> ## Documentation Index > Fetch the complete documentation index at: https://promptwatch.com/docs/llms.txt > Use this file to discover all available pages before exploring further. # How actions are prioritized > The severity levels and action types on the Action Board, and the rules the agent follows when assigning them. Every action on the [Action Board](/docs/academy/action-board) carries a severity and a type. Severity decides order; type decides what kind of work it is and where its deep link takes you. ### Severity levels Four levels, in board order: * **Critical**: confirmed issues with direct visibility impact. The agent reserves this for problems the data clearly supports. * **High**: strong evidence, meaningful impact. Setup actions (missing sitemap, crawler logs, visitor analytics, or Search Console) are always created at high severity. A missing sitemap and Search Console skip the content and performance discovery runs; the other focuses still run. * **Medium**: worth doing, used sparingly by design so the board doesn't drown in maybes. * **Low**: useful non-urgent follow-up, the rarest level. Rules the agent follows when assigning severity: * **Evidence only**: severity comes from tool data, never speculation. The agent is instructed to skip preventive "this might become a problem" actions entirely. * **Prefer the top**: when choosing what to save within its per-run cap of 5, the agent prioritizes critical first, then high, and cuts noisy medium candidates. New actions are never created at low; low only appears when the agent later downgrades an existing Suggested or To Do item. * **Severity is revisited**: on later runs, the agent re-checks open items against fresh data and can raise or lower severity. Only actions still in Suggested or To Do can change; once you pull something into In Progress, its severity stays put. On the board, columns sort by severity first (critical, high, medium, low), then newest first. You can filter the board by severity, type, and assignee. The Promptwatch Action Board filter bar with the Severity dropdown open, listing Critical, High, Medium, and Low. ### Action types Nine typed actions plus one free-form type. Eight of the typed actions come from a rotating discovery focus, one focus per UTC day, Monday through Saturday: Performance and Site Health share a day, as do Untracked and Untapped. Setup sits outside the rotation: * **Content Gap**: create or improve content where AI answers a question but your site barely covers it. Grounded in [content gap analysis](/docs/academy/content-gap). * **Sentiment**: push back on unfavorable AI framing of your brand before it takes hold. * **Reddit**: join Reddit threads AI frequently cites where your brand currently has no voice. * **Outreach** (offsite mentions): get your brand included in high-authority third-party articles AI keeps citing. * **Untracked**: start monitoring pages that get AI crawler traffic but aren't tracked yet. * **Untapped**: strengthen pages AI crawlers visit often but rarely cite or drive clicks from. * **Performance**: speed up slow pages so AI crawlers can reliably read and index your content. * **Site Health**: fix structural SEO issues that confuse how AI classifies your pages. * **Setup**: connect missing integrations (sitemap, crawler logs, visitor analytics, Search Console) so the analysis covers the complete picture. Created programmatically, not by the agent. * **Custom**: free-form actions created through Agent Chat when no typed action fits. The discovery agent never creates these. ### Why a rotating focus The discovery agent runs one focus per day (content, performance, page signals, Reddit, offsite, sentiment) instead of everything at once. This is deliberate: each focus gets a full evidence-gathering pass with its own tools, and the daily cap on new suggestions keeps the board reviewable instead of overwhelming. The agent also checks how many open items each type already has and avoids piling onto a type you haven't cleared. ### What this means for you Trust the top of the board. A critical or high action passed an evidence bar and a dedup check before it reached you, and the agent sees how many open items each type already has when it decides what to save (at most 5 new suggestions per daily run). If you disagree with a severity, work or dismiss the action anyway; a dismissed item stays off the board for 30 days before the agent can suggest the same issue again. See [Working an action to done](/docs/academy/working-an-action) for the status flow.