How to Fix Pages Competing for the Same Query

To fix pages competing for the same query, start from the query instead of the pages. List every page on your site that shows up in the top 30 results for it, check whether two of them actually answer the same question, pick one to be the canonical answer, then either fold the other one in behind a 301 redirect or rewrite it to do a different job. Rewrite both pages, not just the one you kept, because a page that still opens with the same promise keeps competing. Then re-score the canonical page against the query and watch its rank on the searches ChatGPT, Gemini, and Perplexity run, for three to six weeks. Most content teams find two or three of these splits on any site older than a couple of years, and fixing one is usually cheaper than writing anything new.

Here's the version of this that shows up in a content meeting. Somebody notices that ChatGPT names a competitor for a question your site has covered twice. Someone else pulls up both of your pages and they're both fine. That's the confusing part. Nothing is broken, nobody wrote anything bad, and you still lose.

What's happening is a split. Two pages answering the same question don't stack up into a stronger signal. They divide one. Search engines have to choose which of your pages to rank, AI assistants have to choose which of your pages to cite, and when the two look interchangeable, the safest choice for the machine is to go with somebody else's page that has no such ambiguity.

This is an old SEO problem with a sharper edge. In classic search you could survive it, because a results page has ten slots and your two mediocre pages might both scrape onto it. An AI answer names one source per point. There's no second slot to land in.

This guide works with any AI visibility setup; the callouts show how each step runs in Spyglasses.

Step	What you're doing	What tells you it's working
1. Find the split	Listing your own pages on one query	Two or more of your URLs on the same search
2. Confirm it	Checking body and summary, not just topic	You can say what makes them the same
3. Pick the winner	Choosing one canonical page	One URL written down, with a reason
4. Choose the fix	Consolidate or differentiate	The call is made before editing starts
5. Rewrite both	Merging or repositioning	Neither page opens with the same promise
6. Re-score	Measuring after recrawl	One page rising, the other out of the way

Find the query your pages are splitting

The mistake almost everyone makes here is starting from the content inventory. You open a spreadsheet of 400 URLs, sort by title similarity, and try to spot duplicates. It takes a week and it finds the wrong things, because pages with similar titles are often fine and pages with different titles are often the ones fighting.

Start from the query instead. Pick a question you should be winning, one that matters to your business and that you've written about more than once. Then ask a simple thing: which of my pages come back for this?

If the answer is one page, you're done here. Go work on that page. If the answer is two or three of your own URLs sitting near each other in the results, you've found a split.

Do this on the searches that AI assistants actually run, not the ones you wish they ran. When somebody asks an assistant a question, the assistant turns it into several web searches with their own phrasings and qualifiers, and those are the searches your pages get retrieved for. A page can look great for the phrase in your content brief and never surface for the phrase the assistant used. If that distinction is new to you, winning a target AI grounding search covers the mechanics.

Expect to find two to four real splits on a mature site. Older sites that went through a rebrand, a product renaming, or a period of publishing three posts a week tend to have more. Content teams that inherited a blog from a previous agency almost always have some.

Confirm the overlap is real

Two pages on the same topic aren't automatically a problem. A category overview and a detailed how-to can both cover pricing without competing at all, because a reader who needs one doesn't want the other.

So check two things before you touch anything.

First, the body. Read the first 200 words of each page side by side. If they make the same promise to the same reader, that's a split. If one says "here's what this costs and why" and the other says "here's how to set it up," they're doing different jobs even though a topic model groups them together.

Second, the summary. Look at the title and meta description of both pages, which is what a retrieval system sees before it decides whether to open anything. Pages get flagged as duplicates far more often for having near-identical summaries than for having similar bodies. Two posts can be very different inside and still present themselves as the same page in the one place that determines whether either gets read.

Both signals have to agree before you call it. Body-only similarity gives you false alarms on any templated site, where every page shares navigation, boilerplate, and a footer full of the same words. That's why a single similarity check flags product catalogs where nothing is actually wrong. When the body and the summary both say the same thing, you have a real split.

Write down the sentence that describes the overlap. Something like "both of these promise a buyer's guide to the same product category, in the same voice, for the same reader." If you can't write it, you don't have a split, you have two pages on one topic, and the work you were about to do would have made things worse.

Pick the page that should win

Now pick one. Not "we'll see how it plays out." One URL, written down, with a reason next to it.

Three inputs, in this order:

Current position. Whichever page is already closer to the top for the target search has a head start you can't buy. It's been crawled more, it's been linked to more, and the algorithms already trust it more. Promoting the page you like better over the page that's already ranking costs you months.

Links and traffic. Check what points at each URL from the outside world and what each one already brings in. A page with three references from other sites is worth more than a page with none, even if the second one reads better. You can improve writing in an afternoon. You can't manufacture five years of links.

Fit with the query. If the target search asks for a comparison and one page is a comparison while the other is a product page, the comparison wins, even if it's newer and weaker. Fighting a page's format is harder than improving its content.

When the three inputs disagree, current position usually wins. It's the one you can't recreate.

