AI website builders are impressive. Type a few sentences about your business into ChatGPT, Claude, Lovable or the WordPress.com AI builder, and a few minutes later you’ve got something that looks like a real website, with colours, headings, photos and a contact page. I use these tools myself, and the first time I saw one work I had the same reaction most people do: that was fast.
After 25 years of designing websites, though, I’d say what AI produces is roughly the 10% of a website you can see. The other 90% is the thinking that decides whether it works for your business. This post walks through that 90%, then covers some practical ways to get better results from AI if you decide to build your site with it yourself.
What AI does well
AI is very good at getting you from a blank page to something you can react to. It’s quick at layouts and handy for exploring ideas before you commit to anything. If you’ve used it to work out what you like and don’t like, that’s time well spent.
It’s also a brilliant copyist. Point it at something that already exists and it will imitate it convincingly. Where it struggles is coming up with something original that’s right for you. Ask it to create a website from nothing and it gives you the average of every website it has seen, and the trouble starts when people treat that average starting point as the finished product.
Why AI websites all look a bit the same
Once you know what to look for, you’ll start spotting AI-made websites everywhere: a soft gradient glow behind a centred headline, two buttons underneath, a perfectly even row of three icons, text that’s a touch too small, and copy about “elevating your experience” and “seamless solutions”. Everything is technically in the right place, yet nothing looks like anyone made a decision. I’ve heard it described as designed but not thought through, which I think is spot on.
That sameness is a business risk, too. If your website came from a prompt, your competitor can get something very similar from a prompt. Ask yourself whether your customers would recognise your website as yours if someone copied it tomorrow. If they wouldn’t, the site isn’t doing much to represent your business.
The 90%: what designers think about that AI doesn’t
Who the website is for
Before a designer opens any software, they think about who will be using the site. How old are they? Are they on a phone at a job site, or at a desk with time to read? Are they comfortable with technology, or easily put off? Are they comparing three quotes, or looking for reassurance before they book?
A website for retirees looking for a local podiatrist and one for tradies who need a part in a hurry might start from the same template, but they need very different things. The first needs larger text and a phone number up front, while the second needs fast search and stock levels.
AI doesn’t know your customers unless you tell it, and even then it tends to write for a general audience, so the result rarely speaks to the specific people you want to reach.
How it should feel
People decide how they feel about a website in about half a second, well before they’ve read a word. They sense straight away whether it seems trustworthy or cheap, warm or cold. That reaction comes from colour, spacing, photography, fonts and tone of voice working together.
A funeral director’s website needs to feel calm and respectful, while a kids’ party business should feel bright and fun. AI knows the textbook version of this, like blue suggesting trust, but it has never felt anything looking at a page. It tends to land on a safe, pleasant average that rarely feels like you.
A lot of good design comes down to taking things away, like a background that competes with the photo or a font that’s slightly too fancy for your audience. Often a designer senses that something is off before they can explain why, then works backwards to fix it. AI isn’t built to make that kind of judgement call.
What it needs to do
Every business website has a job. It might be getting the phone to ring, taking bookings, collecting quote requests or selling products. A designer works backwards from that job and plans the path a visitor takes to get there: where they land, what they need to know, what might make them hesitate, and what makes the next step obvious.
AI drafts often look complete but have no real path. The buttons are there, but they don’t lead anywhere in particular, and sometimes they don’t lead anywhere at all.
The story it tells
Good pages follow a deliberate order, so a visitor goes from “I’m not sure I need this” to “I need this” by the time they reach the bottom. That usually means opening with the problem your customer has, showing you understand it, explaining how you fix it, and backing it up with reviews and examples.
AI arranges information neatly, but persuasion is a separate skill, and a tidy page of information rarely gets anyone to pick up the phone.
The words
AI was never in the room with your customers. It hasn’t heard how they describe their problem in their own words, so it writes the average of every website it has read. That’s where all the “tailored solutions” and “passion for excellence” comes from.
The copy that works best uses your customers’ language. A plain line like “avoid getting ripped off by your builder” will do far more than a polished tagline, because it shows you understand what worries them.
Whether everyone can use it
Accessibility means making sure people can use your website regardless of ability. That includes people with low vision or colour blindness, people with shaky hands, and people who use a keyboard or screen reader instead of a mouse. It also includes a lot of older Australians who simply find small grey text hard to read.
Good design handles this with enough contrast, readable text sizes, clear headings, descriptions on images, and forms that work without a mouse. In Australia, the Disability Discrimination Act applies to websites, and the WCAG guidelines are the standard most people use to meet it. AI-generated designs often use pale text on light backgrounds because it looks stylish, and they rarely check any of this. It’s one of the most common gaps I see.
What it’s like to use, especially on a phone
A website can look great in a screenshot and still be frustrating to use. Designers think about the moments that go wrong: someone who can’t find the price, a form that clears itself after an error, a menu that’s hard to tap, a page that takes eight seconds to load on a patchy connection.
Mobile is where AI websites most often come unstuck. A layout that looks lovely on a laptop can end up with overlapping text, tiny buttons and awkward gaps on a phone. Most visitors to a small business website are on their phone, so that’s the version that matters most.
The hundred small decisions
This part is the hardest to explain, because it’s made of things you’d never think to ask for. How do you break up a long page so people can scan it? Which details belong up front and which can be tucked away? Where should the eye land first? Is the spacing consistent from one section to the next?
