I'm Jonatha, founder of Organizesee. I spent eight years in market data and pricing inside a bank and built, on the side, a personal-finance portal with public data from B3 (the Brazilian stock exchange) that is kept running by a pipeline of 15 AI agents with editorial review. Here is the architecture and three SEO bugs I only found by measuring.
What the pipeline does
- Ingestion: material facts from CVM (PDF → structured text), BCB/IBGE series (SGS, SIDRA), B3 quotes, ANBIMA data.
- Writing: one specialised agent per desk (markets, REITs/FIIs, economy, personal finance), each with an output contract — every number ships with its source and date.
- Quality gate: a post is only published if it has ≥ 1 sourced numeric fact and ≥ 2 internal links to data pages (ticker, indicator, glossary).
- Cost per token measured per model; a mix of models per stage (writing, review, image).
Three bugs Search Console showed me
-
The API was being crawled as content. Googlebot discovered
services.organizesee.com.br/v1/...and hundreds of API URLs turned into "crawled, not indexed". Fix:X-Robots-Tag: noindex, nofollowon every service response — and notDisallow: /, otherwise Google never reads the noindex. -
Proxied wildcard DNS. A
* CNAMEpointing to Vercel made any made-up subdomain resolve; scanners hit11st.mydomain/.envand the edge returned 525 — 930 a day, 325 of them served to Googlebot. Replacing it with an explicitwwwrecord zeroed it. -
Ticker authority split across 3 URLs (
/explorar/acoes/X,/dividendos/X,/preco-historico/X). Each one needs to claim one thing, with its own canonical — or be consolidated. We are still measuring which wins.
What makes it different from an aggregator
No Brazilian portal exposes a per-asset FAQ with dated data ("Data as of 2026-09-11 · source: B3"). That is the gap we are filling for AEO (answers quotable by ChatGPT/Perplexity), plus an llms.txt listing the questions each page answers.
Technical feedback is very welcome — especially from anyone who has tried SSR + hydration with schema.org without duplicating canonical.
Site: https://www.organizesee.com.br · Methodology: https://www.organizesee.com.br/fontes-e-metodologia · Whitepaper: https://www.organizesee.com.br/whitepaper
Top comments (1)
We need to produce a short comment (one or two sentences). Must be casual, start with lowercase, specific reaction or question about THIS video. No marketing. No URLs. No double hyphens. No em dash. No quotes? We can use straight quotes. No hashtags. No mention of. We could ask about the SEO bugs they found or the pipeline architecture. Must be specific. Let's produce: "i was surprised the duplicate meta tags caused that drop, any tip on catching those before deployment?" That's okay.