Glossary

What is an action card in Keats?

In short

An action card is one piece of work Keats proposes when questions fall below their aim: the gap, the diagnosis, the evidence, the change to make and the questions it targets. Cards are ranked by the visibility they could recover. When the change is live, the card's verdict compares its questions with untouched ones.

Keats, from PressRoom. Updated Read as markdown

Each week Keats reads the answers, finds the questions below their aim, and works out why. The result is a set of action cards, each one a change someone can make.

A card has:

  • A type, from the four fixes in order: fix a fact, update a page the assistants already read, write a new page, or get onto a third-party page. A fifth type, the fact check, flags a wrong or inconsistent statement about you wherever it appears.
  • The diagnosis and the evidence: who is named instead, what the answers say, which pages they cite.
  • The target questions, with their rates and aims.
  • A brief: how the page should be written and what counts as done. An agent can work from it directly.

Cards are grouped by the fix, so one page that would move four questions is one card, and ranked by the visibility they could recover, buying questions first.

When the change is live, you or your agent report its URL and date. That starts the evaluation against control questions at day 21 and day 35, and the card ends with a verdict. Working cards through an agent is described in AEO with your AI agent.

Sources

  1. Keats: llms.txt, the operations and workflows for agents, read November 12, 2026

Track these questions every week

Keats asks your buyers' questions on five assistants every week, shows where you're left out, and has your agent do the work. A person reads every early-access request.