AI microsites let a small website target one service in one town, and the work is deciding where to build, giving the agent a custom brief and checking its output. A transcript from SEO consultant Jesse Cunningham shows a build on Anthropic Claude in 2026, plus the limits no model removes: search competition, thin copy and phone handling.
What Are AI Microsites and When Do They Work?
AI microsites are small websites built around one service in one location, usually a suburb rather than a metro core, and assembled with an AI coding agent. They work when several are built at once, because a single site rarely earns enough to justify the effort.
Jesse Cunningham, an SEO consultant who runs the Rank Expand agency and academy, describes maintaining more than 50 of these sites in a July 2026 YouTube walkthrough. His example is slab repair in Houston, Texas, a high-ticket service with local competition low enough to enter. The money model is simple: each site targets a service-area keyword, captures calls, and the leads are either worked in-house or resold.
The transcript describes this as a repeatable tactic rather than a tested program. Fewer than a handful of builds, plus an unverified comparison against one large multi-city site, is not evidence that microsites outperform. Treat each site as a small searcher-demand bet to be checked before investing hours.
Three conditions decide whether a microsite is worth building.
- Confirmed search demand for the service and place.
- Enough competition-free angle to rank for that query.
- A phone line or intake process that answers when it rings.
The third condition is the one most builders skip. Cunningham says he does not answer calls himself and relies on an AI agent on top of each site to qualify and route inquiries.
Which Claude Code Tier Should Run a Microsite Build?
Model tiers should match the job: a mid-tier model for long unattended builds, a fast mode for time-sensitive edits, and a stronger reasoning model for hard design problems. Cunningham's ranking puts the workhorse tier first, a fast mode second, and a strongest model as a specialist for narrow cases.
His setup in the video is the Claude family of models made by Anthropic moves his default to the tier he calls the workhorse because it is cheaper than the strongest model while handling most of the site work, and describes the fast mode as a paid trade of quality for speed. The transcript's model names and price comparisons could not be verified against current listings, so treat the tier ranking as the creator's working preference.
Model choice buys a few minutes and a smaller bill. It does not replace the brief. A prompt that specifies the section, the conversion goal and the number of form fields produced a usable hero rewrite in a single pass; a vague prompt to improve the same section would not have.
How Do You Choose a Service Area for an AI Microsite?
Service-area selection should follow measured competition data, and a major metro is acceptable when that data shows it is as easy to enter as the surrounding suburbs. Cunningham ran keyword and competition research on local service queries before choosing Houston for the slab repair build.
Suburbs remain the usual choice for a reason. Competition in the towns around a metro is typically thinner than for the metro name itself, and the searchers still live there. A Dallas or Indianapolis map makes the pattern obvious: one central city, a ring of named towns, each a candidate for its own site.
The research step is not optional, and it is where most of the risk sits. He uses a scored research tool inside his academy to rank candidate locations, which is a paid and unverified process from the outside. A free substitute: check whether local businesses are bidding on the service in each town, and whether the top results are thin pages that can be beaten.
How Do You Prompt an AI Coding Agent to Improve a Page?
Prompt the AI coding agent with a concrete output requirement instead of an adjective, and give it a visual reference whenever layout matters. Cunningham's first hero-section prompt named four form fields and a call to action, and the section came back rewritten in under a minute of his own time.
Long builds run in the background. While the agent worked, he switched to voice dictation with a speech-to-text tool so he could issue longer instructions quickly, then edited other parts of the site. That overlap is the practical benefit of an agent that keeps working without supervision, and it is the reason cheap models matter across many sites.
When a generated map came back too abstract, the fix was a screenshot of an existing map plus an explicit ban on third-party tiles. The agent then drew the Houston metro as an SVG, placing each city at its real latitude and longitude on a simple projection. Accuracy claims of that kind are unverified, but the approach matters more than the output quality.
What Separates an AI Microsite That Ranks From a Thin Page?
Custom structure, internal linking and sourced data are what separate a ranking microsite from a thin page. The build in the video shipped roughly ten service-location pages, a primary service page, and internal links between the cost guide and the location pages.
The strongest element was outbound citation. The agent researched documentation and linked to it from the site, including a USDA document surfaced as a PDF. Most small business websites cite nothing, which makes a sourced page easier to distinguish.
