A Keyword Cannibalization Audit That Explained a Flat Traffic Chart

The client's traffic chart was the strangest kind of flat — not declining, not growing, just a perfectly straight line for twelve months despite publishing two to three new articles a week the entire time. Something was eating the gains as fast as they were created.

We pulled every page targeting overlapping primary keywords and found 34 pairs and triplets of pages all competing for the same core terms, built up gradually because different writers over different quarters didn't know a page on that topic already existed. None of them individually looked like a problem. Together, they were splitting ranking signals three or four ways instead of consolidating them onto one strong page.

How cannibalization actually shows up

It rarely looks like one page stealing another's rankings outright. More often it looks like Google oscillating which of the competing pages it shows for a given query week to week, with both landing somewhere in the teens instead of either one breaking into the top five — which is exactly the pattern we found across most of the 34 pairs.

Page A (published 2024): ranks #14-19, fluctuating weekly
Page B (published 2025, same topic): ranks #11-22, fluctuating weekly

Neither wins consistently. Combined traffic < what one strong page would get.

Cannibalization doesn't look like losing. It looks like two mediocre rankings pretending to be a strategy.

The fix and the result

• Merged the weaker page's unique content into the stronger page and 301-redirected it
• Where both pages had genuinely different angles worth keeping, differentiated the target keyword and internal linking so they stopped competing
• Built a shared keyword map that writers check before pitching new topics, to stop the pattern recurring

Within four months of consolidating the clearest cases, overall organic traffic grew about 18% with zero new content published in that window — the content had been there the whole time, just fighting itself instead of ranking.