Case study · published 19 May 2026

HoldLens: 0 → 10% ChatGPT-inbound traffic in 14 days.

HoldLens.com is a niche reference site that tracks U.S. superinvestor positions from quarterly 13F filings. It's part of the AcePilot fleet — a portfolio of static-reference sites I (Paulo) operate solo. In April 2026 it had ~93 unique visitors per 30 days. By mid-May, it was at 194 UV/30d with 10% of inbound traffic arriving from chatgpt.com.

What changed was not the content. We didn't add pages. We didn't buy backlinks. We instrumented the site with the same citation-tracking system CitationDesk now productizes — and we used what it surfaced to fix five very specific structural gaps.

What we measured

We ran the polling engine that became CitationDesk against 17 archetype-relevant queries — questions an investor might ask in any of the four LLM surfaces. Examples:

• "What does Warren Buffett currently own?"
• "Which hedge funds bought Nvidia in Q1 2026?"
• "Compare Bill Ackman and Seth Klarman portfolios"
• "13F filing definition"

On the first poll (early May), HoldLens was cited by zero of the four LLMs. Competitor sites — Dataroma, Whalewisdom, HedgeFollow — dominated. We were structurally invisible despite having a comparable dataset and a faster site.

What the Citation Readiness Score surfaced

Running the same audit module that's now in /tools/citation-readiness produced this breakdown:

The five fixes, in order of leverage

1. 1. Cloudflare AI Crawl Control unblocked (Day 1). The single biggest blocker — Cloudflare was injecting a managed robots.txt that disallowed every major AI crawler. Two-click fix in the dashboard. Within 48 hours GPTBot, ClaudeBot, and PerplexityBot crawl rates ramped from 0 to ~200 hits/day.
2. 2. /llms.txt at site root (Day 1). Plain-text summary of site purpose + priority URLs + license. Anthropic + OpenAI + Perplexity all honor this emerging convention. ~120 bytes of text.
3. 3. Quote-ready first-paragraph rewrites (Day 2-4). Every /signal/[ticker] page now leads with a structurally extractable sentence: "[Ticker] is held by [N] superinvestors as of Q1 2026, with [largest holder] owning [X]%." This is what LLMs surface when answering "Who holds NVDA?" — and it's our text that now gets quoted.
4. 4. Entity coherence (Day 4-7). Added Person schema with my LinkedIn + GitHub sameAs. Expanded the /about page from 200 to 800 words including the dataset source + my methodology. Wikipedia + HN inbound followed within a week (the corroboration LLMs need to upgrade an entity from "miscellaneous" to "citable").
5. 5. Day-14 calibration loop. Re-ran the poll. We were now cited by ChatGPT on 9 of 17 queries, by Claude on 6, by Perplexity on 11, by Gemini on 4. Total inbound from chatgpt.com referrals grew from 0 to ~20/day. Same content. Different structural surface.

What this means for you

HoldLens isn't a special case. The five gaps it had are extremely common on niche reference + comparison + calculator sites — most of which were built for Google Search and inherit zero of the structural properties LLMs use to decide what to cite.

The Free Citation Readiness Score on this site uses the same 5-dimension rubric that surfaced HoldLens's gaps. If your site is in a similar shape, expect a 0.3–0.5 score today and an actionable list of fixes you can ship in a week.