What Railway’s AI search growth can and can’t teach a home app

Railway, a cloud platform where developers run their apps, has grown fast this year, and that growth has been credited online to AI search. Part of that story is real, and the real part is worth learning from. It just doesn’t carry over to a house the way it looks like it should.

How Railway grew, and what it publishes

On Knuckle Up in October, the host put Railway at about 40 people and near $100 million in annual revenue, and its CEO, Jake Cooper, answered “pretty close, just shy” (Knuckle Up, episode 15). In January, VentureBeat reported that Railway “continues to expand at 15 percent month-over-month” (VentureBeat, January 22, 2026).

Cooper would rather solve a problem himself than study how someone else did: “it’s almost like cheating on your homework when you go and look at like somebody else’s solution.” He’s also open about the models: “on the go-to-market side, we’re figuring out how to like incept the models to like, recommend us more and more and more.”

Look at what Railway publishes. Its sitemap lists more than 6,800 one-click templates, each a page that deploys a piece of software when you click it (the one for n8n shows more than 17,000 deploys), plus 312 changelog pages going back to October 2020 (railway.com sitemap, n8n template). Its docs keep a list for AI tools with 414 entries, 127 guides among them, and 10 pages comparing Railway with other platforms (docs.railway.com/llms.txt).

And the models do pick it, sometimes. In a February 2026 study of the tools Claude Code chooses, Railway was picked for 32 of 112 deploy answers (28.6%), against Vercel’s 86 (76.8%), and for 82% in the study’s one Python project (Amplifying.ai, February 2026). In the July edition, Docker led Python at 43% (Amplifying.ai, July 2026).

Why Railway’s pages work: each one does the job

A developer asks an assistant how to deploy n8n. The best possible answer is a link that does it. Railway’s template page is that link: read it, click Deploy, and it runs.

The page isn’t an article about the product. It’s the product doing the thing. The guides and compare pages work the same way, because a developer choosing a host wants exactly those. The answer and the action are the same click.

That’s the part worth admiring. Railway didn’t write pages about deploying. It turned deploying into pages.

Why home questions are different

“How long does a water heater last?” is the first kind. Every assistant already knows. The answer is the same for every house, and thousands of pages already say it. One more page adds nothing.

“When was our water heater put in?” is the second kind. No public page can answer it, because the answer is in one house’s papers.

So a Railway-style pile of pages for homes would be either the same as everyone’s or impossible to write. Google is direct about the first case. Its spam policies name “attempting to manipulate generative AI responses in Google Search”, and scaled content abuse, when “many pages are generated for the primary purpose of manipulating search rankings and not helping users”, typically “large amounts of unoriginal content that provides little to no value to users, no matter how it’s created” (Google, spam policies, updated August 28, 2026).

A file just for AI tools doesn’t change that. Google says you don’t need one, “as Google Search itself doesn’t use them” (Google, AI optimization guide, updated July 10, 2026). Ahrefs found that of about 38,000 domains with a valid llms.txt file, “97% saw no requests for it whatsoever in May” (Ahrefs, June 15, 2026). Google’s John Mueller, quoted there, called it a “temporary crutch, perhaps to save some tokens”, and Ahrefs reads that as a help for coding tools reading developer docs. That fits Railway, whose readers include coding tools. It doesn’t fit a homeowner.

What gets a home page cited by AI search

An assistant cites a page when the page says something it can’t get anywhere else, and shows where that came from.

For homes, that’s sourced facts. How long things last uses only Department of Energy and Housing and Urban Development figures, and every figure names the rule or table it came from. An assistant can repeat a number from it and point to the source.

Why we don’t publish gas, water and electric how-tos

The worst version of a page farm for homes is how-tos on gas, water and electric. A wrong step there isn’t a lost ranking, it’s a hazard. When we’re not sure a step is right for every house, we don’t publish it.

How we’ll measure it: citations, not traffic

Each month, the same set of homeowner questions, asked of the assistants people use, and two counts: whether Homaic is named, and whether the answer links one of our pages as its source.

Not traffic, and not rankings. A citation is the one signal that a page said something worth citing, and it’s the one we’ll count.

Where the answers about your own house live

Railway’s pages work because its product is public. A home’s record is private, and that’s the point. The answers to the second kind of question are in it, and only you and the people you let in can open it.

In Homaic, an AI assistant you use reads your record only when you connect it, and you can disconnect it in Settings (homaic.io/data). The record is yours: you can take it as a download that opens without Homaic, any size, and leave with it (same page).

Look up your home at homaic.io.