Can AI Write Your Homepage? We Tested Claude on a Toronto Brief

Can AI write your homepage? It can write one in about ten seconds, and the result reads well. We tested it on 1 October 2026 with a fictional GTA physiotherapy clinic and a short, honest brief. It did not invent hours or awards, which surprised us. But it guessed at details we never gave it, wrote a headline that would fit any clinic in Ontario, and had no proof to show. A person still has to supply the facts, the proof and the decisions.

What we asked it to do

We invented a business: Maple & Pine Physiotherapy, a small clinic in Oakville, Ontario. It is a fictional brief, not a client. We gave the AI the kind of message a busy owner would type. Three services, three treatment rooms, a short walk from the Oakville GO station, patients in Oakville, Burlington and Mississauga, direct billing for most insurance plans, and one goal: online bookings. We left out the opening hours, the years in business, any reviews, any credentials and the owner's name on purpose, to see what it would do with the gaps.

Why now? On 28 September 2026 Anthropic released Claude Sonnet 5.5, and a lot of owners will be wondering if AI is good enough yet to write their site. So we ran the test on that exact model. We ran it once, in a clean session with no other context, and show it untouched. The prompt, for the record:

What it wrote in under a minute

Here is the raw output, word for word. It took about nine seconds, and it was produced by Claude Sonnet 5.5.

Credit where it is due. The grammar is clean, the tone is warm, the structure makes sense, and the button is repeated so a ready visitor never has to scroll. It also added a note listing what it had left out because it did not have the details: credentials, years in practice, testimonials, and whether a referral is needed. That is more careful than we expected. As a first draft, it is good.

Where the draft goes wrong

Here is the teardown. We only list what we actually found in the output above.

• Details we never gave it. “Hands-on treatment,” “practical exercise plans,” and working “with your surgeon's recommendations” in the rest of the draft all describe how the clinic works. The owner never said so. A real clinic might not treat that way, and a regulated health service can't publish a guess about its own care. The fix: check every sentence about method against how the clinic really works.
• Claims stretched past the brief. We said “a short walk.” It wrote “steps from.” We said “most plans.” It promised “fewer out-of-pocket surprises” and “a few minutes” to book. Small shifts, and each one is a promise the clinic now has to keep.
• A promise about results. The sports injury blurb says it will help patients “keep it from happening again.” No clinician should print that without being sure of it. Health copy needs a person who knows what can be promised, and what cannot.
• A headline that fits any business. Swap “Move better. Get back to what you love.” into a chiropractor, a personal trainer or a yoga studio and nothing changes. It says nothing about Oakville, nothing about physiotherapy, and nothing a patient could not read on ten other sites.
• No proof. No review, no credential, no years of experience, no named physiotherapist. The AI could not know any of it, so the page asks a worried person with a sore back to trust a clinic it has told them nothing about.
• Almost nothing local. The GO station is the only local detail, and it came from our brief. Nothing about Oakville in the headline, nothing about what someone would type to find the page, such as “physiotherapy Oakville.” The AI offered to add that as a follow-up, which is honest, but it is a decision about search, and it should be made first.

Notice what did not go wrong. It did not make up opening hours, awards or a review count. When we asked for less, it gave us less. If you hand it a thin brief, the risk is not wild invention. It is a smooth page that is too generic to win a booking, with a few small guesses mixed in that are easy to miss.

What a person had to decide

Fixing the wording is the small part. Four choices sit underneath, and no draft makes them for you. What is the one thing a visitor should do (book online, or call first)? Which proof leads (a credential, a patient story, how quickly you can see someone)? What does this clinic say about its area that another clinic cannot? What stays off the page?

Here is just the headline, to show the difference (our example, for the fictional clinic):

• Before: “Move better. Get back to what you love.”
• After: “Physiotherapy in Oakville, a short walk from the GO station.” It names the service, the town and the one convenience commuters care about, and it matches how a patient would search. It is plainer, and it is more useful. It also needs only facts the owner has already confirmed.

What to let AI draft, what must come from you, and what to hand off

This is the split we would use, based on what we saw.

Does this replace a web designer?

No. We are a small, founder-led studio and we use AI where it helps. It drafts quickly. But every page still needs its facts checked, its local detail added, its search intent chosen and its layout built around one action. That is the judgment part, and it is the reason a custom web design project is more than a draft in a template. We explain how a small studio uses AI on its own page. AI is a decent way to get a first draft. Then give the draft to someone who will turn it into a page that works.

What to do this week

• Write down the facts no AI can know (hours, service area, proof, the one action you want) and keep them in one place.
• Before any AI-written copy goes public, underline every claim you could not back up in front of a customer.
• Read your current homepage and swap in a competitor's name. If nothing breaks, the page is too generic.

Common questions about AI-written website copy