AI wrote one version because you gave it one business
One prompt describes one company, and a generator answers exactly the question it was asked: one address, one phone number, one opening-hours block, one team photo. Everything it produces is internally consistent, because it was built around a single point of contact that does not exist anywhere in your business.
A heating company with a workshop in Brno and four smaller depots usually receives this treatment without noticing. The site lists the head office, a central number and the hours of the reception desk. A customer in Jihlava calls Brno, gets told to bring the boiler in person, and drives ninety minutes to a building where the service entrance is around the back. Nothing on the website was wrong. It was simply written for one location, and the rest were never mentioned.
The tool did not fail. It cannot know that the depot in Tábor only sells spare parts, that Ostrava works Saturdays and Brno does not, or that the technician the customer asks for left the company last year. A prompt is not where a company keeps branch data, so the generator fills the gap with the only thing it has: the average of every similar business it has ever seen.
Swapping the city name is not a location page
Five pages with identical paragraphs and a different city in the heading are worse than one honest page, because they promise something the branch cannot keep. Changing the name in the H1 takes a second; changing the content underneath it is the work, and that is where generated location pages usually stop.
What a person actually needs from a branch page is narrow and practical: the exact address with the right entrance, the opening hours of that building, a phone number that reaches someone standing there, and a straight answer to whether this branch does what they came for. Add directions that mention the parking behind the building, and the page has done its job.
Search engines and AI assistants read the five pages side by side. Near-duplicates with a city name swapped are treated as one page repeated, and the filler that follows — we serve customers across the country, our quality is the same everywhere — tells both of them that nobody wrote this. The pages that get quoted in an answer are the ones that contain something true only about that place.
One site with location pages, or a site per branch
Up to about a dozen locations, one website with a locations section wins on every count that matters later: a single address to promote, one analytics account, one place to fix a typo. The alternative — a separate site per branch — splits the domain’s authority across five weak addresses, gives you five sets of hosting and certificates to renew, and multiplies every future edit by five.
The middle option suits companies whose branches offer the same thing and differ only in logistics: one site, a location picker near the top, and forms routed to the branch the visitor chose. It keeps the shared content shared while still letting someone in Plzeň reach the Plzeň team without passing through a central switchboard.
AI changes how fast you can produce each option, not which one is correct. Five variants of a location section can be generated before lunch, which makes the structural decision the one worth slowing down for. Once the site is published, the structure decides the cost of every change you will make for the next five years — a branch moves, closes, or changes its name, and the structure determines whether that is one edit or nine.
Five pages from one table
Every fact that differs between branches belongs in one table: address, opening hours, phone number, who works there, what that location actually does, and how a visitor reaches the right door. When that table is the source, the footer, the contact block and the individual location page cannot disagree, because they were generated from the same row.
This is where AI genuinely earns its place. Generate all five pages from the table so nothing drifts. Ask it to compare the whole site against the table and flag every number that survived a branch move. Let it write the driving directions from a real description of the site, then have a person read them, because the delivery entrance is on the side street and the map pins the front of the building.
What stays human is the commercial judgement. Which branches differ in what they sell, not just where they are. Who at each branch owns their row of the table and answers when it changes. And the proof that makes the page believable: photographs of that building, projects delivered from it, reviews that mention it by name. None of that can be generated, because none of it has been written down yet.
What breaks after the first month
Consistency is the first casualty, and a branch that moved is the second. One depot extends its Friday hours to six in the evening; three of the five pages still say five o’clock. A visitor who trusts the page arrives at a locked door, and the review that follows blames the company rather than the stale paragraph nobody was assigned to update.
Forms fail the same way, quietly. Every location page has a contact form. Two of them still send to a person who left the company in spring, and one goes to an inbox nobody opens because it was set up during the launch and never given an owner. The enquiries do not bounce; they simply arrive where nobody reads, which for the business is the same as not existing.
AI is good at the audit that prevents this: crawl the five pages, compare hours, phone numbers and service claims against the table, and report what no longer matches. The audit is only as current as the table behind it, and the table changes only when someone at the branch reports it. Tools can detect that the website is stale; they cannot detect that the Ostrava depot changed its hours on a paper notice by the door.
The ten-minute check before somebody drives to the wrong branch
Open each location page from a phone, standing in the city it claims to represent, and answer five questions: can I reach the right door, call someone who is there, see the hours of this building, confirm this branch does what I need, and send a message a person will actually read? If any answer is generic, the page is a template with a city name, and the customer will find out at the worst possible moment — after the trip.
Check the parts nobody sees too. Does each form reach an inbox someone opens? Does the map pin point at the entrance customers use rather than the middle of the block? Do the hours on the location page match the hours in the footer? Five minutes per branch is enough, and it is the difference between a site that routes customers and a site that describes them.
A multi-location website is where AI’s speed and a business’s reality visibly disagree. The machine can produce five consistent pages in an afternoon; only the company knows which facts differ between the branches and who keeps them true. That knowledge lives in the business, not in the generator — which is why a website built from it looks like the company, while one built from a prompt looks like every other company in the country.
Related service: Website and web application development