AI’s Website Advice Is Free. The Second Opinion Isn’t.
The short version
- AI’s website advice is a mix: some is right and worth doing, some is flat wrong, and some is right in general but wrong for your business.
- Telling them apart takes knowing your site, your market, and your goals. That’s our expertise, not an AI’s (yet).
- Plenty of clients ask us for an opinion on what AI suggested, and so far it has rarely been worthwhile.
- We much prefer creating over double-checking, so bring us the goal or the question, and we’ll figure out the best way to get you there.
- Yes, we’ll still double-check your AI’s to-do list, with a fair warning that the time spent may not be fruitful.
It usually starts with a forwarded list
Every so often, a client forwards us a list of website recommendations from an AI tool and asks, “Should we do this? Can you make these changes?”
Sometimes they went looking for advice. Other times they asked about one thing and the AI led them into a conversation of recommendations.
You have probably seen the viral screenshots. In 2024, Google’s AI search told people to keep the cheese on a pizza with a little glue, and to eat a small rock a day for minerals (Forbes).
It sounded authoritative, but of course it was nonsense.
That’s funny when it’s pizza (as long as no one ate glue!). It’s not funny when the confident, wrong answer is about your website, and using that answer could cost you business.
Some of the website advice an AI gives is right and worth doing. But two things can go wrong with the rest. Some can be flat wrong. Some can be right in general but wrong for your business.
Telling them apart takes knowing your site, your market, and your goals – our expertise!
There’s a tradeoff worth understanding. AI is supposed to save time. But a long list of AI recommendations can do the opposite, creating hours of work on our end that wouldn’t have existed without the AI in the picture.
We love that our clients are curious, and we’re always happy to answer a question. So please read what comes next as an explanation, not a complaint.
Here’s how AI recommendations can go wrong:
1. An everyday AI isn’t looking from all angles
When you ask a general AI tool, like ChatGPT, Gemini, Copilot, Grok, or Claude, how to improve your website, it’s working with far less than you might imagine.
When web access is turned on and used, it may read your public pages and some of the page code a browser receives. Sometimes it doesn’t even do that, and answers from general knowledge, from search results, or from whatever you pasted in.
What it normally can’t see is how your site is built. That includes:
- your analytics setup
- your CMS backend (the system you log into to edit the site)
- your hosting settings
- the code that runs on the server
- your private files and databases
- all of the ways that one feature depends on another
Most of the data one would want to know from these items sits behind a login, so it needs a deliberate (and potentially risky) connection for someone without experience managing it.
Even when it can read your code, unless you’ve spent a lot of time training your AI to do otherwise, it’s missing the context that decides whether a recommendation is good: what your business needs and what a given change would cost.
So it fills that gap with generalizations. It might spot something wrong on your site and have no idea that fixing it would take forty hours and bring you no more visibility, leads, or sales.
2. AI only understands how you work if you tell it
Even if AI knows a little bit about your business, we’ve found that most companies have not shared enough information to make AI website recommendations useful.
For most of our clients, working with AI is a bit like handing a contractor a single photo of your kitchen and asking for a remodel plan. You never told them how you cook, what you hate about the current layout, or what matters most to you.
So they design around the one photo. They might hand you a beautiful plan that puts the stove where you’d hate it, or takes out the one thing you love about the room. It looks right on paper. It isn’t built for the way you cook.
AI treats your website the same way. It can only plan around the little it was shown, so its recommendations can look reasonable and still miss what your business needs.
3. AI sounds confident, but that isn’t the same as correct
An AI can hand you an answer that sounds confident.
In 2025, Cursor’s AI support bot made up a usage policy, telling customers they could not use the product on more than one computer. No such policy existed, and Michael Truell, cofounder of Anysphere, the company behind Cursor, publicly corrected the response (The Register).
That same confidence is what makes AI recommendations expensive for the rest of us. They sound good, so they feel ready to act on.
But they tend to land in three predictable buckets, and all three can waste time.
Where AI recommendations can miss the mark
Here are the three, and why acting on each one costs you time.
1. Work you’ve already done.
AI tells you to claim your Google listing, gather reviews, add a page, or write an FAQ, without realizing you already have all of it. It’s just reciting general best practices.
2. Right advice for the wrong business.
Sometimes a recommendation is valid in one situation and still not right for yours. AI hands back broad marketing and SEO tactics without accounting for a small local business’s market, service area, budget, or staffing.
That can look like:
- aggressive link building (getting lots of other sites to link to yours)
- publishing a new article every day
- chasing the most competitive search terms in the industry
- building separate versions of your site for other countries
Some of those make sense in specific situations. Others can be ineffective, or even risky, without a clear reason to use them.
For a local business, they can cost you money and hours with little to show for it.
3. Changes that could break your site.
Sometimes the advice is specific and technical.
A business owner asked an AI how their website could be faster. It offered a confident fix: move part of the site’s code so it loads later, making the site seem quicker when a person first arrives.
In this case, the site certainly would have loaded faster, because the essential parts wouldn’t have loaded. A visitor wouldn’t have stayed long either. They would have quickly found that much of the site was broken and that they couldn’t make a purchase.
The same blindness can take a site down completely. We know a business that followed an AI’s step-by-step instructions to change their own domain settings, the behind-the-scenes settings that tell the internet where to find your website and your email.
Their website went down. Their email stopped arriving.
What looked like a quick do-it-yourself task became an emergency, plus hours on the phone with support working out what happened. AI couldn’t see how their domain and email were configured, and it never warned them what those settings were connected to.
Diligently checking the AI list takes time
Every one of these recommendations has to be checked before we can know whether it’s useful, already handled, pointless for your particular business and goals, or unsafe.
That checking, and then explaining what we found, takes time. That means you likely just unintentionally took some of your problem-solving marketing budget and put it into your “audit AI” budget.
It’s worth being clear about the tradeoff. When we vet and explain the AI’s list, you learn something. But learning why a recommendation is off, especially when it’s a technical reason that most clients have no interest in, is not the same as changing your site or marketing in a way that helps your bottom line.
That’s why we regularly report on the effects of our work
On a set schedule, we look at your real numbers, your traffic, how people are finding you, and the site’s technical health. Then we advise you on what we recommend doing next, and why.
Some of it keeps your site modern and running well. Some of it is SEO, the work that helps the right customers find you.
Because that advice starts from your own data and what we already know about your business, it’s ready to act on. An AI’s list has to be checked first.
We aren’t AI skeptics
We want to be clear about that. We use AI every day.
The difference is the kind of AI, and how it’s used.
We use paid AI tools and AI coding tools, but we make all the decisions and judgment calls. That’s an entirely different thing from an everyday AI chatbot answering questions from generic patterns about a site it spends little time analyzing.
Used the right way, on the right problem, AI makes us both faster. Used as a stack of unexamined recommendations to be checked one by one, it can slow all of us down.
Bring us the goal, and let us take it from there
Keep bringing us your questions and ideas. We’re always glad to talk them through.
Just keep the tradeoff in mind. The more of our time goes to vetting an AI’s list, the less goes to the changes that grow your business.
So when you can, lead with the goal or the problem, not the AI’s step-by-step. “We’d like more people to reach out through the site,” or “We’re launching a new service and want it to show up when people search,” gives us something we can act on.