Post-mortem: how four near-duplicate pages cost one site its best keyword

Cannibalization is an intent collision, not a word collision

Most definitions of keyword cannibalization are wrong in a way that makes the problem harder to fix. It is not two pages that mention the same keyword. Every page on an SEO site mentions SEO. That's a topic, not a fault.

Cannibalization is two or more URLs on your own domain that satisfy the same search intent. Same question, same expected answer, same page type. Google has to pick one, and because the two are so close, the pick isn't stable. It flips with each recrawl, each minor edit, each wobble in whichever signal was breaking the tie.

That instability is the damage, and it's why the problem is so easy to miss. A page sitting steadily at position 6 accumulates clicks, engagement and — eventually — links. A query where your position swaps between two URLs every fortnight accumulates none of that, because neither URL is ever around long enough to become the thing people bookmark or cite. You don't see a crash. You see a page that never quite arrives.

How a content calendar builds this without anybody making a mistake

What follows is a composite. We've run this cleanup often enough that the details repeat, so we've reconstructed one version of the pattern rather than attach numbers to an account we'd have to anonymise anyway. Nothing here is a figure lifted from a specific client.

A B2B services company starts publishing. Month 2, the pillar page goes live — the definitive guide to the thing they sell. It's good. It ranks. Month 9, a new writer joins, gets handed a keyword-volume export, and briefs a beginner's guide against a variant of the same phrase. Also good. Also live. Month 14, someone reads that checklists earn links, so a checklist post ships against the same query with the word "checklist" appended. Month 22, a marketing manager refreshes the original pillar, and — because the CMS makes it easy and the year looks tidy in a slug — publishes the refresh at a new URL. The old one stays up.

Four URLs. One intent. Twenty-two months. Nobody did anything stupid; every one of those four commissions was defensible on the day it was made. The failure is structural. No document anywhere said which URL owned that intent, so four briefs were allowed to answer the same question, and the site quietly started competing with itself for the only keyword that was ever going to pay for the content programme.

The symptom: URL flip-flop in Search Console

Search Console will not hand you a cannibalization report. It hands you the raw material, and the whole diagnosis lives in the relationship between one query and the pages that serve it. Five minutes, no tools, no subscription.

Do this for the ten queries you most need to own. Not all of them — the ten that matter.

1. Open the Performance report, set the range to the last 16 months, and stay on the Queries tab. Pick a query you believe you should own.
2. Click into that query, then switch to the Pages tab. You are now looking at every URL that earned an impression for that single query. More than one URL with real impressions is your candidate cluster.
3. Export it by week or month and chart average position per URL. Cannibalization has a shape: URL A holds position 7 for three weeks, URL B takes over at 11, A comes back at 9. A healthy page is a flat line with a slope.
4. Check click-through rate against what that position should return. A split intent usually under-performs its position, because the URL Google surfaced is frequently the wrong one for the query the person typed.
5. Sanity-check with site:yourdomain.com "the exact phrase" in Google. Crude, unreliable as a ranking read, but it surfaces the candidate set in about eight seconds.

Merge, redirect, differentiate or delete — the decision rules

Once you have the cluster, each URL in it gets exactly one of four outcomes. Choosing wrong is considerably more expensive than choosing slowly, so this is the part to think about over a coffee rather than in a sprint planning session.

The default instinct — delete the duplicates — is the one to resist hardest. A deleted URL takes its links with it.

• A canonical tag is not a substitute for a 301 when the page should die. Canonical is a hint; a redirect is an instruction. Use canonical when both URLs must stay reachable for users. See canonical versus noindex for which tool fits which situation.
• Never blanket-redirect a cluster to the home page. Google treats an irrelevant redirect as a soft 404 and the link equity evaporates. One-to-one or nothing.
• Don't differentiate two pages that genuinely have one intent. Rewriting titles to look different doesn't create a second reader.

How we pick the survivor

The obvious move is to keep the page with the most traffic. That's the wrong first criterion roughly a third of the time, because traffic is a symptom of the split you're trying to fix and it can be sitting on the wrong URL for accidental reasons — an internal link from the nav, a slightly older publish date.

