how an "impossible" seo bug turned into a 20-minute win with claude code
the structured data issue that kept our site health stuck at 92%, how swapping models fixed it fast, and a founder-ready seo checklist.
share this post
stay in the loop
Get fresh Prism notes via email
When we publish new experiments or playbooks, we’ll send you the highlights so you can apply them faster.
Key takeaways
the real problem: a structured data bug that would not go away
why structured data matters more than most founders realize
enter claude code (opus 4.5): $20 well spent
what this means for you as a founder (not a developer)
Was this post helpful?
Your feedback helps us improve how we deliver practical playbooks.
Free to start
See your biggest growth opportunities — free.
Create your free Prism Growth Dashboard and get a focused Light Audit of what is holding your website, search, and local presence back. No card required.
Search behavior is changing, but the fundamentals remain: clearer intent, stronger proof, and consistent technical quality. Prism operationalizes that into predictable results.
What this improves
Build a stronger SEO base with cleaner indexing and AI discoverability.
Turn search visibility into booked calls and qualified leads.
Scale content and local trust updates without losing consistency.
How we apply this at Prism
SEO servicesComprehensive local and organic visibility support.
Local SEO servicesListings, maps, and trust layer upgrades for local demand.
AI SEO servicesAI visibility execution for search and assistant surfaces.
We hit a structured data bug that would not die. Multiple fixes, multiple Semrush crawls, and site health stayed around 92%. One model switch turned it into a 20-minute win. Here is the story, the lesson, and a checklist you can use.
Video breakdown: the live walkthrough sits below. This post adds context, lessons, and the exact playbook.
quick background: what prism actually does for founders
I run Prism, a web studio for founders who care about growth, not just pretty pixels. Every project ladders up to three levers:
increasing leads (more qualified people discovering you)
improving conversion rate (more of those people becoming customers)
increasing lifetime value (customers sticking around and spending more)
Technical SEO (opens in a new tab) sits in the middle. If your site is hard for Google to understand, you cap all three.
the real problem: a structured data bug that would not go away
We were working across a 120-page site (our own plus some client work). The loop was simple:
run a Semrush site audit
fix errors
re-run the audit
watch site health climb closer to 100%
Except it did not. Health stuck near 92%. A handful of structured data errors refused to clear. Every time we shipped a fix and waited another 20-minute crawl, the same issues returned.
After a couple of days, it was obvious we needed a second opinion from another model.
why structured data matters more than most founders realize
Structured data (schema) tells Google exactly what is on a page:
"this is a product with a price and reviews."
"this is an article with an author and date."
"this is a local business at this address with these hours."
When it is implemented correctly, you get richer search (opens in a new tab) results, better eligibility for AI search (opens in a new tab), and a clearer picture of your business. When it is wrong, you get warnings, missed visibility, and less trust from Google - real revenue left on the table for a 120-page site.
enter claude code (opus 4.5): $20 well spent
I had hesitated to add Claude Code (opens in a new tab)'s Pro plan because we were already spending about $200 a month on OpenAI and had paid for Anthropic (opens in a new tab) and Gemini before. But this bug overstayed its welcome, so we tested Claude Code for roughly $20.
We fed Claude the exact same inputs we had given Codeex - Semrush error details, code snippets, and schema context. Claude spotted the underlying mistake, suggested a fix that matched the spec, and we implemented it.
We ran another Semrush audit:
the bug disappeared on the first try
site health jumped
no more recurring structured data errors
The takeaway is not that one vendor "won" forever. Being dogmatic about a single model wasted days; being flexible solved it in one shot.
what this means for you as a founder (not a developer)
1) keep a multi-model mindset
The AI landscape moves too fast to bet on one vendor. Your team should switch between OpenAI, Anthropic, Gemini, and whatever works best for the problem. Treat models like interchangeable power tools, not religions. If an agency insists one tool is "best at everything," that is a red flag.
2) "we use ai" is meaningless - how you use it matters
Codeex gave plausible fixes that passed its own reasoning but failed reality. Claude gave a fix that matched the schema spec and Semrush's behavior. Good AI workflows stay grounded in real-world feedback (Semrush, Search Console, analytics), generate options, and test until the bug is actually gone.
3) technical seo stops being optional once you scale
A five-page brochure site can get away with a lot. A 50-200+ page site that should feed you leads daily is a different game. You need clean site health scores, structured data coverage, and crawl/indexability dialed in - especially if you are paying to send people to the site.
a simple checklist: is it time to get help with your site health?
Answer honestly:
do you have 20+ pages (blog, product, locations, case studies)?
do you rely on the site for leads, demos, or sales?
do you skip Semrush/Ahrefs/Search Console checks for months at a time?
has your developer or agency ever explained structured data and site health?
when they did, did it sound like jargon or like they actually knew the system?
If you are nodding along, you likely have hidden technical issues costing you money every month.
how we approach this at prism
baseline audit: run a full crawl, check site health, structured data, speed, and core technical issues.
prioritized fix list: rank by revenue impact, not what looks cool in a report. Explain each issue in plain English.
ai-powered implementation: use multiple AI coding tools (Codeex, Claude Code, Gemini, etc.) and validate fixes against Semrush and Search Console, not just model output.
growth alignment: tie every fix back to leads, conversion, and LTV so the site is built to generate business.
ongoing iteration: as models improve, we keep testing them on real problems so you do not have to care which tool we use - only that the site keeps improving.
why we share this stuff publicly
This work is not magic. The video shows the exact issue, tools, and outcome in real time so you can see what is happening under the hood and the level of care you should expect from whoever owns your site.
if you want your website to actually pull its weight
If you are a founder with a site that should be producing leads - and you suspect hidden technical debt is holding it back - let us handle the ugly parts.
We will run a site health and structured data audit, fix the high-impact issues, and turn your website into something that supports your growth targets. Worst case, you get clarity on what is slowing you down. Best case, we unlock traffic and revenue you are currently leaving on the table.
