How This Website Was Built with PolyDraft — The Full Log

This website was built with PolyDraft. Not "designed by humans and assisted by AI" — the same agent our customers run produced the research, the keyword plan, the page structure, the copy and the code you are reading right now. This post is the log, because a builder that cannot build its own site is not worth trusting with yours.

Step one: the brief, not a template

The build started with a plain-language brief describing what PolyDraft is, who it serves and what it must never say. That brief became a verified knowledge base — company entity, address, plan prices, product capabilities, traction figures — and the agent treats it as the only source of business truth. Anything not in it does not appear on the site. That single constraint is what prevents the fabricated testimonials and invented statistics that plague AI-generated marketing pages.

The brief also carried explicit prohibitions: no ranking guarantees, no naming of AI vendors or data providers, no social profile links until real profiles exist. Those became structural rules the agent honored on every page rather than a style note somebody forgets by page nine.

Step two: competitor research on the live web

The original build brief described a review of eight competing website builders. This log does not include their dated source pages or a reproducible comparison dataset, so that historical review should not be used as a current comparison of their capabilities.

The resulting positioning focused on ownership, source portability and inspectable search setup. Evaluate those questions directly in the current product and contract. Competitor support can change; the absence of a feature in an old review is not evidence of its absence today.

Step three: one money keyword per page

The agent pulled live search demand data — monthly volume, difficulty and intent — for the whole topic cluster, then assigned exactly one primary keyword to each page. The homepage owns the flagship term. The features page owns the SEO-specific variant. Each audience page owns its own commercial term, and the three free tools capture high-volume utility demand at difficulty scores the main pages could never reach quickly.

This discipline matters more than any individual keyword choice. When several pages chase the same term, they split their own signals and none of them wins. A site-wide internal-link dictionary was generated at the same time, mapping every money keyword to the single page that owns it, so links inside the copy always point at the right target.

Step four: writing, then checking

Each page was written from its own research brief: recommended sections, target word count, structured-data type, internal links to place, and the questions real people search that the FAQ should answer. Key sections open answer-first — the claim in the first sentence, the supporting detail after — because that structure is what AI assistants can lift and quote accurately.

When the pages were done, an automated quality gate swept the whole site for content depth, SEO completeness, accessibility and design consistency, and the issues it found were fixed before anything shipped. The technical layer went out by default: unique titles and meta descriptions, single H1 per page, Schema.org structured data, breadcrumbs, canonical URLs, an XML sitemap, an RSS feed, a live 404 page, an llms.txt file and a robots.txt file, which Google documents as the control for crawler access, that explicitly welcomes AI crawlers. Both are first-party evidence on this site: you can open our own llms.txt and our own robots.txt and check them against this claim right now.

What it means for your build

Nothing in this process is specific to us. The same agent, given your brief, does the same work on your market: real competitor research, a real keyword plan, original copy, a complete site, an automated quality pass, and a one-click deploy to a cloud account you own. You can watch it happen in a live preview and change anything by asking in chat.

Building is free — you only pay to publish. If you want the mechanics in detail, read how it works, or start a build and read your own log.

Continue with the next task

Apply an independent checklist when assessing the build described here. Read the website builder evaluation questions.

The build log records a launch; use a separate workflow to maintain the published site. Read the post-publication content refresh workflow.

Inspect the published examples in the PolyDraft website showcase and compare the stated scope with the pages you can see.

Sources

• Schema.org — the shared structured-data vocabulary
• Google Search Central — robots.txt introduction and guide

Update note — October 8, 2026: AI-assisted maintenance clarified source boundaries and added related practical guides. This update does not claim a new human expert review or measured search results.