Fixing a keyword cannibalization problem: 14.5K to 82.7K monthly clicks in three months

The problem

Through November 2025, the site pulled over 330,000 monthly clicks at an average position of 6.3. Then in December, Google started oscillating between the homepage and the /product-a/ landing page for the same search terms. The homepage was getting 209,500 clicks at position 17.9. The dedicated landing page sat at position 7.8 and pulled only 9,489 clicks. Neither page could win because they were splitting authority. By January, site-wide clicks had fallen to 208,000, a loss of 122,000 visitors compared to November.

An additional finding

The cannibalization audit surfaced a second problem: the link-building team had been sending nearly all backlinks to the homepage with branded and semantic anchor text for two years. The company's own brand terms dominated the anchor profile. Product-intent anchors like "Keyword A" or "Keyword B" were a small minority.

That imbalance hurt the homepage's product rankings. It also created headroom for the link-building component of both plans. Plan A called for building exact-match anchors ("Keyword A") to the homepage to reinforce the canonical signal. If Plan A failed and Plan B ran, those same anchors would instead point to /product-a/. Either way, the 62% branded buffer meant the site's overall anchor footprint wouldn't tip into over-optimization territory. The concentrated exact-match push was a calculated risk, backed by that 62% buffer.

Only 14% product exact-match across 910 referring domains gave significant headroom to run an exact-match campaign (to the homepage under Plan A, or to /product-a/ under Plan B) without triggering over-optimization signals.

The strategy

The approach: start with Plan A, move to Plan B if it didn't work. Plan A went first because deoptimizing the homepage carries real risk; it's the single most important page on the site, and stripping its product signals is not something to do on a hunch. Consolidating onto it was the lower-risk move to try first. Both plans were scoped before Plan A launched so the pivot required no re-planning. Plan A was implemented February 20; the formal proposal documenting both plans was completed in March 2026.

A third option, a 301 redirect collapsing /product-a/ into the homepage, was scoped as a last resort. Plan B worked first, so it was never needed.

Plan A results

The canonical worked. By March 2026, traffic recovered to 331,000 clicks and average position improved to 7.2. The February 2026 keyword data confirmed the mechanism at the keyword level.

Source: Ahrefs, March 16, 2026. /product-a/ lost these keywords entirely after the canonical was processed.

Then Google ran two core updates back-to-back. The March 2026 core update rolled out March 27 and completed April 8. The April 2026 update followed within days. During that window, Google stopped honoring the canonical tag. Canonical tags are hints. Google overrode them wherever it had its own view of which URL should rank. Traffic fell again. May came in at 225,000 clicks.

The homepage copy refresh, the final piece of Plan A's roadmap, was ready to deploy. Core updates shelved it. Reinforcing the homepage's copy would have fought the same battle twice. Plan B was already scoped. The team activated it without re-planning.

Plan B results

With Keyword A signals stripped from the homepage, the full link-building budget redirected to the spoke pages, and competing content removed from the site, Google had an unambiguous answer. The homepage stopped competing for terms it shouldn't rank for. The landing pages started ranking for what they were built for. The AEO outreach added a second channel: placements in the sources that AI models cite, building brand presence in both traditional search and AI-generated responses simultaneously.

Non-branded keyword footprint

The hub-and-spoke structure is working. September (full month) delivered 45,928 clicks on /product-a/ and 36,774 on /product-b/, up from 6,180 and 8,378 in June. Below are the Sep 30 Ahrefs positions for the non-branded keyword sets on each page.

Source: Ahrefs, Sep 30, 2026.

September: the payoff

September closed as the strongest month for spoke page traffic since the strategy started. /product-a/ finished at 45,928 clicks, 7.4× its June level. /product-b/ hit 36,774, also 4.4× June. Homepage held at 107,145 clicks with an average position of 2.5, continuing to consolidate brand search authority. Total from the three core pages: 189,847 clicks.

The three-month execution window (July to September) is what drove this: link-building fully redirected to spoke pages, secondary cannibalization pockets cleaned up, and AEO citation outreach running in parallel. On the Ahrefs side, "Keyword A" closed September at position 7 on /product-a/ (it had dipped to 17 mid-month as Google re-evaluated, then recovered sharply). "Keyword B variant 1" (a 37K/mo term) reached position 6 on /product-b/, up from position 23 in June.

How the keyword transfer played out

Lime lines are the target spoke pages. Gray dashed lines are the homepage. Rising toward the bottom of each chart is the intended outcome of Plan B. The green band marks top 10. By July 2026, the homepage no longer appears in its top-50 keywords for "Keyword A variant 4" or "Keyword A" at all; only brand terms remain. The spoke pages have absorbed both.

• Cannibalization identified
• Canonical live (Feb 20)
• Core updates
• Plan B implemented

Why a lite tool was part of Plan B

Competitive analysis ran alongside the deoptimization work and surfaced a consistent pattern: every major competitor offered a working version of the core tool directly on their spoke pages. The company's /product-a/ and /product-b/ pages had no tool, just content. The initial concern from leadership: give users a free working tool and they won't sign up for the paid product.

The solution was a lite version with intentional limits, built to engage users without substituting for the paid product. Google weighs dwell time and on-page engagement as quality signals. A page about this kind of tool where users read and leave scores worse than one where they actually interact with something. Competitors had already figured this out. The lite tool was added to both spoke pages as part of the Plan B rollout. Dwell time went up, engagement improved, and rankings followed.

What this case illustrates

1. 01

   Cannibalization looks like a penalty

   The traffic drop had the same signature as an algorithm hit. Identifying it required looking at which pages were ranking for which terms and noticing the oscillation between them.

2. 02

   Build the fallback before you need it

   Plan A was right. The core update breaking it wasn't predictable. Having Plan B already scoped meant no delay when the pivot became necessary.

3. 03

   Audit the anchor profile first

   Knowing that 62% of refdomains used branded anchors made the exact-match campaign a calculated risk with the numbers behind it. The buffer already existed; it just needed to be found.