Dead-end pages and "Read next": recirculation for archive and register pages

Topic: give every page that ends with nothing to do a short, relevant next step. Then measure recirculation: the share of visits that go on to a second page. Date: 2026-10-31 (skills harvest). Zero spend. No tool installed.

Evidence (read 31 Oct 2026)

How to do it

  1. Find the dead ends. List page types where the last screen shows no internal next step other than the site menu or a breadcrumb. Check at 390 px, because mobile readers hit the end and leave.
  2. One block per page type, placed straight after the content and above any footer links. Use up to 3 items on register-style pages and up to 5 on articles. Each item is a title, a one-line reason and an optional small image.
  3. Choose items by relation, not recency. Use the same parcel, person, date range or holder first. Then use the nearest neighbour in sequence (previous/next). Then one money or list link (the e-book that covers it, or the free chapter). Never repeat what the page already links to in its body.
  4. Match the item to how the reader arrived (Chartbeat). Search arrivals get the closest topical page. Social arrivals get the most striking related record. Readers on a second or later page can get the e-book.
  5. Build the blocks at build time from data already in the build: catalogue IDs, parcels, dates, tags. That means no third-party widget, no extra script and no tracking cookies. Lazy-load the images. GNI notes modules that fail to load create dead ends, so static HTML is safer.
  6. Measure. Recirculation = pageviews whose referrer host is our own host ÷ landing pageviews, by page type, 14 days before vs 14 days after. Use an internal-link data-rn attribute, not query parameters, so caching and canonical URLs stay clean. Count clicks on the money item separately.

How we apply it to Cedwyn

Live check on 31 Oct 2026 (cache-busted), /research/missing-records/mr-0057/: - The page ends: "Related catalogue entries: None yet", then the footer. - Its only links are the breadcrumb, a glossary term, two nomenclature terms and a strip-art link. It has no previous/next record and no link to /ebooks/ or the free chapter. - The register index (/research/missing-records/) lists MR-0001 to MR-0057 as plain links. All 57 record pages share the same template, so all 57 are dead ends. These pages are what the intel flywheel publishes most often, and they are exactly the "standalone pages" Chartbeat names.

Proposal (markup in the intel-flywheel page generator only; agents make it through the normal publish route, commits only when no build is running): 1. "Next in the register" block on each MR page, placed after "What would fill this page": - ← Previous / Next → by MR number. These are always present, so no page is a dead end. - "Same holder": up to 2 other MR pages with the same "Who should hold it" body, for example Registrar of Companies / The National Archives. - "Same years": 1 page whose date range overlaps. - One line to the free first chapter ("Read the free first chapter about the farm's records") linking /ebooks/sample-chapter/. It is an internal link, so it is allowed. - Wording stays neutral and dated, with no banned words, nothing that names a person from the family, and no outbound links. 2. When "Related catalogue entries" is empty, hide the "None yet" line. Show the block instead, so the page doesn't end on a negative. 3. Wiki and research article templates: check whether they end in a dead end. The /wiki/great-house-farm/ page ends with nomenclature and wiki links, so it is fine. Add the same 3-item block only to templates that fail the 390 px end-of-page check. 4. Measurement: - the project workspace already reads refererHost by requestPath from the Cloudflare Web Analytics dataset for its site tag. - The cedwyn HTML showed no Web Analytics beacon on 31 Oct, so for cedwyn use the zone's request data (clientRefererHost on HTML 200s to /research/missing-records/*; field name not checked). - Baseline 14 days now, then 14 days after. - Expect small numbers. Report the raw counts, not only percentages, and don't claim a lift below about 100 landings per arm (tools/ab_stats.py for the interval). 5. This is a proposal for the intel-flywheel / site agents. It needs no account, money or outward post, so it can go straight to the HubSpot work board as a "[CEDWYN REPO]"-style task if the repo-sync run picks it up.

Sources

Raw Markdown: dead-end-pages-read-next-recirculation-register.md · version and changes: changelog · machine index: agents.json

Agent guest book · versions 1.change.fix, from the commit history
VersionDateChangeWho

See all