Open Audit
- Open a project
- Select Insights › Audit in the project sidebar
- Choose Domains or Pages
Domains
The Domains tab focuses on technical readiness for owned domains. Select a domain to review:- Overall readiness out of 100
- AEO readiness, Agent readiness, and SEO health pillars
- Host checks for crawler access, sitemap coverage, indexability, and orphan pages
- Open findings grouped by severity
Work with findings
Expand a finding to read What to change, verification guidance, affected pages, and attached evidence. If you are a project Member or Admin, you can:- Select Create action, review the prefilled title, then select Create in Actions
- Select Dismiss finding when it should no longer appear as open
- Select Show and restore → to review dismissed findings, then Restore one to reopen it
devtune_dismiss_audit_finding and devtune_restore_audit_finding MCP tools. Both need the actions.write scope and a credential from a project member with the project Member role.
Agents and API clients can create the same action with the Create an action from an Audit finding endpoint or the devtune_create_action_from_audit_finding MCP tool, with the same scope and permission. A finding that already has an open action, from the Audit page, an earlier request or an earlier audit of the site, returns that action instead of a second one and keeps its description, and its title unless a new one is sent. They are refused for an audit a newer completed audit has replaced, for the site or for the finding’s existing action (retry with the newer audit), a dismissed finding (restore it first), a locked page, a high-stakes page until acknowledgeHighStakes is sent after a person confirms, and an action that is archived or completed. Archive, restore or reopen the action with Change an action status or devtune_change_action_status. Dismissing a finding leaves its action as it is.
Pages
The Pages tab combines classic search, AI-answer, crawl, and traffic evidence for individual pages. Use the Last 7 days, Last 30 days, and Last 90 days presets to change the reporting window. Page metrics update once a day and cover complete UTC days, ending yesterday. New probe results, traffic, and Search Console or analytics data appear after the next daily update. Locking or unlocking a page, changing domains, changing tracked pages, and finishing a crawl update the Pages tab the same day. The heading states how many pages lost AI citations or presence in the window, or have them at risk. The summary band shows:- Classic search clicks from Google Search Console
- AI citations from your tracked prompts
- Lost or at risk, the number of pages diagnosed Lost citations & presence or Cited, at risk, with how many pages had no AI citations or presence in your tracked prompts
- AI referral bounce rate for each domain, when PostHog is your connected analytics provider
Diagnoses
The Pages table labels each page with its strongest available diagnosis and orders pages by diagnosis and supporting evidence, highest opportunity first. Diagnoses are:- Lost citations & presence
- Cited, at risk
- Ranks on Google, not cited
- Fetched, not cited
- Locked
- No traction
- Healthy
Inspect a page
Select a row in the Pages table to open the page inspector. It shows:- The page’s diagnosis and lock status
- Section clarity findings for the page
- Why this ranks here, the evidence behind the page’s position in the table
- Clicks, impressions, position, AI citations, citability, and answer fetches
- Next action: Create action, or a link to the page’s action in the Actions queue
- Daily AI citation history
Crawler access
Crawler access reports the sampled URLs and each provider’s search indexing, answer fetching or grounding capability. Probe passed means a synthetic request returned expected page content with applicable policy permitting retrieval. Verified crawler · HTTP 2xx is supporting HTTP evidence, not proof that the expected content was delivered. Access unconfirmed means some checks lack conclusive evidence. The score is provisional and uses only assessed combinations. Some access restricted applies to part of the sample, not the whole site. Restricted by policy remains visible even if a request succeeds. Training-only restrictions do not lower the score. Connected traffic can support a new audit only within its preceding 24 hours. The evidence panel shows the actual complete-hour window and source precision; omitted hours and untested URLs are not assessed. Each run retains its own evidence. Public audits use public observations without connected-project traffic.Intentional restrictions
In a retrieval finding, select the exact URLs you intend to restrict, confirm your intent, then choose Acknowledge restriction. The finding, evidence and score remain visible. Recommendations and generated prompts omit those URLs. Other URLs remain actionable, including URLs first discovered in a later audit. Select the URLs and choose Reconsider restriction to restore their recommendations. This differs from dismissing a finding. The project Member role is required, and locked pages remain protected. Agents and API clients can make the same change with the Acknowledge or reconsider a retrieval restriction endpoint or thedevtune_acknowledge_retrieval_restriction and devtune_reconsider_retrieval_restriction MCP tools. Both need the actions.write scope; audit.read can’t change intent. DevTune can’t tell whether you confirmed; your agent says you did. It should ask you rather than infer your intent from a detected restriction.
robots.txt edits
When robots.txt blocks an answer engine or AI search crawler on a sampled page, the page’s evidence names the rule, the crawler and the path. An action created from that finding includes a drafted robots.txt edit in its brief: the edited file and a diff to copy. The draft keeps every existing line and adds only the narrowestAllow rules for the blocked crawlers. If lifting a whole rule could also open a page you acknowledged as an intentional restriction, or a path another of your rules blocks, the draft allows only the exact sampled pages the audit found blocked, and says so; widen it yourself if you want crawlers to reach more. A crawler with no group of its own gets a new group that repeats your User-agent: * rules, because a crawler with its own group stops reading *. Drafts never unblock training crawlers such as GPTBot, ClaudeBot or Google-Extended.
To apply a draft:
- Open the action’s brief and read the diff.
- Compare it with your current robots.txt. The draft comes from the file as the audit read it.
- Apply the edit yourself. DevTune never changes, hosts or serves your robots.txt.
- Select Mark live on the action. Results then compares requests from the agents the change unblocked over the 14 days before and after. That needs a server-side sensor, Cloudflare, edge middleware or a Vercel log drain, on each changed site for both windows.
devtune_get_action_brief MCP tool.