You Cannot Attribute a Competitor's Gain to Their Links

A competitor acquires 40 referring domains over a quarter and their rankings improve. Both halves are observable from outside. The inference joining them — that the links caused the improvement — requires knowing what else they changed, and that is precisely the information an external view cannot supply.

This isn’t a counsel of despair. Competitor link data is genuinely useful; it’s just useful for a narrower purpose than the one it usually gets put to. The distinction is between using it to explain their past and using it to find work worth doing.

What you can’t see about a competitor

Attribution on your own site is already difficult — How to Tell Whether a Link Did Anything is largely a list of reasons why. On a competitor you’re missing the inputs that make even a weak analysis possible.

Their content changes. Page rewrites, expansions, retitling, new internal links, consolidation of thin pages. Some of this is recoverable from an archive if you go looking page by page; most of it — the timing, the scale, which pages — isn’t practically recoverable across a whole site.

Their technical changes. A fixed canonical, a resolved crawl problem, a template change that surfaced pages that were previously buried. Any of these can produce a step change with no link involvement, and none of them are visible externally.

Their commercial activity. Brand campaigns, paid media, a product launch, PR that generated searches for their name. Demand changes rankings on brand-adjacent terms and changes engagement everywhere.

Whether the links are actually new. A link index’s “first seen” is when the index saw it, not when it was published. See What First Seen and Last Seen Mean. A crawl that reaches a neglected corner of the web produces a cluster of apparently simultaneous new links that were published over two years. That single artefact accounts for a good share of dramatic-looking competitor link spikes.

What the search results did independently. If the whole result set reshuffled, their gain may be the ranking system changing rather than them changing.

The counterfactual problem

Even with all of that, the question “did the links cause the gain” asks what would have happened without the links, and there’s no version of the competitor that didn’t get them. This isn’t a data-quality gripe; it’s the structure of the problem. There is one timeline, several simultaneous changes, and no control.

Which is why the honest form of the claim is comparative and weak: of the changes I can observe, links are the one I can quantify. That’s not evidence links were decisive. It’s evidence they’re the variable your tooling happens to measure, which is a fact about your tooling.

Why Link Correlation Studies Prove Less Than They Claim makes the same argument at population scale: aggregate correlations between link metrics and rankings are consistent with links mattering, with links being a symptom of the things that matter, and with both. A single competitor is a sample of one, so it can support less, not more.

Sample size, on a single competitor

The numbers involved are small enough to be dominated by noise. A quarter’s worth of movement on a few dozen tracked keywords, against a link change you’ve observed once, in a period where you can’t hold anything constant. Small Link Sets and Statistical Limits covers what that supports: direction at best, and only when the effect is large relative to how much the numbers move on their own.

A worked illustration, with entirely hypothetical figures — the point is the arithmetic, not the values. Say a competitor’s tracked keyword set averages position 14, and typical week-to-week drift on that set is 3 positions. If they gain 40 referring domains and the average moves to 11, the movement is one drift-width. If it moves to 4, that’s a change worth explaining — and it still doesn’t establish which of their changes explains it.

The general point: know the noise level before reading the signal, and on external data you often can’t even establish the noise level, because you’re sampling their rankings through your own tracker with its own variance.

What the data is actually good for

Three uses that don’t require attribution at all.

Finding pages worth citing. A competitor’s referring domains are a list of publications, roundups and resource pages that have demonstrably linked to something in your category. That’s a prospect list, and its value doesn’t depend on knowing what those links did for them. This is the whole content of What a Link Gap Analysis Supports — the output is a work list, not a diagnosis.

Understanding what gets cited in your space. Sort their pages by referring domains and read the top few. Whatever they are — a dataset, a calculator, a definitive explainer, a piece of original research — is evidence about what earns citations from this particular set of publishers. Far more actionable than the count.

Recognising a profile you shouldn’t emulate. A competitor whose link growth is concentrated in a handful of hosts with high authority scores and negligible traffic has a profile that tells you something about them and nothing you’d want to copy. Read the growth with Anchor Text Distribution as a Dataset and Referring Domains Versus Total Backlinks — the shape is usually obvious.

Reporting it without the causal claim

If competitor link data is going into a report, the defensible version separates observation from inference explicitly:

Observed: competitor X’s referring domains grew from A to B between these dates, per this index, retrieved on this date.
Observed: their visibility on this tracked keyword set moved from C to D over the same period, per this tracker.
Not established: whether one caused the other. Content, technical and commercial changes are unobserved, and result-set-wide movement in the period was E.
Actionable: these N referring domains are relevant to us and are a prospect list.

That structure is the standard in Building a Link Report You Can Defend, and it’s the version that survives someone asking how you know. The alternative — “they gained links, they gained rankings, therefore” — reads as stronger and collapses under the first question, which is why it’s a bad thing to build a budget request on.