A crawl export shows hundreds of broken links after a site update. Some rows point to the same retired page from a shared footer. Others come from individual articles, including one link customers need to book a service. Fixing rows in export order can bury the booking issue and send the same footer problem to several people.
Keep each link’s source and destination, test each destination efficiently, then prioritise by the journey and the component that owns the link. The existing AI Visibility internal linking guide covers which connections to make and how to label them. This walkthrough covers what to do when those connections fail across many pages. AI Visibility internal linking guide
Define the journeys and the crawl boundary
Pick the journeys where a broken route would matter. For a local business, that might include service selection, branch details, an enquiry form and a booking page. Include the pages that send people into those routes, such as a homepage card or a popular guide. Add representative templates and language versions. A crawl can find failures across a wider inventory, but those starting points make it possible to explain why a particular issue deserves priority.
Record which origins, paths and page types are in scope. A booking application can live on another host while still being important to the customer. Identify who owns it and include it in a separately authorised check if needed. Limit the crawl rate and avoid running automated checks through actions that submit forms, create orders or change accounts. Use published pages and safe retrieval requests. If access is restricted, work with the owner to define which observations can be made.
| What to keep | Why it matters | Example decision |
|---|---|---|
| Source page and link occurrence | Tells the editor where the link lives | Repair a booking link in a specific service page |
| Unique requested destination | Avoids repeating the same status request | Check a shared footer target once |
| Template or component | Groups repeated failures by owner | Update a single navigation component |
| Final response and expected content | Separates a working destination from an error behind a redirect | Correct a link that lands on the wrong guide |
| Fragment and destination element | Finds failures invisible to status checks | Restore the missing section ID |
Collect links without losing where they came from
Start with ordinary HTML anchors. Google’s crawlable-link guidance describes anchor elements with an href as the reliable starting point for its link extraction. Resolve relative addresses against the document base. Preserve the original value, the source page and the resolved target. Keep an anchor label or surrounding context so the editor can find the link and decide where it should go. If a page uses a base element, apply the browser’s base resolution rather than assuming the final document URL is the base. Google crawlable link guidance · W3C guidance on link purpose
Keep a destination list and a separate occurrence list. Remove the fragment for the HTTP request while keeping it in the occurrence record. This lets you fetch a document once and check several section targets against it. Keep query parameters where they affect the destination. If the report groups tracking variants later, retain the exact requested values so a developer can reproduce a failure. Record both the requested URL and any final URL returned after redirects.
Compare source and browser-rendered link inventories for representative templates. A menu may add links after rendering or when opened. A button that navigates with a script may work for an ordinary user while the initial markup supplies no anchor. Record the state that exposes it and check it as a separate journey when it is important. An HTML-only extraction is still useful if its coverage is clearly described.
For large inventories, decide how you will handle a URL that can generate thousands of combinations. A filter page can add parameters for every available selection. Keep the actual targets for the customer journeys under review and apply a documented scope limit before expanding the crawl. A list of expected public routes from the publishing system can help you cover important content while keeping the automated requests bounded. When you exclude a family of URLs, state the reason and how the team will check representative examples.
Store source-to-target rows before grouping. If one article has two links to the same booking page, the target test can be shared. Keep both occurrences because one might be an outdated call to action and the other might appear in a shared menu. When a single broken destination has different intended meanings on different source pages, the editor may choose a different replacement for each group.
Check statuses, redirect destinations and fragments
Fetch each unique target using a bounded request. A GET supplies the body you need to confirm that the expected page arrived. Save the redirect chain and final address. Flag a 404 or 410 where a current source still links to the resource. Put temporary service errors and rate limits into a separate retry category. A network failure without an HTTP response belongs in a separate category as well. Google documents how its crawler treats common HTTP statuses, and that platform guidance is distinct from the status you observed. Google HTTP status and network guidance
Inspect priority targets for content. A link to “Book a consultation” can return 200 and still land on a general homepage after a redirect. Compare the final title or a stable body marker with what the source promises. For same-page links, look for the named target element. If the markup is generated after load, test the relevant rendered state too. For high-priority cross-page fragments, check the element on the final document and then follow the link in a browser to verify the landing behaviour.
