Keyword cannibalisation: how to find it and when to fix it

Keyword cannibalisation is when two or more pages on your site serve the same search intent for the same query, so Google keeps switching between them and neither ranks as well as one stronger page would. The fix is usually simple. The hard part is knowing whether you have the problem at all, because most overlap between your pages is harmless and some of it works in your favour.

Many SEO tools flag every query where more than one of your URLs appears. Act on every warning and you may merge pages that were earning traffic side by side. Google's John Mueller made this point in September 2025, in an answer reported by Search Engine Journal: three pages from one site appearing in the same results did not seem problematic to him "just because it's 'more than 1'".

Who this is for: small-business owners and in-house marketers who have seen two of their pages taking turns for a query, or whose audit tool reports cannibalisation. What you will be able to do: test in Search Console whether the overlap costs you anything, using thresholds you set, then choose the right fix and measure it. Our post on choosing between a service page and a blog post covers the basic definition and a quick merge; this guide goes further on diagnosis and decisions.

What keyword cannibalisation is, and what it isn't

Keyword cannibalisation is a conflict between your own pages: they answer the same question for the same searcher, so Google has to choose, and its choice keeps changing. Keyword overlap is different. It is two pages that share some queries but serve different needs, such as a guide and a service page, and it usually needs no fix.

Ahrefs draws the same line in its keyword diversification study, first published in 2024. It describes cannibalisation as a search engine constantly exchanging rankings between pages, or pages ranking together that are similar enough to combine. Diversification is two or more pages ranking at once where consolidating them would likely bring no benefit. When the author reviewed a sample of 80 keywords for which Ahrefs' own site had multiple rankings, only one needed action.

What Google has said about multiple pages ranking for one query: our comparison

Google has no documentation page about "keyword cannibalisation" as such, so its position has to be pieced together. Method: in October 2026 we read the sources below and recorded what each says about several URLs from one site appearing for one query, and what that implies for a site owner. Documentation changes, so check the links yourself.

Read together, the sources agree on three points. Overlap is not penalised. Near-duplicates are filtered rather than rewarded. And the fix should match the actual fault, whether that is duplication, a weak page or poor internal linking.

When overlap costs you, and when it helps

Overlap costs you when it splits clicks, links and Google's attention between pages that do the same job, or when the wrong page ranks for a valuable query. It helps when two different pages give the searcher two useful options and take up more of the results page.

Signs it is costing you

• Unstable rankings. The URL that ranks for a query changes every few days, and neither version holds a position.
• A ceiling. Both pages sit somewhere between positions 6 and 15 for months, while competitors with one focused page rank above them.
• The wrong page ranks. An old blog post ranks for a buying query such as "back pain physio near me", while the service page that converts sits lower.
• Split links. Other sites link to both versions, so neither page gets the full benefit of those links.
• A drop after publishing. Clicks to an established page fell soon after you published a newer page on the same subject.

Signs it is helping, or harmless

• Both pages hold stable positions, often one high and one lower, and the leading URL rarely changes.
• The pages are different types that match different needs, such as a guide and a tool page that rank for the same keyword without swapping places. Ahrefs' study calls this diversification.
• The second page only picks up a sliver of impressions for queries it was never written for.

A short list of real faults makes a useful checklist before you merge anything. If a page is unfocused, wanders off-topic, is thin or is poorly linked internally, fixing that may solve the problem without touching the other page. Our on-page SEO checklist covers focus, headings and internal links page by page.

Find candidates in Search Console

Search Console is the best free source for this job because it shows which of your URLs Google actually displayed for each query, rather than a third-party estimate. On a small site, use the Queries and Pages tabs; on a larger one, export query-and-page pairs through the API.

First, get the date range right

Google's data anomalies page records that a logging error affected Search Console impressions from 13 May 2025 until 27 April 2026. Impressions, CTR and average position were affected; clicks were not. For this exercise, use a date range that starts after 27 April 2026, and compare any older period on clicks only (checked October 2026).

Know how Search Console counts pages

Search Console counts the same searches in two ways. Google's help page on Performance report data explains that table data grouped by query is aggregated by property, while data grouped by page is aggregated by page. Position is the topmost position held: at property level, that is the best position of any of your URLs; at page level, the best position of that one URL.

Two consequences matter here. When you filter by one query and open the Pages tab, the page impressions can add up to more than the query's total, because both URLs can appear on the same results page. And a query's average position on the Queries tab can look better than either page's position on the Pages tab, because only your best-placed URL counts at property level.

