Skip to content
WholeHomeBuild
WholeHomeBuild › AI-ready guide
Guide · updated 2026-10-04

How AI assistants pick a homebuilder, and the 10 checks that decide it

Buyers now ask an assistant before they ask an agent. Here is what the assistant can and cannot read on a builder's website.

A buyer asks an assistant for "a builder in Leander with homes under $400,000 and a pool in the community." The assistant does not browse the way a person does. It reads text, follows plain links, looks for structured data and gives up on anything it cannot parse. Then it writes one answer with two or three names in it.

Which names make the cut depends less on marketing budget than on whether the builder's website can be read by a machine. We checked 116 US homebuilder websites and 221 building-product manufacturer websites against ten things that decide it. The numbers below are totals. We do not publish individual company scores here.

What we found

  • The median builder site scored 66 out of 100. 50 of 114 scored 70 or better; 16 scored under 45.
  • 61% of builder sites carry Organization schema on the homepage, but only 25% mark up a community, plan or home with Product, Offer or Residence schema. The assistant learns the company exists and little about what it sells.
  • Only 57% show a price in the HTML of the homepage or a community page. Buyers ask about price first.
  • 29% publish an llms.txt file.
  • 40% have a FAQ an assistant can lift answers from.
  • 1 builder site blocks at least one major AI agent in robots.txt; 1 refused a plainly labeled automated request at the CDN level, which robots.txt never shows. 2 more did not answer automated requests at all and are left out of these totals.

The 10 checks

Each check carries points, and the ten add up to 100. Each one is something you can test on your own site in a few minutes.

  1. AI crawlers allowed in robots.txtrobots.txt does not shut out the agents ChatGPT, Claude, Perplexity and Google's AI use to read the web. Fix: Remove Disallow rules for GPTBot, OAI-SearchBot, ChatGPT-User, ClaudeBot, Claude-User, PerplexityBot and Google-Extended, or add explicit Allow groups for them.
  2. XML sitemap publishedA sitemap.xml (or one named in robots.txt) lists every page so crawlers do not have to guess. Fix: Publish /sitemap.xml listing every community, plan and product page, and name it in robots.txt with a Sitemap: line.
  3. Organization schema (JSON-LD)The homepage carries schema.org Organization, LocalBusiness or HomeAndConstructionBusiness markup naming the company, its site and its contact points. Fix: Add an Organization (or HomeAndConstructionBusiness) JSON-LD block to the homepage with name, url, logo, telephone, address and sameAs links.
  4. Home or product schemaCommunity, plan or product pages carry Residence, Product, Offer or Place markup an assistant can quote. Fix: Mark up each community or plan page (or product page) with schema.org Product/Offer or SingleFamilyResidence/Place, including price and location.
  5. Fast, readable HTML without JavaScriptThe homepage answers in under 3 seconds and carries at least 250 words of real text in the HTML itself. Many AI fetchers do not run JavaScript. Fix: Server-render the homepage copy (or pre-render it) so the words are in the HTML, and keep the first response under 3 seconds.
  6. Community or product pages linked in plain HTMLThe homepage links to community, plan or inventory pages (builders) or product pages (makers) with ordinary <a href> links. Fix: Put plain <a href> links to every community/plan (or product line) in the homepage or main nav, not only in a JavaScript map or search widget.
  7. Prices (builders) or spec documents (makers) visibleBuilders: a price ('from the $300s', '$425,990') appears in the HTML of the homepage or a community page. Makers: spec sheets, installation guides or warranty documents are linked. Fix: Builders: print base prices on community and plan pages as text. Makers: link spec sheets, install guides and warranties as plain links.
  8. FAQ contentA FAQ section, FAQ page or FAQPage schema answers the questions buyers ask assistants. Fix: Publish a FAQ page answering the real buyer questions (price ranges, HOA, warranty, lead times) and mark it up with FAQPage schema.
  9. llms.txt publishedA plain-text /llms.txt summarizing the company for language models. Fix: Publish /llms.txt: a short plain-text summary of who you are, where you build or sell, and links to your key pages.
  10. Machine-readable contactA tel: link, or a telephone/address in structured data, so an assistant can hand a buyer a number. Fix: Use tel: links for phone numbers and include telephone and address in the Organization JSON-LD.

How the assistants actually read a builder

There are two paths. Training crawlers (GPTBot, ClaudeBot, Google-Extended) collect pages that shape what a model knows in general. Search and user agents (OAI-SearchBot, ChatGPT-User, Claude-User, PerplexityBot) fetch pages live when someone asks a question. The second path is the one that answers "who builds in Leander under $400,000" with current inventory. Block it and the assistant falls back on listing portals and old news.

Most of these agents do not run JavaScript. If community cards, prices and the "find your home" map are drawn by a script after the page loads, the agent sees an empty shell. Server-rendered text and ordinary links fix that without changing how the site looks to people.

Structured data is the shortcut. A community page with Product or Offer markup carrying price, location and builder name hands the assistant the exact facts it would otherwise have to guess from layout.

Building-product makers have the same problem

Builders and their trades ask assistants which window meets a given code, which shingle carries a wind warranty, which water heater fits a closet. Across 207 manufacturer sites the median score was 68. The common gaps were product schema, llms.txt and spec sheets linked as plain documents. Our directory of building-product makers lists them by category.

Questions

How does ChatGPT decide which homebuilder to recommend?

It answers from what it can read: pages its crawlers fetched, pages its search tool fetches live, and structured data on those pages. A builder whose site shows prices, communities and contact details as plain text and schema gives the assistant something to quote. A site that hides them behind JavaScript or blocks AI crawlers gives it nothing, so the answer names someone else.

Does blocking GPTBot keep my site out of ChatGPT?

Blocking GPTBot stops OpenAI from using your pages for training. ChatGPT search uses OAI-SearchBot and ChatGPT-User. Many sites block all three without meaning to. If you want to be recommended, allow the search and user agents at minimum.

What is llms.txt?

A plain-text file at /llms.txt that summarizes your company for language models: who you are, where you build or what you make, and links to the pages that matter. It is new and optional, and it costs an afternoon.

Is this the same as SEO?

It overlaps. Sitemaps, fast pages and structured data help both. The difference is that an assistant writes one answer instead of showing ten links, so being readable and quotable matters more than ranking position.

Can I check my own site?

Yes. Open your homepage with JavaScript turned off and see what is left. Read your robots.txt for the agents named above. Paste a community page into Google's Rich Results Test to see its schema. That covers seven of the ten checks.

How we measured

Builders were selected from public facts: SEC filers under SIC code 1531 and other listed homebuilders, our own directories, and builders named on master-planned community websites in the 15 largest US metros. We did not use any published ranking or award list. Each site was fetched politely from a workstation, without running JavaScript, on 2026-10-04. A site that did not answer was left out of the totals rather than scored as zero.