The latest articles on DEV Community by Jonatha Oliveira (@jonatha_oliveira_8f9f37b4). https://dev.to/jonatha_oliveira_8f9f37b4 https://media2.dev.to/dynamic/image/width=90,height=90,fit=cover,gravity=auto,format=auto/https:%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Fuser%2Fprofile_image%2F4124709%2Fab698ee3-fa32-4679-9319-2c2abd8df609.png https://dev.to/jonatha_oliveira_8f9f37b4 en Jonatha Oliveira Mon, 14 Sep 2026 14:24:14 +0000 https://dev.to/jonatha_oliveira_8f9f37b4/15-ai-agents-run-a-brazilian-finance-portal-every-day-heres-the-pipeline-and-the-3-seo-bugs-it-1b9g https://dev.to/jonatha_oliveira_8f9f37b4/15-ai-agents-run-a-brazilian-finance-portal-every-day-heres-the-pipeline-and-the-3-seo-bugs-it-1b9g <p>I'm Jonatha, founder of <a href="https://www.organizesee.com.br" rel="noopener noreferrer">Organizesee</a>. 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.</p> <h2> What the pipeline does </h2> <ul> <li> <strong>Ingestion:</strong> material facts from CVM (PDF → structured text), BCB/IBGE series (SGS, SIDRA), B3 quotes, ANBIMA data.</li> <li> <strong>Writing:</strong> one specialised agent per desk (markets, REITs/FIIs, economy, personal finance), each with an output contract — every number ships with its source and date.</li> <li> <strong>Quality gate:</strong> a post is only published if it has ≥ 1 sourced numeric fact and ≥ 2 internal links to data pages (ticker, indicator, glossary).</li> <li> <strong>Cost per token measured per model;</strong> a mix of models per stage (writing, review, image).</li> </ul> <h2> Three bugs Search Console showed me </h2> <ol> <li> <strong>The API was being crawled as content.</strong> Googlebot discovered <code>services.organizesee.com.br/v1/...</code> and hundreds of API URLs turned into "crawled, not indexed". Fix: <code>X-Robots-Tag: noindex, nofollow</code> on every service response — and <em>not</em> <code>Disallow: /</code>, otherwise Google never reads the noindex.</li> <li> <strong>Proxied wildcard DNS.</strong> A <code>* CNAME</code> pointing to Vercel made any made-up subdomain resolve; scanners hit <code>11st.mydomain/.env</code> and the edge returned 525 — 930 a day, 325 of them served to Googlebot. Replacing it with an explicit <code>www</code> record zeroed it.</li> <li> <strong>Ticker authority split across 3 URLs</strong> (<code>/explorar/acoes/X</code>, <code>/dividendos/X</code>, <code>/preco-historico/X</code>). Each one needs to claim one thing, with its own canonical — or be consolidated. We are still measuring which wins.</li> </ol> <h2> What makes it different from an aggregator </h2> <p>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 <code>llms.txt</code> listing the questions each page answers.</p> <p>Technical feedback is very welcome — especially from anyone who has tried SSR + hydration with schema.org without duplicating <code>canonical</code>.</p> <p>Site: <a href="https://www.organizesee.com.br" rel="noopener noreferrer">https://www.organizesee.com.br</a> · Methodology: <a href="https://www.organizesee.com.br/fontes-e-metodologia" rel="noopener noreferrer">https://www.organizesee.com.br/fontes-e-metodologia</a> · Whitepaper: <a href="https://www.organizesee.com.br/whitepaper" rel="noopener noreferrer">https://www.organizesee.com.br/whitepaper</a></p> ai seo python webdev