Method A: the interface, query by query

1. Open Performance, then Search results. Show clicks, impressions and average position for a 3-month range starting after 27 April 2026.
2. On the Queries tab, add a "Queries not containing" filter for your brand name, sort by impressions and note your top 30 to 50 non-brand queries.
3. Click the first query to filter the report by it, then open the Pages tab. Record every URL with impressions, along with its clicks, impressions and position.
4. Repeat for each query. Any query where a second URL has a meaningful share of impressions goes on your candidate list.
5. Work the other way too: filter by a suspect page, open the Queries tab and compare that list with the queries of the page you think it competes with.

Method B: export query-and-page pairs

For larger sites, pull both dimensions into one table. The Search Analytics API accepts query and page together as dimensions and returns up to 25,000 rows per request, with startRow for paging through more. Google notes that it returns the top rows rather than guaranteeing every row. A request body looks like this:

{"startDate": "2026-07-01", "endDate": "2026-09-28", "dimensions": ["query", "page"], "rowLimit": 25000, "startRow": 0}

A developer or any reporting tool that connects to the Search Console API can run this. Paste the rows into a spreadsheet with query in column A, page in B, clicks in C, impressions in D and position in E, then add two formula columns:

• URLs per query: =COUNTIFS(A:A, A2, D:D, ">="&$H$1) counts how many of your URLs reached your impression threshold, typed into cell H1, for that query.
• Share of impressions: =D2/SUMIF(A:A, A2, D:D) gives each URL's share of the query's page-level impressions.

Filter to queries with two or more URLs where the second URL's share passes your threshold. That is your candidate list, not yet your problem list.

Set your own thresholds

No universal cut-off separates noise from a real conflict: a query with 40 impressions a month behaves differently from one with 40,000. Set thresholds before you look at individual pages, so every query is judged the same way. The starting values below are our suggestions for a small business site.

Confirm it is a problem: four tests

A candidate is real cannibalisation when the pages share an intent and the overlap is measurably costing you. Run four tests on each candidate pair: intent, flipping, cost and content overlap. Test 1 must pass, plus at least one of tests 2 and 3, before you change anything.

Test 1: do the pages serve the same intent?

Ask whether someone who landed on page B would be fully satisfied by page A, and the other way round. Then search the query in a private window and note what kind of page ranks: guides, service pages, tools or local results. If both your URLs are that type and answer the same question, they share an intent. A guide and a service page usually do not.

Test 2: does the leading URL keep swapping?

Flipping is the clearest sign that Google cannot decide between your pages. Ahrefs' study describes exactly this with "seo case studies" and "seo case study", where a newer post kept displacing an older one until Ahrefs fixed it with a pillar page, which then ranked in the top 3 most of the time.

To measure it, you need daily data per URL. In the interface, filter by the query and one URL, open the Dates tab and export; repeat for the second URL and line the two exports up by date. With the API, add date to the dimensions and filter to the one query. For each day, the leading URL is the one with the better average position. Count how often the leader changes over 28 days, skipping days with very few impressions (for example, under 10), because they are noisy.

Check device and country before you conclude. Mobile and desktop results can differ, and a business serving two cities may see a different URL lead in each. Swaps that disappear when you filter by device or country are not flipping.

Test 3: is it costing clicks or positions?

Look for a cost you can actually see. The usual signs are neither URL reaching your position threshold, the pair stuck in the same band for months, or clicks to the older page falling after the newer one was published. Use clicks for any comparison that reaches back before 27 April 2026.

Test 4: how much content do the pages share?

List the H2 and H3 headings of both pages side by side and mark the sections that answer the same question. If more than about half are shared, the pages are close to duplicates and a merge is usually right. If less, each page may have a distinct core worth keeping, which points to differentiating them instead. Test 4 does not decide whether there is a problem; it decides which fix suits it.

Rule out look-alikes

Some problems look like cannibalisation but are not. A sitewide drop in the same weeks as a confirmed Google update is a ranking change to diagnose with our core update recovery guide. A URL that vanished entirely may have an indexing or canonical fault, which a technical SEO audit will catch. And a dip that repeats at the same time every year is seasonal.

Fix options: merge, differentiate, re-link, canonicalise or leave

Once a pair passes the tests, the fix depends on two questions: does each page deserve to exist, and do users need both URLs? The table turns those answers into an action.

