One Intent, One URL: Cleaning Up Keyword Cannibalization Before Answer Engines Choose a Favorite
Cannibalization used to be a ranking problem. Two pages chasing one query meant split signals, alternating positions, and a slow drift downward that was annoying but survivable. In 2026 the cost is higher. When a generative answer assembles its response from multiple sources, a publisher that presents the same claim on four different URLs has given the system four weak witnesses instead of one strong one. The likely outcome is that none of them get cited.
The cleanup below is evergreen in structure and current in emphasis: the goal is not merely to consolidate rankings, but to consolidate evidence so that a single URL is unambiguously the authoritative source for a given intent.
Step 1: Cluster by Intent, Not by Keyword
Keyword-level duplicate detection produces long lists that are mostly noise. Two pages ranking for the same phrase may serve entirely different needs — one answering "what is it," the other "how do I file it." Those are not competitors; they are neighbors.
Cluster instead by the question a reader arrives with. Pull the queries each page earns impressions for, group them into intent families, and look for pages whose intent families overlap by more than roughly 60 percent. Sort each cluster by clicks and by conversion or subscription value. The clusters that combine high overlap with high value are your true cleanup candidates; everything else can wait.
A useful sanity check: if you cannot write a single sentence describing what the cluster's reader wants, the pages probably do not belong together after all.
Step 2: Score Each URL on Evidence Strength
Before deciding anything, grade every page in the cluster on four dimensions.
- Distinct facts. Original data, dated statistics, named sources, first-hand testing, expert quotation.
- Topical completeness. Whether the page resolves the reader's question without requiring a second click.
- Technical standing. Indexability, canonical correctness, internal link equity, page experience.
- Age and maintenance. Whether it has been reviewed within the last twelve months and whether its claims still hold.
Score on a simple scale and be honest about it. Teams routinely protect their newest page when the older one carries the actual evidence. Answer engines reward the page that can be verified, not the one that was published most recently.
Step 3: Choose Merge, Redirect, or Rewrite
The decision tree is narrower than most audits suggest.
Merge when two or more pages each hold unique fragments — a stat here, a worked example there — and no single page is complete. Combine the fragments into the strongest URL, keeping the original sourcing intact and noting the revision date.
Redirect when one page is clearly dominant on all four scoring dimensions and the others add nothing substantive. Consolidate with a permanent redirect, and only after confirming the target genuinely covers the retired page's intent. Redirecting into a page that answers a different question transfers confusion rather than authority.
Rewrite when the winning URL is structurally sound but thin, outdated, or written for a different audience. Keep the URL, replace the substance, and align the headings with the intent family you identified in step one.
What you should almost never do is leave both pages live with light edits and hope the search engine sorts it out. That is the arrangement that produces split evidence.
Step 4: Consolidate the Evidence Itself
This is the step that matters most in an AI-answer environment and the one most often skipped. Consolidation of URLs is not consolidation of content.
Take the surviving page and audit every factual claim. Remove hedged duplicates — the same statistic stated three ways across three sections. Pick one figure, cite its source, and date it. Replace vague attributions with named ones. Where the merged content brought in competing numbers, reconcile them explicitly rather than listing both, because a page that contradicts itself gives an answer engine a reason to look elsewhere.
Add the details that make a page citable: methodology notes, sample sizes, definitions of terms, and a clear statement of what the page does and does not cover. Specificity is what distinguishes a source from a summary.
Step 5: Rewire Internal Links and Re-Check
Cannibalization survives cleanup through internal links. Run a site search for every retired URL and update the anchors to point at the survivor. Where the anchor text used a retired page's phrasing, revise it to match the consolidated intent. Remove the old URL from sitemaps and any navigation, related-content modules, or inline callouts.
Then wait. Give the change four to six weeks before evaluating, and measure three things: the cluster's combined impressions, the survivor's citation frequency in AI-generated answers, and whether the retired intent still draws traffic through the redirect as expected. If the cluster's total impressions fall while the survivor's share rises, the consolidation worked — you are seeing the same audience through fewer doors.
A Maintenance Cadence That Prevents the Next Mess
Set a quarterly review tied to content publishing, not to calendar quarters. Before any new page ships, require the author to name the existing page it is closest to and explain why it is not a duplicate. That single question prevents most future cannibalization and costs almost nothing to ask.
Consolidation is unglamorous work with a measurable payoff: fewer pages, stronger evidence, and a clearer answer to the question of which URL speaks for your publication on a given topic.