Most Shopify Keyword Cannibalization Isn't Cannibalization

Run a cannibalization audit on a Shopify store with 800 products and the tool hands back something like 340 flagged pairs. Open the first twenty. They are the same t-shirt in six colours, the same mug with two prints, and two collections that both happen to contain the word "gifts".

None of that is broken. The tool is describing a catalog.

The awkward part is that cannibalization is real, it does cost stores traffic, and it is buried somewhere in that list of 340. Finding it means throwing away most of what the audit flagged. That feels wrong, so merchants tend to either fix everything or fix nothing, and both are worse than the ten minutes it takes to check properly.

Google is not penalising you for any of this

Worth killing the underlying fear first, because it drives most of the bad decisions here.

There is no duplicate content penalty. Google's canonicalization documentation says so in as many words:

Some duplicate content on a site is normal and it's not a violation of Google's spam policies.

John Mueller has been repeating the same point for years. The most quotable version came in January 2021, when he described duplicate content across pages and said "it's not so much that there's a negative score associated with it".

What actually happens is duller than a penalty. Google groups the URLs it considers duplicates, picks one to represent the set, and shows that one. The others get filtered out of the results. Nothing is deducted from anything. You just do not get to make the choice. Duplicate content is a related problem with its own fixes, worked through here, and it is not the same thing as two pages competing for one query. This post is about the second one.

That reframing changes the job. You are not defusing a punishment. You are trying to stop a search engine from choosing badly, or from choosing again every other week.

The advice you are reading was written for blogs

Almost every cannibalization guide assumes a content site. On a blog, two posts targeting "best running shoes for flat feet" is always a mistake. Someone wrote the second one without checking the archive. One intent, one page, consolidate, move on.

A catalog does not work that way. Forty t-shirt pages that differ by one word are not an error, they are stock. Deleting thirty-nine of them to satisfy an audit tool would be self-harm. The blog rule, transplanted into ecommerce, produces advice like "make every product description unique", which is fine advice for other reasons and almost never the thing that fixes a cannibalization problem.

The useful split is not duplicate versus unique. It is which kinds of pages are overlapping.

Same-type overlap is usually a structure question

When two products read as near-identical, the interesting question is rarely which one to delete. Ask instead whether they should ever have been two products.

Google's ecommerce URL guidance is that each variant should be identifiable by a separate URL, and that where variants are expressed as an optional query parameter, the canonical should be the URL with the parameter stripped. Shopify already does this. A product reached at /collections/winter/products/merino-crew carries a canonical pointing at /products/merino-crew, and the ?variant= version resolves the same way. View source on any product page and you can confirm it in about ten seconds.

The stores that get into trouble are the ones where colours became separate products. That usually happened for reasons with nothing to do with search: a supplier feed imported that way, or the theme's swatch display was easier to build with distinct handles. Now there are twelve product pages for one garment, each self-canonical, all of them live.

That is a structural problem wearing a cannibalization costume. Rewriting the meta description on all twelve will not touch it. Merging them into variants will.

Cross-type overlap is where the real cases live

The pairs worth your time are usually a product page and a collection page that have quietly converged.

Not every such pair is a problem. A collection called "Merino Base Layers" holding nine products, and a product page for one specific base layer, are not competing even though a keyword tool will insist they are. One is browse, the other is buy, and having both rank is the outcome you were hoping for.

The trouble is the collection that contains a single product. Or two collections built around slightly different phrasings that return an identical product set. Or a product page that, after four years of copy edits, has grown into a category explainer with a buy button at the bottom. Those pairs genuinely answer the same question, and something has to choose between them.

Keyword overlap is a poor way to find them

Most tools compare title tags, H1s and target keywords. That combination misses the pairs that matter and flags the ones that do not.

It misses them because two pages can chase one intent in completely different vocabulary. A product page built around "rain shell" and a collection built around "waterproof jackets" share almost no keywords and compete head on. It over-flags because Shopify themes generate titles from templates, so half a catalog matches the other half by construction.

Comparing the body content by semantic similarity instead of shared keywords catches the rain shell case. The catch is where you set the bar. Anywhere near "these two pages are broadly about the same sort of thing" and a 900-product store returns most of itself, which drops you back exactly where the keyword tool left you. The threshold has to be high enough that a match is genuinely surprising.

Even then, similarity only produces a candidate list. It tells you two pages read the same. It cannot tell you whether anyone searches for the thing they both read as, or whether Google is having any trouble telling them apart. That part is not visible in your content.

Search Console can settle it, but not in the view you normally open

The Performance report will not show you cannibalization unless you ask a specific question, because of how position is calculated. Google's documentation for the report explains that the position it shows is the topmost position your site held, averaged across queries. If your product page sits at 6 and your collection at 11 for the same search, the report tells you 6. The second page is not in the number at all.

So ask it differently:

• Filter the Performance report to a single query, then switch to the Pages tab. You are now looking at every URL of yours that took impressions for that search.
• Healthy looks like one URL holding nearly all of them. Cannibalization looks like two URLs splitting the impressions with clicks landing on neither in any real volume.
• Run that same view across two or three separate date ranges. The signature is the swap. Google ranks the product page in March, the collection in April, the product page again in May. That instability is the actual cost, because neither page ever accumulates enough to settle the question.

It is a fiddly view to assemble, one query at a time, with no way to see the whole catalog at once. That is probably why so few people ever look.

The Page Indexing report carries a second tell. URLs sitting under "Duplicate, Google chose different canonical than user" are Google telling you it looked at your canonical and disagreed. It is entitled to. In Google's own wording, indicating a canonical preference is a hint, not a rule.

The fix menu, ordered by how often it is the right one

• Consolidate one page into the other and 301 the loser. Google's documentation calls a permanent redirect a strong signal that the redirect target should become canonical, and it is the only option here that also removes the dead end from a shopper's path. Correct when one of the two is genuinely redundant: a discontinued product, a collection nobody links to.
• Differentiate the intent and keep both. Let the collection do the comparing: sizing, how the nine options actually differ, which one suits which use. The product page can then stop explaining the category and get on with selling the item. More work than a redirect, and more often the right answer.
• Convert separate products into variants. The structural fix for the twelve-colour case. It is an afternoon in the admin and then it stops being a problem.
• Canonicalise one to the other. Fourth on purpose, because it is the first thing most people reach for. Google describes rel=canonical as a strong signal rather than an instruction, and it leaves both pages live for shoppers and internal links, which is occasionally what you want and frequently a way of not deciding.
• Do nothing.

If differentiation turns out to be the answer across several hundred products, that is bulk work, and Seokai's meta title and description tooling exists to do it. Its competing content report will also hand you the shortlist, comparing what pages mean rather than which words they share, and holding same-type near-duplicates to a stricter bar than product-against-collection overlap for exactly the reason this post opened with. What it cannot tell you is which of those pairs is costing you anything. The Search Console check above does that, and it takes ten minutes.

Doing nothing is right more often than SEO advice admits

Two pages splitting impressions on a query with forty searches a month is not worth a redirect. Consolidation carries costs no audit tool models: a 301 kills a landing page someone is running ads to, breaks a link in an email from 2023, and retires a URL your support team pastes into replies. If a pair has been stable for six months, one page clearly owns the query, and the other collects a trickle, that is a search engine doing its job rather than a problem you have.

Fix the pairs that flip. Leave the pairs that have settled.

Cannibalization is one of the few SEO problems where the diagnosis and the cure are the same act. You decide what each page is for. Most pages that compete with each other do so because nobody ever did.