Decision table

How to differentiate two pages

1. Write one sentence for each page: who it is for and what question it answers. If you cannot make the two sentences different, merge instead.
2. Rewrite each title and H1 around its own question. Check the new titles in our free Google SERP Preview to make sure they are not cut off on mobile, and avoid the patterns in our post on common title tag mistakes.
3. Cut or move sections that belong on the other page, and link to that page at the point where you removed them.
4. Run each page with its own target query through the On-Page SEO Checker. Look for the target term in the title, H1 and opening paragraph, and a self-referencing canonical. If both pages score well for the same term, they are still too alike.
5. Give each page different long-tail queries to own, and check in Search Console or Keyword Planner that those queries have real demand before you rework a page around them.

When a canonical tag is the right tool

Use rel="canonical" only when both URLs must stay live for users and their content is the same or nearly so. Google's documentation on specifying a canonical ranks redirects as the strongest signal, rel="canonical" next and sitemap inclusion as weak, and says Google may still choose a different canonical. Pointing one distinct article's canonical at another asks Google to treat different pages as duplicates, so do not rely on it.

What not to do

• Do not noindex one page as the fix. Google says it does not recommend noindex to prevent selection of a canonical page within a single site, because noindex blocks the page from Search completely.
• Do not hide a page with robots.txt. Google has to crawl a page to see its redirect or canonical tag.
• Do not redirect to the home page. Google's site move guidance warns that redirecting many old URLs to one irrelevant destination, such as the home page, might be treated as a soft 404. Redirect to the page that now covers the topic.
• Do not mix signals. Google advises against specifying different canonical URLs for the same page through different methods, such as one URL in your sitemap and another in rel="canonical".

Worked example: a physiotherapy clinic's overlapping pages

A worked example shows how the thresholds and tests combine into a decision. Example (hypothetical): a physiotherapy clinic in Bengaluru has three pages that appear for back-pain searches. Page A is a 2023 blog post at /blog/lower-back-pain-exercises, page B a 2026 post at /blog/best-exercises-for-back-pain, and page C the service page at /back-pain-physiotherapy. All figures are invented to show the method, using 90 days of data from July to September 2026 and the suggested thresholds above.

Reading the example

1. Query 1 passes every threshold. It has 7,400 page-level impressions, B's share is 3,300 ÷ 7,400 = 45%, the leader changed 5 times in 28 days and neither page is better than position 3. Both pages are exercise guides, so test 1 passes, and 6 of their 8 H2 sections answer the same questions, so test 4 points to a merge.
2. Query 2 is diversification. A's 27% share passes the share threshold, but the service page leads every day at position 3.1 and the intents differ. The only action is a contextual link from A to C for readers who need treatment rather than exercises.
3. Query 3 changes the merge plan. B ranks at 2.4 for a question A barely touches. Choosing the survivor on query 1 alone would put that at risk.
4. Choose the survivor on whole-page value. Suppose that across all queries A earned 1,240 clicks in the 90 days and has links from 14 other sites, while B earned 610 clicks and has links from 3. A survives. B's walking section moves across intact, B's URL is 301 redirected to A, and query 3 goes on the tracking list.

Merge pages without losing traffic

A merge combines the useful content of two pages on one URL and permanently redirects the other. Done carefully, the surviving page keeps the links, bookmarks and rankings of both; done carelessly, it loses the very passages that were ranking.

Step by step

1. Pick the survivor. Choose the URL with more clicks across all its queries, more links from other sites and the better fit with the intent. If the weaker page is better written, keep the stronger URL and move the writing.
2. Inventory both pages. List each section, the queries it ranks for and the links pointing at it. Run both pages through the free AI Answer Snippet Extractor, which scores every paragraph out of 100 as a quotable snippet, and carry the strongest passages across rather than rewriting them from scratch.
3. Rewrite the survivor. Combine the sections into one structure that answers the intent better than either page did. Remove repetition, update facts and keep the survivor's URL unchanged.
4. Redirect the old URL. Use a permanent, server-side 301 or 308 redirect. Google's redirects documentation says Google Search uses permanent redirects as a signal that the target should be canonical.
5. Update internal links and the sitemap. Point every menu, footer and in-text link at the survivor and remove the old URL from your XML sitemap. Avoid redirect chains: Googlebot can follow up to 10 hops, but Google recommends redirecting straight to the final destination.
6. Check the redirect. Enter the old URL in our Redirect Checker, which tests the http, https, www and non-www versions. You want one 301 hop from the old URL to the survivor, no chains or loops on the variants, and a 200 status at the end.
7. Ask Google to recrawl. Inspect the survivor with Search Console's URL Inspection tool and request indexing.