One caution. Teams often want to pick the newest page because it matches the current brand voice and the old one is embarrassing. That's a real reason to rewrite something, but it's a bad reason to make it your canonical page. Keep the URL with the history and rewrite it into the voice you want. You get both.

Choose between consolidating and differentiating

There are two fixes and they lead to very different weeks of work, so make the call before anyone opens a doc.

Consolidating means the second page stops existing as a separate answer. You move whatever it does well into the canonical page and redirect the old URL. One page, one job, all the signal in one place.

Differentiating means both pages survive with clearly separate jobs. You rewrite the second page so it answers a different question, with a different title, a different opening promise, and a different reader in mind.

	Consolidate	Differentiate
Use when	Same reader, same moment	Different readers or stages
The second page has	Little traffic or links of its own	Its own audience worth keeping
The overlap lives in	The substance	The titles and intros
Work involved	Merge, redirect, fix internal links	Rewrite the second page's angle
Risk	Losing the redirect or the links	The two pages drift back together

A third move gets suggested in every one of these meetings, and it's worth naming so you can rule it out. Someone will propose a rel=canonical tag pointing the weaker page at the stronger one, or a noindex on the loser, as a way to avoid the work. Both are real tools and neither does what people hope here. A canonical tag is a hint that search engines can ignore, and it does nothing about the second page still sitting there making the same promise to any assistant that fetches it directly. Use them alongside a real fix, not instead of one.

Content teams reach for differentiating more often than they should, usually because consolidating feels like admitting a page was a mistake. Here's a test that settles it. Write a one-sentence job description for each page, in the form "this page is for someone who wants X." If the two sentences are the same sentence with different words, consolidate. If you can't write the second one at all, definitely consolidate.

There's one more option nobody likes but that's sometimes right. If neither page is any good and the query matters, consolidate both into one new page at the stronger URL and treat it as a rewrite rather than a merge. That's more work than either fix above, but it beats stitching two weak drafts together and hoping.

Rewrite both pages, not just one

This is the step teams skip, and skipping it is why cannibalization keeps coming back on the same two URLs.

If you consolidated, the work has four parts and all four matter:

1. Move the useful sections from the second page into the canonical one. Not everything. The parts that answer the question better than what's already there.
2. Redirect the old URL to the new one with a 301, not a 302. A temporary redirect tells crawlers the old page is coming back, so it keeps competing, and a page that returns a 404 throws away every link pointing at it.
3. Update internal links so they point at the canonical URL directly instead of bouncing through the redirect.
4. Rewrite the canonical page's opening so it clearly owns the whole question now, including the part the other page used to handle.

If you differentiated, the work is smaller but more delicate. Change the second page's title, its opening paragraph, and its headings so the promise is different. Cosmetic edits won't do it; if the first 200 words still make the same offer, the pages will keep competing no matter what you did to the middle. Then link from each page to the other with text that says what the other one is for. That signals the split to readers and to retrieval systems at the same time.

Either way, spend some of the effort on the canonical page itself. A page that just absorbed a competing URL is usually a bit of a mess, with two introductions and a middle section that repeats itself. The habits that make a page easy for an assistant to cite are the same ones that make it readable; answer the question early, use the words the search uses, and keep the extractable parts as tables or short lists. There's more on that in optimizing existing content for AI citation.

Re-score and watch the rank

Publish, then wait, then check. In that order, and the waiting is longer than it feels like it should be.

Nothing changes for anyone until both pages get recrawled. The canonical page has to be re-read with its new content, and the redirect has to be processed before the old URL stops competing. Three to six weeks is normal. Checking after four days will show you nothing and make somebody nervous.

Watch three things:

• Your page count on the target search. This is the direct measure of whether the split closed. Two of your URLs became one.
• The rank of the canonical page. Merges usually produce a dip for a week or two and then a recovery above where either page started. The dip is not a sign the fix failed.
• Citations in answers. This moves last, because answers get rebuilt from whatever the assistant retrieved most recently. Ranking improvement shows up first, citations follow.

Re-score the canonical page against the same query you started with so the before and after are comparable. If the score improved and the rank didn't, the page is now good enough to be retrieved and something else is holding it back, often a technical issue like a page that only renders in the browser or that blocks the crawler. An AI readiness audit rules those out in one pass.

Then do the next one. Splits come in small numbers, so a site with four of them is maybe six weeks of work total, and it's the cheapest visibility work available because you're not commissioning anything new. When you do go back to publishing, check drafts against their target query before they go out; that's how you stop creating the next split, and checking a draft before publishing walks through it.

Start with one query you should already own

Pick the single question your team is most annoyed about losing. Look at which of your pages come back for it. If two of them do, you've found work worth a week, and it's work that doesn't require a new brief, a new budget, or a writer.

The reason this stays hidden is that nothing about it looks like a problem. Both pages are decent, both get some traffic, and no report has a row that says "these two are canceling each other out." You only see it when you stop auditing pages and start auditing queries.

One question, one canonical page. Do that four times and you've fixed more AI visibility than a quarter of new posts would have.