Each of these decisions is small, but together they decide whether people read a page or leave it. You can’t ask AI for a design decision you’ve never heard of, and you can’t fix a problem you can’t see. A good designer makes those calls on your behalf.
Trends versus design that lasts
Design trends move quickly, and AI learns from whatever is popular right now. A website built entirely around this year’s look can feel dated surprisingly fast. A designer balances a fresh look with choices that will still work in three or four years, because most small businesses don’t want to rebuild their website every year.
The unglamorous bits
Then there’s everything nobody notices until it goes wrong:
- A domain and email address that you own and control
- Hosting that’s fast, secure and backed up
- Contact forms that arrive in your inbox
- A privacy policy, especially if you collect names and email addresses
- Search engine basics, so Google understands what your site is about
- Answer engine optimisation (AEO), so AI tools like ChatGPT and Google’s AI Overviews can find and cite you
- Updates and security monitoring after launch
None of it is exciting, and AI builders mostly leave it to you, but a website won’t keep working for long without it.
How to get better results from AI
If you’d like to build your website with AI yourself, you can close a surprising amount of that gap by changing how you use it. These are the same habits I use when I work with AI.
1. Do your homework before you prompt
Most disappointing AI websites start with a two-line prompt. Before you open any AI tool, write down who your customers are, the one thing you most want visitors to do, what makes you different from the next business, and three competitors whose websites you’d want to beat. That page of notes will improve the result more than any clever prompt.
2. Gather your customers’ real words
Pull together enquiry emails, reviews, notes from phone calls and the questions people ask you over and over. Paste them into the AI and ask it to list the exact phrases customers use, the problems they mention most and the worries that come up before they buy. AI is excellent at finding those patterns. Use the phrases it finds in your headings and copy. You gather the material and AI helps organise it, but you decide what goes on the page.
3. Give it references to work from
AI copies well and invents poorly, so give it something to work from. Share two or three websites you like and explain what you like about each, whether that’s the photography on one or the layout of another. Don’t ask it to clone a single site. It’s also worth looking at what your customers read and use, like a trade magazine or a brand they love, because that tells you a lot about the look they’ll trust.
4. Write a proper brief
Treat the AI like a new designer on their first day. A good brief covers:
- What your business does, where, and for whom
- Three words for how the website should feel
- The main action you want visitors to take
- The pages you need and what each one is for
- Your real photos, logo and brand colours if you have them
- Things to avoid, like gradients, stock-style images or buzzwords
5. Ask it to critique its own work
AI is better at reviewing than you might expect, as long as you ask. Try prompts like “What would an experienced web designer change about this page?”, “Check every text colour for contrast against its background”, “How will this layout behave on a small phone screen?” and “Where might a first-time visitor get confused?” You won’t agree with everything it says, but it will point out things you’d otherwise miss.
6. Be specific when you ask for changes
“This looks weird, can you fix it?” leaves the AI guessing, and every guess can break something else without you noticing. Describe what you see, where it is and what you want instead. For example: “The gap between the heading and the three cards is much bigger than the gap between the cards. Make it the same.” If you’re not sure what to call something, take a screenshot, circle the problem and attach it.
7. Change one thing at a time
Ask for one change per prompt, check the result on desktop and on your phone, and keep a copy of any version you’re happy with. AI has a habit of fixing one thing and undoing another, which is much easier to spot when you’ve only changed one thing.
8. Strip out the AI tells
Go through the finished draft and remove the giveaways: decorative gradients, generic icons, stock-style photos and words like “elevate”, “seamless”, “unlock” and “empower”. Replace them with your own photos, your own examples and your customers’ language. Most people are surprised how much difference this makes.
9. Test it on a real person
Hand your phone to a friend or a customer who hasn’t seen the site. Ask them to find your prices, make a booking or send an enquiry, and watch without helping. Wherever they hesitate, your website needs work.
10. Decide where it will live before you fall in love with it
Before you spend weeks refining a design, check whether you’ll own it, whether you can move it, and whether you’ll be able to update it without going back to the AI every time. A beautiful prototype that only exists inside one tool can be harder to turn into a working website than people expect.
A quick self-check for your AI website
If you’ve already made something with AI, try answering these honestly:
- If a competitor copied your site tomorrow, would your customers know it was yours?
- Does the home page make it obvious what you do, who it’s for and what to do next?
- Does the copy sound like you and your customers, or like every other website?
- Can you read every piece of text comfortably on your phone, outdoors?
- Have you sent a test enquiry through every form and received it?
- Is the site on your own domain, with hosting you control?
- Do you know how you’ll update it in six months’ time?
- Does it show up when you search for your business name and what you do?
If you answered no or “not sure” to a few of these, you’re not alone. Most AI drafts I see fall short on at least half of them.
So where does that leave you?
I’m not suggesting you avoid AI. Used well, it’s a great way to explore ideas and work out what you want, and the tips above will get you further than most people go. Just treat the draft as the start of the process.
If you’d like to finish the job yourself, my post on what to do after you’ve made a website with AI walks through the steps. If you’re still weighing it up, I’ve written a longer piece on whether you should build your business website with AI. And if you’d rather hand your draft to someone who does this every day, that’s what our AI to Reality service is for. We take what you’ve made, keep what you like about it, and turn it into a real website built around your customers.
