# Link-building statistics: refresh plan Owner: SEOlutions editorial team. Edition: 9 September 2026. The portal article and the current downloadable reference contain 107 distinct findings from 33 studies and research releases. The collection is a review of published research, not a reproduction or independent audit of the publishers’ underlying datasets. ## Review dates and priorities - By 9 December 2026: recheck source availability, changed claims, AI-citation research and marketplace listing data. Record the actual check date; an accessible page is not automatically a reverified claim. - On release of a new annual edition: review BuzzStream, Muck Rack, Cision, Orbit, Aira, Ahrefs and WebAIM research. Confirm actual fieldwork dates and response counts before changing the displayed year. - Every six months thereafter: review ranking, internal-link and link-retention research. Keep historically useful results dated rather than relabeling them as current measurements. - Immediately after a reported error: inspect the original table or chart, correct the claim and preserve the correction in the change log. This document specifies editorial work. No automation or recurring notification has been created. ## Per-finding checks 1. Open the original source and any linked methodology. Check the figure, denominator, population, collection window and definition of the metric. 2. Distinguish a page-update date from a study date. Keep missing dates marked as undisclosed. 3. For comparisons, preserve the whole comparison under one finding key. Do not count each percentage, its complement or a reworded takeaway again. 4. Check whether reported costs are quotes, transactions, service retainers or campaign cost estimates. Preserve currency and included scope. 5. Keep association, self-reported opinion and controlled experimental effect distinct. Show confidence limits or inconclusive results beside the estimate. 6. Hold the record if the original source is unavailable or contradictions cannot be resolved. Document why, then update the public count. ## Records and publishing workflow The full working ledger is `output/citation-hubs/link-building-statistics/evidence.json`. It retains excluded and held candidates. Stable evidence IDs such as S038 are independent of the visible ordering; links such as `#finding-s038` must keep resolving when ordering changes. The formatter at `artifacts/link-building-statistics-insights/build_research.py` contains the reviewed source metadata, new findings and reader-facing interpretations. Edit the matching source record or formatter entry, then run the formatter to keep the public JSON, Markdown reference and article data synchronized. Update its count assertion only after an editorial recount. The MDX overview is separate and must be checked for repeated claims and count changes too. Run the citation-hub skill’s offline evidence audit against the working ledger and Markdown review copy. This checks fields, unique keys and citation coverage; it does not replace the source review. Run `npm run build`, then check the initial HTML, filters, source disclosures, downloads, finding anchors, canonical metadata and mobile layout. Preserve old download URLs. The original 80-finding collection is an archived, broader SEO edition; its evidence is stored separately from the current 107-finding reference. ## Important source-specific cautions - Ahrefs link rot: 66.5% is cumulative loss across a historical, roughly nine-year window beginning January 2013. It is not an annual attrition rate. - Ahrefs AI citation overlap: 37.1% uses organic blue links; the rounded headline also counts other result blocks. Older parsing methods are not directly comparable. - BuzzStream guest-post costs, June 2026: “direct” is a lowest-vendor-quote proxy. Some high-tier and news-pricing figures conflict between summary and body. The selected quote average and traffic findings must not be mixed with those unresolved figures. - Muck Rack AI, May 2026: the companion methods gives more than 25 million citations for the latest edition, while the summary introduction repeats an older one-million description. Keep that discrepancy visible. The prompt set intentionally excludes branded transactional prompts. - Siege organic link ratios: the numerator is new referring domains and the denominator is organic visits. A falling denominator can raise the ratio. Do not call it a raw backlink growth rate or attribute the change to one content tactic without further evidence. - SearchPilot: preserve test-specific populations and significance. The breadcrumb case also changed the consistency between visible navigation and its schema. ## Change log ### 9 September 2026 — plain-language edition Removed all ten findings from the excluded publisher and removed that source from the current article, source index, evidence download and working ledger. Added seven previously verified findings from Ahrefs, Orbit Media, Pew Research Center and WebAIM, leaving 107 findings from 33 studies and 33 original research URLs. Added a separate plain-English explanation to every finding, simplified study-type and disclosure labels, and updated the article, generator, downloads and checks for 107 explanations and 214 action steps. Existing evidence IDs remain stable; the three removed generated IDs stay unused so later anchors did not change. ### 9 September 2026 — expanded portal edition Replaced the article’s limited selection and 80-finding download with a full on-page collection of 110 findings. Retained 54 relevant findings, removed 26 general SEO findings from this edition, and added 56 source-checked findings. Added ten topic groups, search and evidence filters, practical interpretations, source disclosures, stable finding links and a source index. The largest study supplies ten findings; AI occupies ten of the 110 entries. During the final source review, corrected the description of Reboot’s 12% subject-line association: the relevant grouping was celebrity names, events and special dates, not the separate “study/survey” wording category. Corrected inherited sample descriptions for subset findings and retained uncertain or historical study labels.