How to Benchmark Two Link Indexes on Your Own Site

You cannot determine which link index is “more accurate” in general, because there is no reference copy of the web’s link graph to check against. What you can do is measure how two indexes differ on a site whose links you partly know — your own — and get a specific, local answer about coverage and staleness.

That answer is far more useful than a vendor comparison article, because it’s about your sector, your site, and the neighbourhoods of the web your links live in.

Why “which is more accurate” is unanswerable in the abstract

The ground truth would be the complete set of links to a site at a moment in time. Nobody has it. Google doesn’t publish it, Search Console shows a capped sample, and each vendor has a crawl rather than a census.

So a claim like “Tool A found 40% more links” is a statement about crawl size and retention policy, not accuracy. More rows can mean better coverage or slower purging of dead links, and from the outside the two look the same. That’s the mechanism laid out in why two tools report different backlink counts; this post is the procedure for quantifying it locally.

The overlap test

Work at the referring-domain level. Link-level comparison drowns in URL normalisation differences and tells you less.

Export referring domains from each tool for the same target domain, on the same day, with matching filters. Record the filters: all attributes or followed-only, subdomains included or not, live-only or historical. Mismatched filters invalidate everything downstream, and the defaults usually differ between vendors.
Normalise the domain strings. Lowercase, strip www, strip trailing dots, decide on subdomain handling and apply it identically to both files.
Compute three sets: in A only, in B only, in both. Report the three counts and the Jaccard-style overlap if you like, but the three raw counts are the finding.
Sample each exclusive set. Take twenty from “A only” and twenty from “B only” and open them. For each, record: link present, link absent, page gone, page unfetchable.
Tabulate. You now have an estimate of what fraction of each tool’s exclusive rows are real, current links versus stale or phantom rows.

That’s the whole test. It takes an afternoon and it produces numbers you can cite for months.

What the results tell you

High overlap, small exclusive sets. Either tool will do for headline reporting. Use whichever has the better interface for your workflow, and note that switching mid-year still breaks your time series.

Tool A’s exclusive set verifies as mostly live links. A has better coverage of your neighbourhood. Worth knowing which neighbourhoods — check whether the exclusive rows cluster by language, TLD, platform, or site size. They usually do, and the pattern is the real insight.

Tool A’s exclusive set verifies as mostly dead pages. A retains links longer. Its higher total is partly historical, which matters if you’re reporting growth.

Both exclusive sets verify as live. Neither index is close to complete, and the union is a better estimate of your profile than either. This outcome is common for sites with links in non-English or platform-heavy neighbourhoods.

Low overlap overall. Check your filters again before believing it. Genuinely low overlap usually means one tool is including something the other excludes at a structural level — subdomains, or a category of source.

Add Search Console as a third leg

Google’s Links report is capped and undated, so it can’t be compared on counts. It can be compared on membership, which is the useful part.

Take twenty of each tool’s exclusive domains and check whether they appear in the Search Console linking-sites table. A domain absent there might be below the cap rather than unknown to Google, so absence proves little — but presence is a first-party confirmation that the link exists and Google knows about it. The report’s specific limits are in reading the Search Console Links report.

What to do with the numbers afterwards

Pick a primary and stick to it. Time series across tools are not time series. The benchmark tells you what you’re giving up by choosing, which is precisely the value of running it.

Quote the disagreement in your reporting boilerplate. Something like: “Referring domains per [primary vendor]: 812. A July overlap test against [secondary vendor] found 640 domains in both, 172 unique to the primary, 268 unique to the secondary; a 20-domain sample of each exclusive set found most secondary-only rows to be live links. Treat 812 as a conservative count.” (Illustrative figures, not measurements.)

That paragraph does something rare: it puts an honest uncertainty band on a link count, derived from a test you actually ran.

Re-run it annually, or after any vendor announces an index change. The result has a shelf life, for the same reasons covered in when a metric moves but nothing changed.

Don’t extrapolate to other sites. Your result is about your link neighbourhoods. A site in a different market may see the opposite ranking of the same two tools.

The traps that ruin the test

Mismatched filters. The most common failure by a wide margin. Write the filter state next to each export before you compute anything.
Different dates. A week’s gap adds churn to the difference and you can’t separate the two.
Export caps. If either file hit a cap, you compared samples, not sets. Check for round row counts, per sampling, export caps, and missing rows.
Subdomain handling. Whether blog.example.com and example.com are one referring domain or two is a vendor decision, and it can shift counts substantially in either direction.
Judging from the exclusive sets alone. Twenty rows out of a few hundred is a small sample; treat the verified proportions as rough, and say so.

What this doesn’t settle

Which tool is better. Whether either is close to complete. What Google actually knows, and — the separate and unanswerable question — what Google actually counts.

What it does settle is the local, practical version of the question: for your site, right now, how much do these two datasets disagree, and how much of the disagreement is real links versus index residue. That’s a number you can put in a report and defend, which is more than any vendor comparison can give you.