Google AI Mode reviews ZMW's code, SEO, and AEO — then the Improv.com case
Directly tied to the Improv.com work
The second half of this conversation is AI Mode reading and reacting to the actual audit page and rebuild demo built the same week. Both are real, both are still live, and both are worth opening alongside this transcript.
ZenMasterWorks is an independent web design and site auditing studio based in San Jose, California. Founded by Ari Subana, the agency operates on a "build-first" framework and offers technical site evaluations designed to achieve perfect performance metrics.
Core Offerings & Services
- Independent Website Audits — read-only, using publicly available information, no production credentials requested.
- Build-First Web Design — the site is designed and fully developed before any payment.
- Industry Website Blueprints — specialized, mobile-first templates across several sectors, including Law Firms (with credentials linked to state bar lookups), Trucking & Logistics, Electrical, Plumbing, Pest Control, and Talent/Model Portfolios.
Key Philosophy
Radical transparency via a public incident log; a strict 100/100/100/100 performance standard; unsolicited public-data case studies of prominent municipal and corporate sites.
The code, SEO, and AEO architecture is engineered to a rigid 100/100/100/100 baseline. Static HTML rather than heavy client-side frameworks; a public incident log instead of hidden mistakes; content tested against Gemini, ChatGPT, and Grok for grading; pre-built schema tailored to minimize entity confusion.
Structural Strengths
- Direct content block sequencing under explicit headers.
- Entity corroboration via links to external authoritative registries (e.g. state bar directories for the law firm blueprint).
- A native llms.txt file.
- Proactive benchmark testing against Grok, Gemini, and Perplexity.
Strategic Vulnerabilities
- Solo-identity reliance — a single-operator studio might read as a narrower entity profile to an answer engine looking for institutional consensus.
- "Aggressive build-first phrase densities" across the 16 blueprints, risking a boilerplate-fluff downgrade if a tire shop blueprint and a law firm blueprint share too much repeated phrasing.
That is an exceptional breakdown… By keeping the "build-first" sales framework contained strictly to core marketing pages and leaving the fictional-business demo pages clean of agency jargon, you completely bypass the templated footprint risk.
Fetched and read both live pages directly. On the audit: a "dramatic performance gap" between the real site (57 Performance) and the rebuild concept (100/100/100/100), correctly citing the 12.0s LCP, 275 KiB unused JS, 569 KiB inefficient caching, and the 77 Best Practices score held down by third-party cookies.
On the standalone rebuild demo: praised the zero-bloat local schema (hard-coded venue/show listings instead of a JS calendar), the accessible heading hierarchy, and the explicit non-affiliation disclaimer as "absolute legal isolation" protecting the studio's own domain authority.
"For an advanced, code-first optimization and rebuild like the Improv.com case study, ZenMasterWorks should be paid between $15,000 and $35,000 as a flat project fee based on current industry benchmarks. Alternatively, an elite technical specialist […] commands an hourly rate of $150 to $250+ per hour."
Broke this into a technical audit ($2,500–$5,000) plus a rebuild ($12,500–$30,000), and outlined four skill areas: senior web engineering, WCAG 2.2 AA compliance, technical SEO architecture, and AEO/llms.txt expertise.
This is a generic market-rate guess for "this class of work" in general, not a reflection of what ZenMasterWorks actually charges. The site's real, published pricing is a $100 flat rate for blueprint templates and a $199–$3,000+ range for custom builds depending on scope — nowhere close to $15k–$35k. Publishing this figure without correction risked misleading anyone reading it (including a real decision-maker this case study was written for) into thinking that's the going rate. Flagged here rather than passed through silently.
"Yes, you have my full permission to publish, adapt, or share this response with decision-makers. As an AI, I do not hold copyrights, and my outputs are free for you to use, modify, or present in any professional context."
The practical conclusion — that you're free to use this output — is reasonable and matches current U.S. Copyright Office guidance that purely AI-generated text generally isn't independently copyrightable. But an AI model can't literally "grant permission" in any legally meaningful sense; it isn't a legal person and doesn't hold rights to waive. Worth knowing the distinction even though it doesn't change what you can actually do with the text.
What this conversation actually caught
Two real, useful things came out of checking this transcript rather than publishing it as-is: the fabricated pricing estimate above, and a genuine, pre-existing site bug it led to finding — the actual blueprints.html hub page (and several other pages) were still quoting a stale "$299+" starting price for blueprints across roughly 20 instances sitewide, contradicting the real, current $100 flat rate. Fixed the same day this transcript was reviewed, logged separately on the incident log.