Ranking a single-location clinic with the Local SEO Starter

A solo marketer, one medical practice, and the five-skill Local SEO Starter loaded into Claude Code. Here is the afternoon, start to finish.

Most local SEO advice for healthcare is a checklist someone else wrote and never ran. This is the opposite: one real workflow, in order, with the handoffs shown. The client is a single-location dermatology clinic that has a Google Business Profile nobody has touched in a year, a handful of unanswered reviews, and a website that buries the phone number. The budget is small, so we are using the $49 Local SEO Starter, which is five skills: GBP Service Optimization, GBP Posts, Review Response Generator, Reputation Pattern Analyzer, and CTA Optimizer.

The point of running them as a chain instead of five separate sessions is that each one produces structured output the next one can read. You are not re-explaining the clinic five times. You explain it once, and the context travels.

Loading the bundle into Claude Code

Claude Code reads a CLAUDE.md file at the project root and pulls in anything you import into it. After buying the bundle you get five markdown files. Drop them into the project and point CLAUDE.md at them.

mkdir -p client-derm/.claude/skills
cp local-seo-starter/*.md client-derm/.claude/skills/
cd client-derm
echo "@.claude/skills/17-gbp-service-optimization.md" >> CLAUDE.md
echo "@.claude/skills/18-gbp-post-generator.md"       >> CLAUDE.md
echo "@.claude/skills/19-review-response-generator.md" >> CLAUDE.md
echo "@.claude/skills/20-reputation-pattern-analyzer.md" >> CLAUDE.md
echo "@.claude/skills/09-conversion-cta-optimizer.md" >> CLAUDE.md
claude

That is the whole install. The skills now load every time you open Claude Code in this folder, and you can call any of them by name.

Get the profile right before you drive traffic to it

There is no reason to generate posts for a profile that has the wrong categories and a thin service list, so the profile comes first. GBP Service Optimization asks for the practice name, the primary location, the actual services offered, and the primary category. It gates on those. If you leave the service list blank it stops and asks, because guessing the services is how you end up with a profile optimized for a clinic that does not exist.

For the dermatology clinic it returns a structured set: a primary category of Dermatologist, a short list of secondary categories, and a service block with a plain-language description for each treatment. That service block is the part that matters for the next steps, because it is the vocabulary the rest of the chain will reuse. The treatments named here are the same treatments the posts will talk about and the same ones the CTA work will point at.

Why the output is structured, not prose

The skill does not hand back a paragraph telling you to "optimize your categories." It hands back the categories, the service names, and the descriptions as discrete fields. You paste them straight into the profile, and you keep the file, because the review skill and the post skill both read it.

Read the reviews for patterns before answering any of them

The instinct is to answer reviews one at a time, oldest to newest. That is backwards. Reputation Pattern Analyzer takes the full set of reviews and looks for what repeats, because seven reviews mentioning a long wait is an operational fact, while one is noise. Paste in the reviews and it returns the themes ranked by frequency, separating what the clinic can act on from what it cannot.

For this clinic the analyzer surfaces two clear patterns: people love a specific nurse by name, and several mention confusion about parking. Neither of those is something you would reliably notice reading reviews top to bottom. Both are about to be useful.

Answer the reviews without breaking HIPAA

This is the step where healthcare is genuinely different, and where a generic AI tool will get you in trouble. The Review Response Generator carries a compliance block that does one critical thing: it never confirms the person was a patient. Even when a reviewer opens with "I had my mole removed here last week," the safe response does not acknowledge the visit, because confirming someone is a patient in a public reply is a privacy disclosure regardless of what they revealed first.

So the skill drafts replies that thank the reviewer, address the theme, and invite a private channel, without ever confirming care. It pulls the themes from step two, so the reply to a parking complaint references the validated parking the clinic actually offers, and the replies to the reviews praising that nurse stay warm without naming the patient. Every draft comes with the claims that need a human to verify flagged at the bottom. You read them, you confirm the parking detail is true, and you post.

Turn the patterns into a month of posts

Now the GBP Post Generator has something to work with. Instead of generic "we are open" filler, it writes posts that answer the questions the reviews revealed people actually have. The parking confusion becomes a short post explaining exactly where to park. The interest in a specific treatment becomes an educational post about what that treatment involves.

The skill produces a month of posts as a structured set, each with a suggested image direction and a call to action that matches the profile's service names from step one. Because it inherited that vocabulary, the posts and the profile describe the same clinic in the same words, which is what local ranking signals reward. The medical-content guardrail still applies here: anything that edges toward a clinical claim gets flagged for review rather than stated as fact.

Make sure the traffic has somewhere to go

The first four skills earn attention. The CTA Optimizer makes sure the attention converts. It audits the on-site path a new visitor takes, the dermatology clinic's habit of hiding the phone number below the fold being the obvious problem, and returns specific changes: a click-to-call button in the header, consultation framing that names the treatments from the profile, and a removal of the three competing buttons that currently split the visitor's attention.

It works because it knows the services. The CTA it writes is not "Contact us." It is a call to action built around the treatments the rest of the chain has been talking about, so the message a patient saw on the profile is the message they see on the page.

What the afternoon produced

By the end you have an optimized profile, a set of posted review responses that stay on the right side of patient privacy, a month of scheduled posts, and a fixed conversion path, all describing the same clinic in the same language because each skill read the one before it. Five separate prompt-pack prompts could not have done this, because none of them would have carried the service vocabulary or the review themes forward. The chain is the product.

The Local SEO Starter is the entry bundle, built to be useful on day one for a single location. When the clinic wants content at scale or wants to prove the work is paying off, that is where the larger bundles come in, and the same structured-handoff model carries straight through.