Case studies · lionelz.com

Technical SEO and visibility in AI search engines, with Astro and Claude

My own site as a testbed. From WordPress to Astro, from near-zero to inbound calls.

Project
lionelz.com (my own site)
Sector
Freelance PPC and CRO consulting
Goal
Technical SEO, content and AI visibility (GEO)
Stack
Astro v5 (static) + Claude as assistant
Languages
Spanish, English, French
Period
Early 2026 · ongoing

The case in 30 seconds

lionelz.com is my personal site and the brand for my freelance PPC and CRO consulting. It was running on a WordPress setup more complex than I needed, I had stopped publishing, and it had barely any traffic or leads. In early 2026 I rebuilt it on Astro v5 (static) and started using Claude as an assistant for technical SEO, new content, updating old content and internal linking, publishing again in three languages. In the first three months of Search Console data (since 16 March 2026) it accumulated around 400 clicks and 319,000 impressions, with average position around 9.7. CTR on Google is low (~0.1%), consistent with the shift toward zero-click searches; the real conversions come as inbound calls with prospective clients in Europe, the US and Asia, and several mention having arrived through AI platforms (ChatGPT, Perplexity, Gemini). It's an experiment on my own site, not a client account: small sample and short period, but the direction is clear.

Signals from the experiment · 2026
0 → 319k
Search Console impressions in ~3 months, starting from near-zero
99–100
Lighthouse performance on mobile and desktop (static Astro)
AI
Several leads cite ChatGPT, Perplexity or Gemini as their discovery channel
The traffic and performance figures come from Search Console and PageSpeed Insights. The AI attribution is what the contacts themselves report: it's my own site, over a short period and with a small volume. What matters here is the change of direction.

The starting point

The site was on WordPress, with the usual stack of plugins. For a personal site it was a more complex setup than necessary. I had started publishing in three languages back around 2020, but when I no longer needed extra clients, that publishing habit faded until it stopped.

In early 2026 I read quite a bit about the results Claude was giving a lot of people (development, SEO, operations). After years of watching marketing fads come and go, I tend to take this kind of enthusiasm with a pinch of salt. But I wanted to try it firsthand.

When I started, the site had close to zero real organic traffic; the little that moved in analytics were spikes from AI bots crawling the site. I had set it aside and no longer worried about it. That made it, precisely, the perfect site to experiment with: no real risk. And, above all, it generated almost no leads.

Why I rebuilt the site on Astro

After accepting that my WordPress was probably not the ideal foundation to compete well on performance and results, I decided to go all in with Claude and use it as an assistant. We started by choosing a simple, flexible-enough technical stack. I have a developer background, so I didn't need a back office or anything sophisticated to manage the site and the content.

The full stack essentially became Astro v5: static output, fast, without the plugin overhead, easy to version-control and edit directly. That simplicity shows in performance: on Lighthouse the site scores 99–100 on performance, 100 on SEO and best practices, and 95 on accessibility, on both mobile and desktop.

Lighthouse · PageSpeed Insights Scores on mobile and desktop · agentic navigation 2/3 Mobile Desktop 99 95 100 100 100 95 100 100 Performance Accessibility Best practices SEO

For a content site in three languages, that simplicity is exactly what reduces friction and lets the SEO and content work be what leads, not the infrastructure.

Claude as an SEO assistant

For the analysis side I relied on claude-seo, an open source SEO analysis plugin for Claude Code created by Daniel, easy to connect to the workflow. From there, the site started to get, on a regular basis, new content, updated old content, improved internal linking and more.

The working pattern is consistent: short, repeated cycles rather than one big one-off overhaul. Writing and translating into the three languages, reviewing what's published, spotting broken links and content gaps, and starting again. It's exactly the kind of repetitive task where an assistant like Claude provides real leverage.

Why SEO isn't what it used to be

There's an underlying shift that conditions everything above: being in Google's number 1 spot has stopped guaranteeing free traffic. AI Overviews take up more and more space on the results page, and a growing share of searches ends without a click, because the user gets the answer without leaving the SERP. SparkToro and Datos' work on zero-click searches documents this trend well.

On top of that, more and more people use AI platforms like ChatGPT, Perplexity or Gemini directly as a search engine. And a result obtained through an AI carries, in many people's eyes, a high recommendation value: it's not an ad or just any link, it's an answer.

That makes SEO more complex and shifts the weight toward other signals: demonstrable experience and authority (E-E-A-T), well-placed structured data (Schema.org), entity clarity about who you are and what you do, and content written so an AI can cite it in specific passages. Optimizing to get cited by AI (what's starting to be called GEO) is a new layer on top of classic SEO, not a replacement.

What has happened since

Search Console started recording data for the site on 16 March 2026. Over the following three months, impressions and clicks grow steadily from near-zero, with average position stabilized around 9.7.

Google Search Console — organic performance 16 Mar – 7 Jun 2026 · since Search Console was connected 400 Clicks 319k Impressions 0.1% Avg CTR 9.7 Avg position Clicks Impressions 16/03 26/04 07/06

These numbers deserve an honest read. CTR on Google is low, around 0.1%: many impressions stay as visibility without a click, exactly what you'd expect in an AI Overviews and zero-click scenario. The real value isn't so much in those clicks as in what doesn't show up in Search Console: from that base, the site has gone on to generate inbound calls on a regular basis from prospective clients in Europe, the US and Asia.

The detail I'm most interested in: several of those contacts mention having been referred by an AI platform, for very specific tasks related to their project. It's not a massive volume, nor is it meant to be. It's a personal site that was switched off and that, with a better technical foundation and sustained content work, has become a discovery channel again, including inside the AI tools where many people now start their search.

Notes on scope and what's not shown

This case is an experiment on my own site, not a client account. The traffic figures come from Google Search Console and the performance figures from PageSpeed Insights; the charts on this page reproduce that real data. The "an AI recommended me" attribution comes from what the contacts themselves report, not from a closed tracking system, and the period is short with a small lead volume. That's why I present it with caution: the direction is clear, but it's early to draw definitive conclusions.

The experiment continues; this case reflects a snapshot from mid-2026.

Free first call

Could your ad campaigns
perform better?

30 minutes to review your situation and tell you exactly what I would change. No pitch, no sales proposal.

Book a call →
30 min · Google Meet · No commitment