the most revenue.
Real jobs in real neighborhoods are proof a roofer works there. Good call.
Pointing links at the pages that make money is in our strategy too.
Where it drifts
These are the hard ones to catch. The recommendation is in the right area, but acting on it as written would waste hours or break something that works.
Localize it, or AI eats it.
A generic cost question gets answered right in the AI response, so nobody clicks. Tie it to the client's city or a priority suburb and we become the local source AI is far more likely to cite.
Those numbers are CallRail, on purpose.
Different numbers let us tell which calls came from organic search. Before we set it up, the client couldn't separate organic calls from paid ones.
People don't memorize a roofer's number. They tap it. The tracking is worth far more than any confusion it could cause.
What it never saw
Nothing in the audit was wrong here. It was just blank. These are the moves most likely to grow leads for this client.
The site desperately needs authority.
The companies holding the local map pack have much stronger sites. No amount of on-page work closes this gap by itself.
One pin can't cover the whole metro.
The only Google Business Profile that ranks sits on one edge of the metro. Its map rankings fade fast with distance, and the money is in suburbs on the other side of town. We need more pins in those priority areas, each with its own profile.
The site was riddled with AI writing, and a lot of it is still there.
Most pages were written with AI before we started. We've rewritten a lot of it, but plenty remains. Generic AI copy drains trust with buyers and holds back performance.
AI audits never flag this. They read AI-written copy as normal copy, so the tool is blind to one of the site's biggest problems.
Why it missed
ChatGPT only saw the pages. Every miss above traces back to something it had no way to know.
The owners told us which suburbs are where the money is.
Who actually holds the map pack, and how strong their sites are.
How far their one pin really reaches.
Why the site has more than one phone number.
Which content AI answers on its own, and which it still has to cite.
Where the copy came from. AI doesn't flag writing that sounds like itself.
The ChatGPT column paraphrases its audit of the client's site. Client details are anonymized.
Spotting the drift
The differences are hard to see, and we have to catch them anyway. Run each recommendation through these before it goes into a strategy.
Check for gaps, not just errors. Links, map presence, and AI-written copy are the usual blind spots.
Priority areas, tracking setup, and past work all live with us, not on the site.
If the idea works for a roofer in any city, AI can answer it without us.
CallRail numbers, redirects, and noindexed pages often look like mistakes from the outside.