Keyword Cannibalisation Looks Like Flapping

The signature is a position that alternates between two distinct levels rather than wandering. Not noise around a centre — two centres, and the series jumping between them. When you look at which URL of yours is ranking on each check, it’s a different one each time.

That pattern has a specific cause worth knowing on sight, because it’s one of the few ranking problems that’s genuinely fixable by decision rather than by effort.

What you’re seeing

Two or more of your pages are plausible answers to one query. Which one gets shown is being decided per retrieval, and the two pages don’t rank equally well, so the reported position lands on one of two levels depending on which was chosen.

A tracker that records only your best position, and not the URL, shows this as inexplicable volatility. A tracker that records the ranking URL shows it immediately — the position column and the URL column change together.

This is why the URL field is the most valuable column in a rank tracker after the position itself, and why it’s worth checking that your tool records it.

Distinguishing it from ordinary volatility

Ordinary volatility wanders around one centre. Values cluster, and the distribution has one peak. The URL doesn’t change.

Flapping has two clusters with a gap between them. The URL alternates.

A genuine ranking change is a step: one level, then another level, and it stays. The URL may or may not change.

A tool artefact is usually a single-check spike with a return, or a suspicious flat stretch. The URL is unchanged or blank.

The two-cluster shape is the tell, and you can see it in a raw series without any statistics. If your tool draws a line chart, flapping looks like a square wave rather than a fuzzy line. Baselines make it obvious, since the “normal range” turns out to be bimodal — you can’t read a position without a baseline.

Why it’s a real problem and not just a display issue

Three consequences worth naming.

The worse page sometimes wins. Whatever page you’d have chosen, you’re not choosing — and roughly half the impressions go to the one you didn’t want.

Diagnosis gets harder. Any change you make to one page produces an ambiguous result, because the measured position depends on which page was picked. You cannot evaluate your own work.

Reporting gets harder. The average across a flapping keyword is a number between two levels the keyword never occupies, which is the ordinal-averaging problem in a particularly annoying form — position is an ordinal, not a quantity.

Confirming it before acting

Flapping is diagnostic of competing pages, but check before restructuring anything.

Look at the URL history for that keyword over as long a window as you have. Two URLs alternating is confirmation. One URL flapping is not this problem.

Check Search Console for the query and see how many of your pages received impressions for it. Two of your URLs both accumulating impressions on one query is the same finding from the other direction.

Check whether the pages genuinely overlap. Sometimes two pages answer different intents that happen to share a query, in which case the answer isn’t to merge them.

Check the canonical relationships. Occasionally this is a directive problem rather than a content problem — two pages that were supposed to be one, with the canonical not doing what someone assumed.

What the fix looks like

In rough order of preference: pick the page that should rank and make it clearly the better answer for that query; consolidate the weaker page into it if they genuinely say the same thing; differentiate them if they’re actually about different things and the overlap is accidental; or, last, use directives to express which one you mean.

Whatever you do, the measurement matters afterwards. Recovery from a consolidation takes time and the series will be messy while it settles, so expect a period where neither the old nor the new pattern holds. Note the change on the chart, because you’ve just invalidated that keyword’s baseline.

And take the keyword off any alerting for a few weeks, since it’ll fire on the transition — rank alerts that don’t cry wolf.

What to actually do

1. Make sure your tracker records the ranking URL, not just the position.
2. Look for two clusters, not wide scatter, when a keyword seems volatile.
3. Confirm with the URL history and with Search Console before restructuring.
4. Check whether the overlap is real — some near-duplicate queries genuinely have two intents.
5. Annotate the fix and reset the baseline, and don’t read the first few weeks after a consolidation as a result.