How many city pages can I build before Google treats them as doorway pages?

The definition that matters

Google's objection to doorway pages is that they exist to capture a query and then funnel the visitor somewhere else, adding nothing along the way. The tell is not the URL pattern. It is that the page has no reason to exist for a reader.

Apply the reader test before the search test. If someone in that town landed on the page cold, would they learn anything they could not have learned on your main service page? If the honest answer is no, the page is a liability.

What actually distinguishes a real location page

• Work performed there. Described jobs, described conditions, described outcomes, without publishing customer identities.
• Local constraints. Permit processes, HOA rules, well versus municipal water, common construction eras and the equipment that goes with them, seasonal patterns specific to that area.
• Logistics. Which crew covers it, typical response expectations, whether you cover it for emergency work or scheduled work only.
• Named geography. Neighborhoods and landmarks used naturally, not stuffed into a list at the bottom.
• Reviews from that area, pulled in and attributed by first name and town.

A workable rule for how many to build

Build a location page when you can defend it with three specific facts and at least one piece of proof of work in that area. If you cannot, the page is premature. That rule naturally produces a small set at first and grows as the business actually grows into the territory.

It also produces a better answer to the question people are really asking, which is how to cover forty towns. You do not, at least not immediately. You cover the ones where you already have jobs, and you let the rest be served by regional pages until you earn them.

Where automation belongs and where it does not

Generating forty near-identical pages is easy and counterproductive. Generating forty pages that each assemble genuinely different underlying data is a different exercise and a legitimate one.

If your field service system knows which jobs happened in which ZIP codes, what work was done and what the seasonal pattern looks like, those facts can populate location pages that differ because the underlying reality differs. That is the useful version of content automation: the model assembles and writes, but the substance comes from your operational data rather than from the model's imagination. Human review still belongs in the loop before anything publishes, and the same discipline applies to answer content intended for AI assistants, which reward specificity for exactly the same reason.