How Staxly Went from 10/100 to Improving Its Citation Score (Real Case Study)

Most GEO advice is theoretical. This isn't. What follows is a documented account of how Staxly.dev — a real product — was misrepresented by AI models, ran a structured audit, applied specific fixes, and measured what changed.

What Is Staxly?

Staxly is a catalog of 137+ developer platforms — BaaS providers, databases, auth services, hosting platforms, AI APIs, payment processors, CI/CD tools, and more. Each platform has structured pricing data, feature lists, and quota breakdowns from vendor documentation. The kind of site a developer bookmarks when comparing Supabase vs PlanetScale vs Neon before making a decision.

Exactly the kind of comparison question people are increasingly asking AI assistants. Which is why it mattered that ChatGPT had no idea what Staxly was.

The Diagnosis: A Score of 10/100

Across 15 test queries — "best tool to compare developer platform pricing," "catalog of BaaS platforms," "compare Supabase vs Firebase pricing," and direct brand queries like "what is Staxly?" — 12 out of 15 either returned no mention of Staxly or returned actively wrong information.

The worst failure: ChatGPT consistently identified "Staxly" as a Minecraft proxy server. Not a developer tool. Not a platform catalog. A Minecraft proxy. Complete brand hallucination, driven by noise from gaming Discord servers and forums.

What Did the Audit Find?

No Brand Disambiguation

Staxly's content never explicitly stated what it was. LLMs, lacking a clear signal, defaulted to the noisiest source: gaming communities where an unrelated "Staxly" project existed.

No Schema.org Markup

Zero JSON-LD structured data. No Organization schema, no WebSite schema, no SoftwareApplicationmarkup. Nothing told a machine "this is a website about developer tools."

English-Only Content

Despite having Spanish-speaking users (including its creator, based in Colombia), all content was English-only. Queries in Spanish returned nothing — the site didn't exist in that language context for AI models.

No Citable Sentences

The content had excellent data (pricing tables, quotas) but poor citable prose. LLMs don't cite table cells — they cite sentences. "Supabase Free: 500 MB" in a table isn't citable, but "Supabase offers a free plan with 500 MB of storage and 50,000 monthly active users" is.

The Fixes Applied

None were architectural overhauls. Targeted content and metadata changes, applied over a few days:

1. Bilingual content: Key pages published in both English and Spanish. A helper function generates tier summaries like "Free gratuito, Pro desde $25/mes, Team desde $599/mes."
2. Organization + WebSite JSON-LD: Structured data identifying Staxly as an organization with inLanguage: ["en", "es"] and explicit disambiguatingDescription.
3. Explicit disambiguation: The phrase "Staxly is not affiliated with Stax Payments, OpenStax, STAX headphones, or Staxel" added in visible text and schema.org.
4. 29 category landing pages: Each tool category (BaaS, databases, auth, hosting, etc.) got a dedicated page with bilingual H1, comparison tables, and ItemList JSON-LD.
5. Long-form guides: 1,760-word guide on best BaaS platforms for startups, with real pricing data from the database and FAQPage schema.

The Result

The re-audit showed a score improvement from 10 to 13 out of 100. Context:

• LLMs don't update in real time — changes take weeks to reflect in responses.
• The Minecraft hallucination disappeared. ChatGPT now identifies Staxly correctly as a developer tools catalog.
• Spanish queries started workingwhere they'd previously returned zero results.
• Citation score noise is ~10 points — the real delta is measured in weeks, not days.

Lessons for Your Site

1. If your brand name is ambiguous, disambiguate. Explicitly. In visible text and in schema.org. LLMs hallucinate when they can't find a clear definition.
2. Schema.org is table stakes for GEO.It's the most direct way to tell an LLM what you are, what you do, and how to categorize you.
3. Language coverage is a competitive moat. Most GEO optimization is English-only. Publishing in other languages gives you a significant advantage in less competitive contexts.
4. Hard data wins. Exact prices, specific quotas, update dates. Not vague marketing — data an LLM can verify and cite.
5. Measure. Iterate. Repeat.A single audit isn't enough. LLMs change, competitors optimize, content ages. Continuous monitoring separates visible sites from invisible ones.

Want to know your site's citation score? Run a free audit on AICite — 60 seconds, no signup, public and shareable results.