Keep the redirect in place

Google's site move guidance says to keep redirects for as long as possible, generally at least a year, so signals can transfer, and to consider keeping them indefinitely for users. For a merge on your own site, there is rarely a reason to remove one, and Google's redirects documentation says permanent redirects are for changes you are sure won't be reverted.

What can go wrong

• A moved section loses its query. If the passage that ranked was rewritten or buried, the survivor may not inherit that query. Keep winning passages close to their original wording and heading.
• The survivor becomes unfocused. Pasting two pages together produces a long, wandering page. Edit until each section earns its place.

Cannibalisation and AI answers

AI answers make duplication harder to justify. Google's page on AI features and your website says AI Overviews and AI Mode may use a query fan-out technique, issuing multiple related searches across subtopics, and that this lets them show a wider and more diverse set of helpful links than classic search.

Our reading of that design: two pages saying the same thing are substitutes competing for one place in that diverse set, while two pages answering different sub-questions can each match a different fan-out search. We found no published study of how often AI Overviews, AI Mode or ChatGPT search cite two pages from one site in the same answer, so treat this as reasoning, not measured fact.

What Search Console can and cannot tell you

• AI Overview positions are shared. Google's help page on impressions, position and clicks says an AI Overview occupies a single position and every link inside it is assigned that position. Two of your URLs in one AI Overview look like one appearance at property level.
• AI Mode follow-ups count as new queries. The same page says each follow-up question in AI Mode is treated as a new query, so a page can appear for a follow-up without appearing for the first question.
• The generative AI report works by page. Google's help page for the Generative AI performance report describes impressions for links shown in AI features, broken down by page, country, device and date, with pages grouped by the final URL after any redirects. It offers no query breakdown, so you cannot see AI cannibalisation for one query directly.

A quick check before you merge

1. If your property has the Generative AI performance report, filter it to each URL in the pair and compare impressions over the same dates.
2. If the page that is weaker in organic results earns more AI impressions, find out why before merging. It may have a cleaner, more direct passage, and that passage should move into the survivor intact.
3. After the merge, expect the old URL's AI impressions to appear under the survivor, and track them alongside its clicks.

Our guide to the Search Console generative AI report covers its other limits, and our guide to getting cited in AI search explains why clear, answer-first passages tend to be the ones quoted.

Prevent it with a keyword-to-page map

A keyword-to-page map is a list that assigns each search intent on your site to exactly one target URL. Checking it before anyone writes a new page prevents most cannibalisation, because the question "do we already have a page for this?" gets asked every time.

A template you can copy

The rows continue the hypothetical clinic above. Add a column for the page owner if several people publish.

Rules that keep the map useful

1. One intent, one target URL. Group close variants, such as "back pain exercises" and "exercises for lower back pain", into one cluster rather than one row each. Our keyword research guide shows how to group keywords by intent.
2. Check the map before every new page. If the cluster already has a target, improve that page instead of writing another.
3. Record supporting pages. Supporting pages target narrower questions and link to the target with descriptive anchor text, which also stops internal links pointing the wrong way.
4. Review it quarterly against Search Console. Rerun method A or B and add any new pairs that pass your thresholds.
5. Be careful with location pages. A page per city is legitimate when each has genuinely local content, such as the team, address and services available there. Swapping only the town name creates near-duplicates and, at scale, the doorway pattern Google's spam policies describe.

Track the fix

Judge a fix on clicks for the queries you targeted, compared with the same length of time before the change. Give Google time to recrawl; ranking changes after a merge are rarely instant.

What to record

A simple timeline

1. Note the date of the change in your SEO log or report annotations.
2. After 2 weeks, confirm the redirect still works and the old URL is fading from the Pages tab.
3. After 4 weeks, compare the 28 days after the change with the 28 days before it. Avoid comparing across a confirmed Google update if you can.
4. After 3 months, compare the full quarter. If the survivor sits below the old combined clicks, check whether a moved section lost a query it used to win, and restore what is missing.

Put these numbers in your regular reporting so the fix is judged on outcomes, not on a tool's warning count. Our post on what to look for in a monthly SEO report shows where they fit.

Diagnosing and fixing overlapping pages is part of our on-page SEO service, alongside keyword mapping and internal linking. If you would like a second opinion on a pair of pages before you merge them, get in touch.