You can list the distinct raw link values on a page in the browser console before comparing them with a crawl. This helps find links added by a script. Keep the source page and resolve the values before grouping by destination.
Array.from(document.querySelectorAll("a[href]"), link => ({
source: location.href,
raw: link.getAttribute("href"),
resolved: link.href
}))
See what a small live sample can establish
On 6 November 2026, a limited audit fetched five public AI Visibility source pages. The sample included the homepage, the second archive page, the technical SEO category, the internal linking guide and the site access checklist. It counted same-host HTTPS anchor occurrences, resolved relative addresses and removed fragments for the document grouping. The five source responses contained 386 occurrences associated with 52 distinct document URLs. Repeated links to the same target were counted as occurrences and retained separately from the distinct URL count. AI Visibility homepage · Technical SEO category
The test requested the 12 most frequently linked destinations, reusing a captured response where it already had one. All 12 returned HTTP 200 without redirects. For example, the technical SEO category appeared 37 times across the five pages. That result illustrates why the number of links and the number of destinations should be reported separately. The sample leaves the remaining 40 destinations outside the status check and does not establish that the site is free of broken links.
The audit also checked 17 distinct same-page source-and-fragment pairs against their source markup. All had a matching target. In a separate 390-pixel-wide Chrome run on the internal linking guide, six resolved same-page fragments had matching elements in the DOM. The check included full-path links as well as links that might be written with a hash alone. It did not test scrolling, visibility under a sticky header or cross-page fragments. AI Visibility internal linking guide
Prioritise fixes by the journey and owner
Group failures by intended destination and by component. One navigation link can account for hundreds of occurrences. A repair in the shared component may resolve all of them. Confirm a few affected source pages before assigning the issue there, since visually similar links can come from different templates. Put an isolated enquiry or booking failure high in the queue even if only one page links to it. A customer blocked at that step has a different problem from a reader whose related-article link takes one additional hop.
Consider a hypothetical site with two findings. Its footer lists a retired guide on 200 pages. Its pricing page links to the wrong form once. The link count can explain the scope of the footer change, while the owner may still repair the form first because it blocks the intended enquiry. Give each issue its affected journeys, sample sources and likely owner. When the right destination is unknown, ask the content owner to choose it rather than substituting a broadly related page.
Studio Simpatico’s year-in-review describes building a publishing site for Iron & Air and embeds a post from the studio’s official Instagram account. That example illustrates why a publishing release may involve both shared components and editorial references. It supplies no link audit for the publication. Apply the audit to the actual journeys and templates under review. Studio Simpatico publishing project and official Instagram embed
Review the cases where a crawl could not decide the result. A timed-out target should be retested under known conditions. A link to an authenticated dashboard should be checked through the expected authorised route. A cross-page fragment might need a browser if the element appears after navigation. Leave these categories visible in the report so the reader can tell a confirmed failure from a limit of the crawl. Record a separate count of unchecked targets. That count makes the audit repeatable and helps the owner decide whether a broader run is needed.
Use a Broken link audit record with the source, original link, intended destination, observed status, final content or fragment, occurrence count, source-page count, owner and priority. Include the crawl date and which pages or rendered states it covered. Where permitted logs help diagnose a reported request, match the exact URL and time and verify the requester before attributing it to a crawler. The public sample used no private logs. Google requester verification guide
Keep the same source sample for the first comparison after release. Report any newly added templates separately so a change in scope does not look like a change in link health.
Retest the link from the original source after the repair. Verify the final page and any section, and repeat the action on a representative narrow screen if the link depends on a menu. For a template fix, check multiple affected pages. For a page that moved, fix the source links you control and verify that the old address still routes appropriately for people using bookmarks or outside references. Record which items were checked so the next audit starts from a known sample.
AI Visibility is published by Lunasol, which offers GEO and AEO services. The worked observations use public HTML, limited HTTP sampling and one browser DOM check. About AI Visibility