We rank the candidates in this order, and we stop at the first clear winner.

1. Referring domains pointing at the URL. Links are the single hardest thing on this list to rebuild. If one URL has eleven referring domains and the other has one, the argument is usually over.
2. Rank history depth. Which URL held a position for the longest stretch? Longevity suggests Google has already decided this URL is the one, and agreeing with that decision is cheaper than fighting it.
3. URL fit. Does the slug describe the intent that's actually winning? A slug reading /blog/2023-beginners-guide-updated is a bad long-term home for your best commercial query.
4. Page-level conversion, if you track it. Two pages can rank identically and convert very differently.
5. Content quality — last. Yes, last. Better content can be moved onto the surviving URL in an afternoon. Earning those links again is a quarter's work.

The 301 plan, in order

Sequence matters here more than it looks. Redirecting first and merging content later means Google recrawls a destination page that doesn't yet answer the query it inherited, and the consolidation reads as a downgrade.

1. Fold the content first. Move anything genuinely useful from the losing pages onto the survivor. Publish it. Let it sit for three to seven days so the improved page gets recrawled on its own terms.
2. Then 301 the losers, one-to-one, each to the survivor. Server-level redirects, not a plugin chain if you can avoid it.
3. Rewrite every internal link that pointed at the redirected URLs. This gets its own section below because it's where most consolidations quietly fail.
4. Fix the sitemap. Remove the redirected URLs, keep the survivor, resubmit. A sitemap that still lists dead URLs slows the recrawl.
5. Update what you control off-site. Guest post bylines, email footers, Google Business Profile links, social bios, the PDF someone sent to 400 people.
6. Request indexing on the survivor in Search Console. It's a nudge, not a lever, and the quota is small — use it on the one URL that matters.
7. Wait four to eight weeks before judging it. Watch impressions for the target query on the survivor, not total site traffic.

Rewriting internal links is the step everyone skips

301s work. They also stop being a fix and start being technical debt the moment you rely on them as architecture instead of as a transition.

Three reasons to do the rewrite properly. First, every redirected internal link is an extra hop for a crawler, and hops compound into chains the next time somebody restructures. Second, chains beyond a couple of hops get treated with less patience than a direct link. Third — and this is the one that actually costs rankings — your internal linking is a statement about which URL you consider canonical. If your navigation, your related-posts block and forty old articles all still point at the URL you just retired, you have told Google two contradictory things about the same page. Contradictory signals are how a consolidation half-works, which is the worst available outcome: you've taken the disruption and kept the ambiguity.

Finding them is unglamorous and finite. Crawl the site and filter for links resolving with a 3xx. Search your CMS body content for the old slug string. Then check the places a crawler flags but people forget: navigation and footer templates, sidebar widgets, hardcoded links inside components, author bios, and the inline links in old posts that nobody has opened since 2024.

While you're in there, point the recovered internal links at the survivor with anchor text that describes the query it now owns. Consolidation is one of the few moments you get to rewrite a site's internal linking with full context — internal linking beats link building more often than founders expect, and this is the cheapest chance you'll get to prove it.

The keyword map that stops it happening again

The cleanup is a week. The prevention is an afternoon, and it's a spreadsheet — which is why almost nobody builds it.

Four columns: query cluster, owning URL, page type, owner (a named human). Every query you intend to win appears exactly once. Every commissioned brief cites the row it belongs to. That's the whole system, and it's the cheapest insurance in content marketing.

• One intent, one URL — written down before the brief is written, not after the second page ranks.
• A refresh never gets a new URL. Update in place. The publish date can change; the slug cannot. This single rule would have prevented the month-22 page in the story above.
• No row, no brief. If a writer wants to cover something that isn't on the map, the map gets a new row first, and adding that row forces the conversation about overlap.
• Quarterly flip-flop check. Pull your top 30 queries, look at the Pages tab for each, flag any query with two URLs earning impressions. Fifteen minutes a quarter.
• Kill the year from your slugs. /guide-2025 guarantees somebody publishes /guide-2026 next January, and now you're back here.