AI policy

Last updated: October 27, 2026.

Indicora (indicora.com) uses artificial intelligence (AI) and automation to find, read and write part of what it publishes. This page says where, how, within what limits and, above all, who is accountable for the result. It complements the Editorial standards.

1. Where we use AI and automation

• Data items built from a fixed template. For official data with a stable format, such as economic indicators, rates and calendars, the text comes from a fixed template filled in with the source's numbers. There is no generative AI writing these items: the numbers come straight from the official data.
• AI drafts and summaries. For notices and documents, language models may extract information, summarize and propose a draft, always from sources we cite. We prefer the model to pick literal, attributed passages rather than paraphrase facts.
• Internal support. AI also helps group items about the same subject, spot duplicates and flag what deserves the newsroom's attention.

The language models come from external vendors and only receive text from public sources and newsroom instructions. We never send readers' personal data to AI models.

2. What AI does not do

• It does not write opinion, analysis or recommendations, which Indicora does not publish as a rule. Texts with terms such as "buy", "sell", "recommends" or "price target" are blocked before going live.
• It does not publish on its own about sensitive subjects, such as accusations against people, politics and elections or conflicting sources: those items get full human editing.
• It does not talk to readers or decide on accounts and subscriptions.
• It does not decide on its own what goes live. A person releases publication; the only exception is the data items in section 4, which follow a publishing rule defined and switched on by the newsroom.

If we use an AI-generated image, it is identified in the caption.

3. How we label content

• Article written or edited by a person. It carries the name of the human editor accountable for it. If AI helped with the text, a short note in the article says so and points to this page.
• Automated data item. It carries the institutional signature "Indicora Dados" (Indicora Data), the Data label, the source, the date of the data and a link to this page. It carries no person's name and does not pretend to be reporting.

The institutional signature does not remove anyone's responsibility: it only tells the reader the text was generated automatically.

4. Human review

• Review before publishing is the rule. Every type of content starts in prior review, and a person approves before the article goes live.
• Automatic publishing is the exception. It is only enabled, type by type, for official data items with a stable format, after post-publication review, quick unpublishing and alerts are working.
• Automated checks. Before any automatic publication, the system checks the source's official origin, the format, the dates, whether the numbers in the text match the data, the absence of banned terms and duplication. Any failure sends the item to a person.
• Post-publication review. Items published without prior review enter a queue and are checked by a person afterwards, with full sampling at the start of each type, reduced only while there are no serious errors. The official source is queried again later to detect whether the original document was revised.
• Safety brake. A serious error in an automated type sends that type back to prior review. The newsroom can also unpublish an item, a type or all automation at any time.

5. Who is accountable

Indicora answers for all content it publishes, including content generated with AI or automatically: it is our own content, not third-party content. The responsible human is the editor named in the Masthead. Saying "the AI did it" is no excuse for a mistake.

6. Limits and mistakes

Language models can make mistakes, invent details and misunderstand a document. That is why we prioritize official sources, fixed templates for data, literal passages and human review. If you find a mistake, including in an automated item, tell us through our contact page. Severity levels and correction deadlines are in Corrections.

7. Changes to this policy

This policy will change as technology and the law change. The date at the top shows the current version. Any expansion of AI use or automatic publishing will be recorded here.