An internal linking agent for orphan pages and AI crawlers
Pages with few or no internal links may never be reached by crawlers. Here is how to find them, fix them with an agent's help, and check with crawler logs whether bots actually arrive.
Every site has pages that nobody links to. A landing page from an old campaign, a how-to guide that the navigation redesign dropped, a comparison page that only exists in the sitemap. People rarely land on them. Crawlers might, if they find them some other way, and that "if" is the problem this post is about.
We should be careful about what this does and does not mean. Orphan pages are a crawl-path problem. We have no evidence that having few internal links lowers how often a page is cited by an AI engine, and you should be wary of anyone who says it does. What we can say is narrower and safer: a page a bot cannot easily reach is a page you cannot count on being read. And since reading is the first step toward citation, it is worth knowing which of your pages sit in that group.
What counts as weak
A few terms, used the way site audits use them.
- Orphan: no internal link points to the page. It can still be reachable through a sitemap, an external link or a search engine's memory.
- Weak: only one or two internal links point to it. A single editing mistake can orphan it.
- Redirect or error targets: links that lead to a redirect chain or a broken page. They waste the crawl and the link equity.
Pages end up here for dull reasons. Navigation changes. Migrations. Content hubs that were never finished. A product line that was retired from the menu but not from the site.
Finding them without a tool
Take your sitemap URLs. Crawl your site starting from the homepage and list every URL you reach by following links. Subtract the second list from the first. What is left are sitemap pages that links do not reach.
Then do the reverse. URLs you reach by crawling that are not in the sitemap are pages you forgot to list, or pages that should not exist.
For weak pages, count inlinks per URL from the crawl and sort ascending. Anything with one or two is on the list.
Finally, decide what each page deserves. Some orphans should die: redirect them or remove them. Some should be linked from a hub. Some should be merged into a better page.
Fixing them by hand
The fix is editorial, not mechanical. A link from an unrelated page is noise. A link from the page where a reader would naturally want the next step is useful.
For each page you want to keep, find two or three existing pages that discuss the same topic, and add a contextual link with descriptive anchor text, not "click here." Add the page to the relevant hub or category listing. If you maintain a "related reading" block, make sure the new page is eligible for it.
It is a job that gets done in a burst and then decays. Sites change, new pages arrive, and a year later you are back where you started.
What Promptwatch's October changes cover
The changelog entry for October 28, 2026 describes a linking map in Site health. It opens with a map of your site by URL path. Orphan pages, pages with only one or two inlinks, and redirects or errors each get their own color. You click a section to zoom in, or filter to just the weak spots.
The rest of the entry is useful for the fixing step:
- Each page's detail view has a Linking tab: the pages that link to it on the left, the pages it links to on the right, and broken links in red.
- Every page list shows where a page came from: your sitemap, a link from another page, Search Console, or a site map crawl. That distinguishes "listed but unlinked" from "linked but unlisted."
- Broken internal link issues list each broken target with its status code and anchor text.
- Sitemap-only audits check only the pages in your sitemaps. New projects start with it on, and pages that drop out of every sitemap leave the audit after the next full sync.
On the writing side, the October 2, 2026 entry says the Content Agent now looks for relevant pages on your site before it starts writing, so internal links are part of the article from the first draft. Pages with many Search Console impressions but few clicks are suggested first, and it stopped linking your homepage every time your brand name comes up. That means new content arrives already linked, instead of adding to the orphan pile.
The audit itself is weekly, with 21 checks and a health score, and an email arrives every Tuesday with the health score, issues fixed since the last audit and the top issues AI crawlers may trip over. Site audits are on paid plans; Essential covers up to 1,500 pages.
Closing the loop with crawler logs
The map tells you which pages are structurally weak. It does not tell you whether a bot ever turned up. For that you need crawler logs, which show real AI crawler requests, IP-verified so spoofed user agents are filtered out, with the actual bot names such as GPTBot, OAI-SearchBot, ClaudeBot and PerplexityBot.
Crawler logs start on Professional. Essential has none, so on that plan the map is the whole picture and you should say so in any report.
With logs in place, the check is simple. Take the list of weak and orphaned pages you fixed. A few weeks later, ask whether those URLs received AI crawler requests, and whether they had before. The crawl-to-citation view gives each page a citation rate, so you can see which fetched pages were also cited. Compare with an unchanged control group of weak pages you left alone, and you will have something close to evidence instead of a feeling.
A workable agentic routine
- Weekly: read the Tuesday audit email. Open the Linking map and filter to the weak spots.
- Triage: for each flagged page, decide keep, merge or retire.
- Draft links: use Agent Chat or the Content Agent to propose contextual links from relevant pages. Review each proposal, because an anchor that reads badly is worse than none.
- Apply: make the edits in your CMS.
- Verify: after the next audit, confirm the page left the weak color. Later, check crawler logs for the URL.
- Prevent: have new articles linked at creation, which is what the Content Agent's first-draft links are for.
For the audit side in more depth, see the weekly site audit, health score and GEO fix queue. For what bots do when they arrive, read how an agent reads crawler logs and proposes robots.txt fixes. And for a migration, where orphans are born, use the site migration checklist.
Recommendation
The combination that matters is map plus logs plus a drafting agent that links as it writes. Promptwatch is the platform we would test first for that, because a standalone site crawler can show you orphans but not whether AI bots came, and a log tool can show bots but not why a page was unreachable. Keep expectations modest: fixing links makes pages easier to reach. It does not buy a citation.