How We Build Contractor Websites With Claude
Every agency now claims to use AI. Almost none say what that means in practice. This is our actual workflow, including the parts where it makes things worse.
By Ryan Goering, CEO & Founder, BaaDigi
Last updated August 2026
The Short Answer
We use Claude for three things when building a contractor website: turning research into structured page briefs, drafting long pages against those briefs, and reading finished pages back the way an automated agent would. We do not use it to decide what pages to build, because that decision depends on real search demand it cannot see — and getting that wrong is expensive. We learned that the hard way on our own site: a page built around a phrase that felt obviously right turned out to have no measurable search volume at all, and earned zero clicks in four months. Now every page starts with verified demand data, and the model only writes once a human has decided the page deserves to exist.
What decides whether a page gets built?
Real search demand, checked before anything is written. We pull volume and difficulty for the exact phrase, and if there is no measurable demand the page does not get made — however obviously right it sounds. This is not theoretical for us. The page you are reading replaced one we built on the phrase "AI prompts for contractors". It reads well, the prompts are good, and the phrase returns no measurable search volume. It sat at position 37 and earned zero organic clicks in four months while collecting six email signups from people who found it some other way. A model would have happily written forty more pages just like it. That is the failure mode nobody mentions when they say they use AI — not that it writes badly, but that it writes confidently about things nobody asked for.
Where does Claude actually help?
Three places, in order of value. **Structuring research into a brief.** We give it the demand data, the competitor pages, and what the client told us on the kickoff call, and ask for a page structure: what question the page answers, what headings it needs, what facts we are missing. This is the highest-leverage step and the least visible. **Drafting long pages against that brief.** Service pages, guides, and troubleshooting content run 1,200 to 2,000 words and need to hold voice and facts across the whole thing. Claude follows a detailed structural brief more literally than the alternatives, which matters when the brief is the thing carrying the SEO work. **Reading the finished page back.** We paste the built page source in and ask what an automated agent could extract — what the business does, where it operates, how to get a quote. If it cannot answer from the source alone, neither can the systems that increasingly visit a site before any human does.
Where does it make things worse?
Anywhere a confident wrong answer costs money. It will invent prices, warranty terms, licensing requirements, and years in business, and it will do it in the same tone as the sentences around them. On a contractor site those are not stylistic errors — publishing a warranty term the client does not offer is a real problem. So every brief includes an instruction to bracket unknowns rather than fill them, and every draft gets read by a person who knows the client. The rule is simple: if the client did not tell us, it does not go on the page. It is also bad at knowing when to stop. Left alone it will produce a service page for every conceivable variant, which is how sites end up with two hundred near-identical pages competing with each other. Deciding what not to build is entirely a human job.
What does the workflow look like end to end?
**1. Demand check.** Pull real volume and difficulty. Kill anything with no measurable demand. **2. Recon.** Read what already exists — the client's current site, the competitors ranking now, what questions the market actually asks. **3. Brief.** Claude turns the research into a page structure. A human edits it, because this is where the decisions live. **4. Draft.** Claude writes against the brief, bracketing every fact it does not have. **5. Fill and verify.** The client supplies the bracketed facts. Nothing invented survives to publication. **6. Build.** The page is built with schema, internal links, and the answer-first structure AI engines quote from. **7. Read it back.** Agent readability check, link audit, verify every internal link resolves. **8. Measure.** Search Console at 30 and 90 days. Pages that do not earn impressions get rewritten or removed. Steps 1, 3, 5 and 8 are where the value is, and all four are human decisions. The model does 2, 4 and 6 faster than we could — which is worth a great deal, but only after the decisions are made.
Should you do this yourself?
For a lot of it, yes, and the prompts across this library are the same ones we use. The parts that are hard to do alone are the demand check, because it needs paid keyword data, and knowing when a page is not worth building, which mostly comes from having built the wrong pages before. If you want to try it, start with the website audit checklist on your existing site. You will find more revenue in fixing what is already there than in publishing anything new — that is true for almost every site we look at, including the ones that hired us to write more pages.
Frequently Asked Questions
Does AI-built content rank on Google?
Content ranks or does not rank on whether it is useful, not on who typed it. What fails is content aimed at phrases nobody searches, or content with nothing in it only that business could say. Both are easier to produce at volume with AI, which is why the guardrails matter more than the tool.
How do you stop AI inventing facts about a client?
Instruct it to bracket unknowns rather than fill them, and have someone who knows the client read every draft. Our rule is that if the client did not tell us, it does not go on the page. That catches nearly everything.
Why publish your own workflow?
Because "we use AI" has stopped meaning anything, and the specifics are the only way to tell one claim from another. Also, most business owners are not going to build their own site — they are going to want to know whoever does it understands what they are doing.
What happened to the page this one replaced?
It redirects here. It targeted a phrase with no measurable search volume, earned zero organic clicks in four months, and is a better example of the failure mode than anything we could invent — so it seemed worth keeping the story rather than quietly deleting it.
Ready to Get More Leads?
Get a free marketing audit for your contracting business. We'll show you exactly where you're losing leads.
Get Your Free Audit