A count is a result of a method
Two dashboards can disagree without either number being a complete census of the web. They may describe different target scopes, different observations or different collections. Start with a testable question: which records explain the difference after applying the same rules? A larger total alone cannot establish greater accuracy, freshness or business value. This guide reconciles reader-supplied files, while provider benchmarking remains a separate research task.
Align the same target and collection scope
Write down whether the target is one document, one hostname or a grouped domain with subdomains. Here exact-document mode removes only the fragment and retains protocol, port, path and query; host mode matches the hostname across protocols and ports. Domain mode uses the stated suffix boundary. Record live versus historical selection, attribute filters, collection timezone and any export truncation. Do not compare an all-history dashboard with a limited active-link file as if they were the same population.
Record the provider, report, index/live-history selection, filters, export limits and collection time in each source’s scope. Match the intended target first. The tool cannot verify that the supplied files actually use comparable provider settings. A collection date is not a link publication date.
Choose one counting unit
URLs use browser normalization. Observation identity preserves fragments and exact anchor/attribute text. Pair and document units remove fragments but keep queries, paths and protocols. Domains use the Public Suffix List including private suffixes; that is a grouping rule, not proof of shared ownership. No redirects or canonical URLs are resolved.
| Counting unit | A | B | Both files | Union |
|---|---|---|---|---|
| Export observations | 9 | 7 | 4 | 12 |
| Source–target pairs | 7 | 6 | 5 | 8 |
| Source documents | 6 | 5 | 4 | 7 |
| Source hosts | 5 | 4 | 3 | 6 |
| Referring domains | 4 | 4 | 3 | 5 |
Ahrefs · Backlinks / Referring domains
Reconcile the worked example
Fictional teaching exports for brand.example, collected on the same example date. A has 12 input records and B has 8; after explicit exclusions they contain 9 and 7 observations. Both yield 4 referring domains, but only 3 domains overlap. These are not measurements of any provider.
Export A starts with 12 records. One repeat, one same-domain source and one outside-target record are excluded, leaving 9 observations. B starts with 8 and excludes one same-domain source, leaving 7. Domain grouping produces 4 for each, but archive.example appears only in A and fresh.example only in B. The union contains 5 domains, with 3 shared. Neither unique domain is proven absent from the other provider’s full index.
Read each source’s own definitions
Google describes its Links report as sampled, with its own canonical grouping, duplicate handling and table limits. Semrush documents filters and export limits, and explains that Backlinks and Backlink Audit can have different scopes even within the same product family. Check those definitions before interpreting a discrepancy. The worksheet does not reproduce either provider’s proprietary index or Google’s normalization; it applies one disclosed rule to both supplied files.
Semrush · Backlinks / Backlink Audit
Investigate a difference at record level
Pick several A-only and B-only records and inspect the original export, the source page and the intended target with appropriate access. Record collection and verification times, the response and the actual link evidence. A provider’s first-seen date means detection in that source, not necessarily publication. Treat redirects, removed links and changed filters as hypotheses until verified. Keep failures to access a page separate from confirmed link absence. A provider-accuracy study would also need a defensible ground truth, representative sampling and a dated evaluation protocol.
How to do a backlink audit: evidence, priorities and an action template
Keep a repeatable comparison record
Changing inputs invalidates the comparison. Unit changes recalculate the grouping. Search and pagination affect only displayed rows; CSV and HTML contain the entire selected unit. JSON restores inputs, not display filters. Missing-file membership remains distinct from zero observed links.
Use Source URL and Target URL; Anchor, Rel and Nofollow are optional. Common English header aliases are accepted. Up to 2,000 records and 1 MB per CSV; JSON drafts up to 5 MB. Unsupported or invalid URL records are excluded with their record numbers. Other columns are retained in the draft but do not define an observation.
Files and notes stay in this browser. Compare supplied exports without contacting a backlink provider or fetching source URLs. Save a JSON draft to retain the exact inputs; nothing is saved to an account.
Download all comparison rows (CSV)
Add 12 navigation examples · CSV
Referring domains vs backlinks: what are you actually counting?