The problem this solves
AI answer engines cite pages they can crawl, parse, and quote with confidence, and most sites quietly fail one of those tests. The blockers are unglamorous: crawler rules that shut out answer-engine bots, headings that bury the answer, missing schema, stale dates, no llms.txt. Each fix is small, but finding them all requires someone who understands both the site's code and how answer engines read pages - and that audit sits in nobody's job description.
So citability work stalls in the gap between marketing, which sees the visibility problem, and engineering, which owns the files. Recommendations arrive as a slide deck, get translated into tickets, and die in a backlog. The shortest path through that gap is arriving with the actual changes: concrete diffs an engineer can review in minutes, with the reasoning attached to each one.
How the mission runs
- Crawl and grade the site. Site Audit crawls the site and grades the mechanics that citability depends on: crawler access rules for answer-engine bots, page structure and heading hierarchy, schema markup coverage, freshness signals, and whether llms.txt exists and says anything useful about the site.
- Test how engines see you today. AI Visibility checks how the site currently shows up in live AI answers for questions in your category, which grounds the audit in outcomes: pages that should be cited and are invisible get priority over theoretical improvements to pages nobody asks about.
- Rank the fixes. Findings are ranked by expected citation impact against implementation risk. Crawler access problems come first, since they gate everything else; structural and schema fixes follow; content-level rewrites are flagged for a separate effort. Each finding cites the specific evidence behind it.
- Write the changes. The agent edits the actual files: robots and crawler directives, llms.txt, schema blocks, heading structure, metadata and date handling. Every change is scoped to be minimal and reviewable, touching what the fix requires and leaving the rest of the file alone.
- Open the pull request. All changes land as a single GitHub pull request with a description explaining each fix and the finding it addresses. Nothing merges without your review - the PR is the safety gate, and your normal checks run against it like any other contribution.
The prompt
This is the exact objective the agent receives. Swap the obvious placeholders for your own domain, segment or channel and run it as-is from the console, Slack, or the API.
What comes back
A GitHub pull request containing the complete set of citability fixes - crawler directives, llms.txt, schema markup, structural and freshness changes - with a per-file explanation tying each diff to the audit finding behind it, plus the audit report itself as the evidence base. Your team reviews and merges on its own terms, and the changes ship through your normal deploy pipeline.
Make it yours
- Scope the audit to your highest-value section, such as docs or comparison pages, and get a smaller pull request that can ship this week.
- Run the audit read-only first to get the graded report with no code changes, then trigger the PR once the findings are agreed internally.
- Re-run quarterly and let each PR handle only the drift since last time - new pages missing schema, dates gone stale, new crawler bots to admit.
Frequently asked questions
Is it safe to let an agent modify my repository?
The agent never pushes to your main branch. All changes arrive as a pull request on its own branch, your CI runs against it, and a human merges it. Reviewability is a design constraint: diffs are kept minimal, and each one carries its reasoning in the PR description.
What evidence backs the audit findings?
Every finding links to the specific evidence: pages where a crawler rule blocks answer-engine bots, files missing schema, live AI answers where the site fails to appear. The PR description pairs each change with its finding, so review is checking a claim rather than trusting a score.
How much difference do these mechanical fixes make?
They remove the failure modes that keep otherwise-citable content invisible - blocked crawlers, unparseable structure, missing freshness signals. Mechanics set the ceiling and content sets the height; the audit is explicit about which of your gaps are mechanical and which need content work to close.