Claude Code SEO: what it found on my hobby site

16 slides · use ← → to step through

Claude Code SEO

What it found on
my hobby site

Free and self-hosted, with OpenSEO. And Umami as the cherry.

Notes

Intro: I let Claude Code do the SEO for one of my hobby sites, using OpenSEO, a free open-source alternative to Semrush and Ahrefs that runs on my own server. First what it found and what changed, then how the setup works, a demo of my dashboard, and how it compares to Ubersuggest (Neil Patel), Ahrefs and Semrush. Visitor numbers come from Umami, my self-hosted analytics.

The hobby site

A hobby site with a live map

10k

visitors a month

2/3

on a phone

$1

of data for everything

Notes

My test case: a hobby site with a live map of A380 routes. About 10,000 visitors in 30 days, 10% more than the period before (visitor numbers from Umami, my self-hosted analytics). 60% come via Google, plus Bing, DuckDuckGo and Reddit; mostly from the US, the UK, Germany and Australia. Two thirds are on a phone. 69% look at only one page. Everything on the next slides, all the keyword research, audits and backlink checks, cost about one dollar of DataForSEO credit.

SEO is a library

Google is the librarian.

Keywords

What people ask the librarian

Backlinks

Other books recommending yours

Ranking

Which shelf your book ends up on

Notes

Quick metaphor before the tools. Google is a giant library and the librarian decides which book you get handed first. Keywords are the questions people ask at the desk. Backlinks are other books that point to yours, so the librarian trusts it more. Ranking is the shelf: eye level gets read, the bottom shelf on page 2 never does. SEO tools tell you what people ask, who recommends whom, and where you are on the shelf.

What Claude found

The problem

The fix

Google can't read an interactive map

Added plain text pages

Google sent people to the wrong page

One clear title per page

Seen in Google, rarely clicked

Answer the question right away

Most visitors are on a phone

Fixed mobile first

Notes

Claude Code pulled the data from OpenSEO and Search Console, I looked at it together with Claude, and Claude made the changes. Left the problem, right the fix. 1) Google can't read an interactive map: a map drawn by code is invisible to Google, like a painting in a dark room. Google reads text. So I added normal pages with text, plus a map as a plain image. 2) Google sent people to the wrong page: searches about one airline landed on my homepage instead of that airline's page, and hardly anyone clicked. Fix: every page gets a title that says exactly what is on it, and clear links between pages. 3) Seen in Google, rarely clicked: popular questions showed my site in the results, but people didn't click. A shop window everybody walks past. Fix: one page that answers the question right at the top. 4) Two thirds of visitors are on a phone, and that is where the site was worst. So mobile came first.

And a few surprises

A forgotten page

A side page I never optimised is one of the most visited.

ChatGPT knows me

It cites my site as a source. Perplexity sticks to big media.

Notes

1) Traffic I didn't expect: one of my biggest entry pages is a side page I never did any SEO work on. Check which pages people really land on; it is rarely the ones you think. 2) ChatGPT already names the site as one of its sources for a common question in my niche. Perplexity doesn't mention it; it mostly cites well-known media. 3) Almost all the sites linking to me are spam, like junk link directories. Competitors get their links from blogs and forums in the niche, so that is where the chances are: ask those sites. 4) The tools have pitfalls: the rank tracker once said "not found" because of an error at the data provider, and a one-off check in Google put me on page 2 while Search Console averaged a top-3 position. Lesson: trust Search Console over one-off measurements.

Before and after the click

The shop window, and the shop.

OpenSEO

Where you show up in Google, and who clicks

Umami

Where visitors land, and whether they stay

Google sent people to my homepage. Umami showed most left after one page. Together: they couldn't find what they came for.

