Usually because you wrote both pages for the same reader with the same question in mind. Google then has to pick one of them. It may pick the weaker page, or it may swap between them week to week. Your rankings look shaky, and the page you actually care about never settles into place.
We see this on almost every site we take over. A blog post from two years ago covers a topic. A newer service page covers the same thing in different words. Nobody ever retired the old post. Now both sit in the index, and neither one wins.
This is called keyword cannibalization. It is fixable, but the fix is not always "delete the old page". In our work, the wrong fix costs more traffic than the original problem did. So the first job is to tell a real conflict apart from a false alarm.
Keyword cannibalization is when two or more pages on one site target the same search intent, so they compete with each other in search results. The site splits its own signals across those pages. Links, clicks, and topical relevance land on two URLs instead of one, and neither URL gets strong enough to rank well.
The word makes it sound like a penalty. It is not. Google Search Central states plainly in its SEO Starter Guide that "duplicate content on your site is not a violation of our spam policies, but it can be a bad user experience." Nothing is deducted from your site. The pages simply dilute each other.
The important word in that definition is intent, not keyword. Two pages can share a keyword and still be fine if they answer different questions. Two pages can use completely different words and still cannibalize if they chase the same job. We go deeper on that split in our guide to search intent and how to match it.
Search engines have gotten better at sorting this out on their own. That is exactly why a manual audit can do damage. You can spend a weekend fixing a problem the algorithm had already resolved in your favour.
No, and this is where most audits go wrong. Ranking twice for one query is often a win rather than a bug. Google's own systems allow it. Before you merge anything, you have to prove the two pages are hurting each other instead of just sharing a results page.
Ahrefs published a study on multiple rankings in February 2024 that changed how we look at this. Their team found roughly 9,700 cases of multiple rankings on the Ahrefs.com domain alone. They then hand-checked a sample of 80 of those keywords. Only one of the 80 turned out to be a case that actually needed fixing. The rest were the site taking up more space, not less.
Google's own documentation sets the ceiling. The Guide to Google Search Ranking Systems says the site diversity system "works so that we generally won't show more than two web page listings from the same site in our top results". Two listings is the normal maximum. If you already hold two spots on page one, you are at the cap, and merging them can drop you to one.
Our rule is simple. Cannibalization is a problem when the wrong page ranks, when the ranking page flips back and forth, or when both pages sit stuck outside the top ten. Two stable, useful listings is a good day, not a defect.
Open the Performance report, filter to a single query, then add the Pages dimension. Search Console will show every URL that has picked up impressions for that query. If two or more of your URLs share meaningful impressions on the same query, you have a candidate. Export that list and repeat for your most valuable queries.
The signal we care about most is instability. Sort by date and look at which page holds the position over eight to twelve weeks. A clean result shows one page owning the query. A conflict shows the URL changing month to month while average position drifts down. That flip-flop pattern is the real tell, and the date comparison view surfaces it faster than any third-party tool.
Crawlers help build the shortlist. Screaming Frog and Sitebulb both flag pages with near-identical title tags and H1 headings, which is the cheapest proxy for overlap. Semrush and Ahrefs both ship cannibalization reports that group ranking URLs by keyword. We treat all of those as a shortlist generator and never as the verdict. The verdict comes from Search Console, because that is your own data. If the tool is not set up properly yet, our guide to using Google Search Console walks through the reports we rely on.
Merge when both pages answer the same question, both are thin, and neither ranks well alone. Take the stronger URL, move the genuinely useful parts of the weaker page into it, and redirect the weaker URL to the stronger one. One strong page beats two weak ones almost every time.
Pick the survivor on evidence, not on how much you like the writing. We look at three things. The first is which URL has more referring domains. The second is which has the longer history in the index. The third is which URL structure makes more sense for the site going forward. Usually those three agree. When they disagree, backlinks win, because links are the hardest signal to rebuild from scratch.
Then do the merge properly. Pasting the old page into the new one produces a bloated result that reads like two articles stapled together. We rewrite the merged page as one piece with one argument. Anything that does not serve the new page gets cut, even if it read fine in the old one.