Page depth ran in the other direction. After the first tour, Cunningham judged the copy too short and asked for the headings and body text to be built out on the theory that short local pages struggle. That instinct matches search guidance in spirit, but no ranking case from the video proves it.
A practical rule from the walkthrough: if a section looks generic, rewrite the brief and regenerate instead of editing by hand. Cheap model time is the whole reason that loop is affordable.
How Should a Multi-Microsite Operation Organize Its Work?
A multi-microsite operation should keep a shared knowledge base, a repeatable build recipe and a single owner of the deploy step. Cunningham keeps what he calls an LLM Wiki that commits to a private GitHub repository and serves as the reference for rolling out the next site.
The build also depends on an established deploy loop built in an earlier session, in his case over SFTP, which lets him describe a change in plain language and have the agent publish it. That infrastructure is the difference between one finished site and twenty.
Guard the deploy credentials. He describes an early push that went to a GitHub account he could not identify. Treat that as a reminder to confirm remote URLs, deploy targets and account access before asking any agent to publish.
AI Microsites, Claude Search Visibility Compared
The table below compares what each layer of the workflow contributes, based only on what the transcript shows and on each product's own documentation. It is not a benchmark of search outcomes.
| Layer | Role in the workflow | Evidence status |
|---|---|---|
| Claude | Writes and edits site code over many turns | Creator's experience |
| Search competition research | Scores candidate service areas | Paid process, unverified |
| Custom map graphic | Adds visual identity to a service-area page | Shown in video, accuracy unverified |
| Deploy loop | Publishes edits without manual upload | Shown, prior-session setup |
| AI phone agent | Handles inbound calls at scale | Described, not demonstrated |
The default habit of copying a stock theme is what the custom-map step is meant to break. Cunningham argues there is a real advantage in a site that does not share a template with thousands of others, and his fix is to ask the agent for a bespoke layout rather than a plugin.
None of these layers substitutes for the others. A custom map on a page with no demand behind the query still brings no calls, and a well-researched area served by a thin page still has to compete.
Frequently Asked Questions
- What is an AI microsite? An AI microsite is a small website built around one service in one location, usually assembled with an AI coding agent. The narrow scope lets a single site target a specific local search phrase instead of competing for a broad city term.
- Can a microsite rank on Google? A microsite can rank when the service-area query has real demand, few strong competitors and a page with enough custom content to compete. No model guarantees a position, and every candidate location needs its own research before a build.
- Does an AI-built website get penalized? Google evaluates the quality and helpfulness of the content, not whether a model drafted the code, so an AI-built site is judged like any other. Thin pages with no original information are the real risk, which is why the walkthrough pushes for longer copy and cited sources.
- Which model should build the site? Match the model to the task: a cheaper workhorse tier for long builds, a fast mode for quick edits, and a stronger reasoning model for the hardest design work. Verify current model names and prices from the provider before you plan a budget.
- How many microsites do you need before the model pays off? The transcript builds more than 50 and treats the concept as a volume business. Most of the leverage comes from reusing one deploy loop and one knowledge base across every new site rather than from any single build.
Turn a Recorded Walkthrough Into a Written Guide
The Houston slab repair build works because every decision is visible: the area chosen from competition data, the hero rewrite with a defined call to action, the rejection of a bad map, the push for longer copy. Those steps are easier to repeat when they are written down and searchable instead of buried in a recording.
The same applies to your own material. Dev Doido do canal do youtube, for instance, covers technical topics at length, and much of that reasoning can serve readers who will never watch the video. A link from the written version also points people back to the source.
If you have a walkthrough, interview or technical explanation sitting in a YouTube upload, Skala blog turns it into a draft article you can review before publishing. Paste the video URL, let the transcript come through, and edit the result into something worth ranking. The Crazystack site is another example of how a technical brand can present written content alongside its other work.
Fork this article
Start a new branch from the same video, shaped your way. You keep the credit; the original keeps the attribution.
A fork in another language is filed as a translation of this article, so the two pages point at each other. You can unlink it later from the editor.
0/240
You are creating
- Format
- For
- Language
- Source
- Your angle
You will be asked to sign in before it is generated.
Buy credits