Notes

Each tool measures half the story. OpenSEO with Search Console is before the click: where you appear in Google, for which searches, at which position, and how often people click. Umami is after the click: where visitors land, on which device, and whether they click on or leave. What the combination showed: 1) Mobile: both showed about two thirds of visitors on a phone. Looking at the mobile version turned up four real bugs. 2) The homepage gets the clicks, then it stops: Search Console showed searches about specific airlines landing on the homepage, Umami showed most visitors view only one page. Together: people don't find the page they were looking for. Hence links under the map and inside the map. 3) Traffic SEO didn't see coming: Umami showed a side page as a top entry page, while the SEO tools said nothing about it. 4) AI as a source: OpenSEO showed ChatGPT names the site, and Umami shows the first visitors arriving from AI tools like Gemini. Next: Umami now counts clicks on the new links under the map. In a few weeks: Search Console shows people coming in, Umami shows whether they click on. A before-and-after for a follow-up video. Note: OpenSEO can combine Search Console with analytics, but only with Google Analytics, not Umami, so Claude combines the two itself.

Two steps up

Google impressions per day, from summer to now.

Flat all summer ×2 Sep 2 · text pages + sitemap +50% Oct 13 · OpenSEO round Jul Aug Sep Oct Nov

Clicks: ×2.5 since summer, and still catching up.

Notes

From Search Console, matched against the dates in the site's git history. Summer: completely flat. Step 1, September 2: Google impressions more than doubled, from about 1,400 to about 3,000 a day within a week. That is when the route lists became plain text, with pages per airline and per airport and a sitemap. Before that, Google only saw a map with nothing readable. Step 2, October 12–13: another 50%, from about 3,300 to about 4,700 a day, after the first round with OpenSEO: robots.txt with the sitemap, IndexNow, stronger airline pages, the fleet count, FAQ markup and share images. Clicks went from about 85 to about 210 a day since the summer, two and a half times. They rise more slowly than impressions because many of the new impressions are on positions 5 to 10, the bottom of page 1. That is what this week's work is for: titles with the exact search phrase, an operators page and links to the right pages. Honest caveat: this is correlation, not proof. But the months before were flat with no seasonal effect, and the steps land exactly on the days of the changes. The chart is drawn from these rounded levels, not from the raw daily data.

OpenSEO: the same toolbox, open source

Free, you run it yourself. Say you have a bike shop:

Keywords

How many people search “bike repair”?

Rankings

Am I on page 1 for it?

Competitors

Which searches bring the other shop its visitors?

Backlinks

Which sites link to my shop?

Site audits

Broken links, missing titles, slow pages

AI visibility

Does ChatGPT recommend my shop?

Notes

github.com/every-app/open-seo, MIT licence. An open-source alternative to Semrush and Ahrefs. Walk through it with one example, a local bike shop: how many people search "bike repair", whether you show up on page 1 for it, which searches bring the shop down the road its visitors, which sites link to you, technical problems like broken links, and whether ChatGPT and co. mention you. You self-host it or use their hosted version. It also has an MCP server, so Claude Code can query the SEO data directly.

Guest or owner

The same hotel as in my last talk, Server security, simply →

Ahrefs, Semrush, Ubersuggest

A guest in their hotel

  • Their rules
  • Their prices
  • Check out, lose it all

OpenSEO, self-hosted

Your own hotel

  • Your rules
  • Your data
  • Rebuild any room
Notes

With Ahrefs, Semrush or Ubersuggest you are a guest: you use the rooms they built, at the price they set, and when you check out your history and reports stay behind. Self-hosted, you own the building. The data sits on your server, nobody changes the plan under you, and with Claude you can rebuild any room: add a report, change a screen, connect it to your other tools.

What owning it gets me

Claude-ready

Claude asks it questions directly

Hackable

I build my own reports on top

Notes

Private: keyword lists, competitors and rankings never leave my server; it sits behind Tailscale. Claude-ready: OpenSEO has an MCP server, so Claude Code can look up keywords and backlinks while it works on my sites. Hackable: it is open source (MIT), so Claude can change the code; my weekly SEO report for 11 sites, every Monday, is built on top of it.

Bonus: it costs almost nothing

You only pay for the data you use.

$0

the app

4¢

one site check

$1.70

my month, 11 sites

Data from DataForSEO, prepaid. Ahrefs starts at $129 a month.

Notes

Self-hosted the app is free; you pay DataForSEO directly per request, no markup. DataForSEO is prepaid: you top up a balance and each call is deducted. The hosted version is $10 a month plus 28% on each request. For comparison: Ubersuggest $29, Ahrefs $129, Semrush $139.95 a month. My weekly report checks domain rank, backlinks and keywords for 11 sites every Monday: about 4 cents per site per run, about $1.70 a month. Ad-hoc keyword lookups are a few cents each. Total spend so far: about $5.50 in five weeks.