Keep both when each page serves a different reader or a different stage of the buying decision. A commercial service page and an educational blog post can both live on the same topic. The fix is not deletion. It is making the difference obvious in the title tag, the H1, the opening paragraph, and the internal links.
This is the option most audits skip, and it is often the right one. A services page and a "how much does this cost" guide overlap in vocabulary but not in job. One exists to convert. One exists to inform. Rewriting each opening so the intent is unmistakable usually resolves the flip-flop on its own, with no redirects at all.
Internal links carry the same message. Send commercial anchor text to the commercial page and informational anchor text to the guide. Google reads those links as a statement about what each page is for. When your own site is inconsistent about which page answers which question, you cannot expect a search engine to be more decisive than you were.
Use a 301 redirect when one page should stop existing. Use a rel="canonical" annotation when both pages must stay live but only one should rank. Google treats both as strong canonicalization signals, and it warns against sending conflicting signals by pointing different methods at different URLs.
Google's documentation on consolidating duplicate URLs describes a redirect as "a strong signal that the target of the redirect should become canonical" and a rel="canonical" annotation as "a strong signal that the specified URL should become canonical". Sitemap inclusion is listed as only "a weak signal that helps the URLs that are included in a sitemap become canonical". So a sitemap entry will never rescue a page you have canonicalized away.
The same page carries the warning we quote to clients most often: "Don't specify different URLs as canonical for the same page using different canonicalization techniques." We have inherited sites where the canonical tag pointed one way, the sitemap pointed another, and an old redirect pointed a third. Google picked its own winner, and it was not the page anybody wanted. Our guide to canonical URLs and duplicate content covers the tag itself in more detail.
Google also notes that these methods stack and become more effective when combined. So the clean version of a merge is a 301 redirect on the retired URL, a self-referencing canonical on the survivor, and only the survivor listed in your XML sitemap. All three saying the same thing.
We think it does, and the mechanism is the same. Answer engines summarize and cite a small number of sources for each question. When your own pages split the signal, you make it harder for any single URL of yours to be the obvious best answer. The fix does not change: one clear page per question.
Here is what we can state as fact rather than opinion. Google's site diversity system already caps how many of your pages show in top results. Google AI Overviews, ChatGPT, and Perplexity all build answers from retrieved pages, so the pool they draw from is shaped by the same ranking and canonicalization rules that shape blue links. If Google has folded two of your URLs into one canonical, that canonical is the URL in play.
Our practical read is that answer engine optimization raises the cost of an unclear page. A human reader can skim two similar articles and work out which one they wanted. A model picking one source to cite does not do that work on your behalf. So we now write one canonical page per question and defend it, rather than covering the same ground three different ways.
Expect four to eight weeks on most sites. Google has to recrawl both URLs, process the redirect or the canonical, and then move the consolidated signals across. Small sites can settle faster. Large sites with slow crawl rates take longer. None of this is instant, and checking daily will only make you anxious.
Speed it up where you can. Submit the surviving URL through the URL Inspection tool in Search Console. Keep the old URL reachable by the crawler for a while so Google actually sees the redirect instead of a dead end. Update your internal links to point at the survivor, because those are the fastest signal you fully control.
Then measure the right thing. Do not compare the new page against the old page in isolation. Add the impressions and clicks of both old URLs together, and compare that combined total against the merged page. We have seen merges that looked like a loss on a single URL and a clear gain on the number that actually mattered.
Start with one query rather than a full audit. Pick the commercial keyword that matters most to your business, check it in Search Console, and see whether one page owns it. If two pages are trading places, fix that single conflict properly. Then repeat. A slow, evidence-led cleanup beats a mass deletion every time.
The mistake we most want you to avoid is the bulk merge. Somebody runs a tool, gets a list of forty flagged pages, and redirects half the blog in an afternoon. Traffic drops, and now there is no clean way to tell which redirect caused it. Fix one, wait, measure, then fix the next one.
If you are staring at a Search Console export and cannot tell whether your pages are competing or simply co-existing, we are happy to walk through it with you. This cleanup is part of almost every rebuild we do, and it is usually the cheapest ranking win available. Let's talk. You can reach us at phoenix.studio.
Tell us where you want to go. We'll tell you how we'd get you there.