Demo

My dashboard

Notes

In the video: a keyword lookup for one of my sites, the backlinks of a competitor, the weekly SEO table next to the analytics overview, and Claude Code asking OpenSEO a question over its MCP connection.

Not on the menu

Own crawler

Years of link history for any site

Content tools

A score for your text before you publish

Ads research

See the Google Ads a competitor runs

Client reports

A monthly PDF with your logo on it

Browser plugin

SEO numbers right inside Google

Support

A helpdesk and video courses

Missing a dish? Ask Claude to cook it.

Notes

What the restaurant has that your kitchen doesn't, and when you actually need it. Short version: you need these when SEO is your job (clients, a marketing team, ads), rarely when you build a few sites of your own.

Own crawler: Ahrefs and Semrush own the farm. They crawl the web themselves, keep years of history and have their own scores (Domain Rating, Authority Score). OpenSEO buys its ingredients from a wholesaler, DataForSEO. You need the farm for serious link building: deep backlink audits, or seeing how a competitor's links grew over the years. For "who links to me", DataForSEO is enough.

Content tools: they score your text against the pages that rank now: missing words, length. Useful when you have lots of articles written, especially by writers who don't know SEO. With Claude you can do this yourself.

Ads research: which Google Ads keywords competitors bid on, their ad copy and an estimated budget. Only needed if you run Google Ads yourself; for organic traffic you don't.

Client reports: a monthly PDF with your own logo, sent automatically, with logins for clients and colleagues. Needed if you run an SEO agency.

Browser plugin: a bar in Chrome that shows the numbers next to every Google result and site. Handy for quick checks, never essential.

Support: a helpdesk plus courses like Semrush Academy and Ahrefs Academy. Useful if you are new to SEO, or in a team with nobody technical.

And the data is often already in DataForSEO, so Claude can build the missing piece. Check the list before recording: features change fast on all four tools.

One country at a time

A local newspaper, not a world map.

320

OpenSEO estimate, US only

5,500

Search Console, worldwide

Your own site: Search Console. Competitors and new keywords: OpenSEO.

Ahrefs and Semrush can add up all countries. Every tool's traffic number is an estimate.

Notes

A surprise I ran into. Every OpenSEO project has one location, and the default is the United States. My site's project was on that default, so the 320 is only the estimated Google traffic from the US, and without the long tail. Search Console showed 5,500, worldwide, real clicks. OpenSEO can't give one worldwide number because DataForSEO works per country. If you want other countries, make a project per country, for example the UK or Australia; each lookup costs a few cents of DataForSEO credit. How the others do it: Ahrefs and Semrush also keep their data per country, but they can add all countries up into one worldwide number and show a global search volume next to the local one. Ubersuggest also works per country . And all of them estimate traffic from rankings; none of them sees your real visitors. Lesson: for your own site, Search Console is the source of truth. Use SEO tools for competitors and for finding new keywords.

Which one is for you

You want to own your tools

OpenSEO, self-hosted

You want zero hosting

OpenSEO hosted, or Ubersuggest

You run an SEO agency

Ahrefs or Semrush save you time

Notes

Cook at home, order in, or go to the restaurant. For most indie hackers and solo founders the first row is you.

Why agencies still pick Ahrefs or Semrush: strictly you don't need them for client reports, I built my own weekly report on OpenSEO. But for an agency: the reports are ready-made, nothing to build or maintain, which matters with 30 clients. Clients know the numbers and often ask for their "Domain Rating" or "Authority Score", which are Ahrefs' and Semrush's own scores and can't be rebuilt. You can look back years ("last year you were here") only if you have collected data that long. And they are built for teams: logins per colleague and per client.

Point your claude to:

github.com/every-app/open-seo

Notes

Call to action: open Claude, paste the repo link and ask it to deploy OpenSEO on your server. Add a DataForSEO key and look up one keyword for your own site. Tell me in the comments what you would build on top of it.