Diagnosing keyword cannibalization across event brands that share a vertical

Two shows in the same group serve the packaging market from different angles, one on materials and one on logistics. Both have an editorial team. Both have published a piece on sustainable packaging trends. An agency deck arrives recommending that one be deleted, and the word on the slide is keyword cannibalisation across event brands.

Nobody in the meeting can say what the pages earned before or after, so the recommendation gets accepted or rejected on how confident the agency sounded. Both outcomes are guesses, and the data to settle it takes about an hour to pull.

What the word usually gets blamed for

The common version of the idea holds that two pages on the same term split your ranking authority and neither can win. Joshua Hardwick's Ahrefs guide, updated in January 2024, takes that apart and gives a narrower definition worth adopting: "Having multiple pages about the same thing can lead to unexpected or undesirable rankings, but it doesn't always mean that something's wrong or needs fixing."

The narrower test is whether the overlap costs the group organic traffic. Two pages both ranking on page one, between them earning more clicks than one page ever did, are a good outcome that a ranking tool will flag as a problem. One page dropping from position 4 to position 12 the month a sibling brand published on the same term is a different situation entirely.

An event portfolio produces the second case more often than most publishers, for a structural reason. Shows in adjacent verticals share a buyer, share a supply chain and share the vocabulary that buyer searches with, so the overlap is not accidental and it will not be designed away. It has to be measured.

The only test that settles it

Export query and page data from both brand properties. Filter each to the shared term. Sum the clicks and the impressions across both. Then do the same for a comparable period before the second page existed.

Three outcomes are possible and they need different responses. Summed clicks up: leave it alone, whatever the ranking report says. Summed clicks flat with impressions up: the two pages are trading positions and you are getting more visibility for the same traffic, which is worth watching and not worth surgery. Summed clicks down while impressions hold or rise: that is the case, and it is the only one that justifies the cost of a merge.

Choosing the two periods is the part that needs care. The natural cut is the publication date of the second page, with twelve months either side, which keeps seasonality out of the comparison because both windows contain one show cycle. Where both pages predate whatever history your query export covers, the before period does not exist and this test cannot be run at all. The fallback then is a page level comparison inside the window you do have: if one page holds a stable share of the term's clicks and the other takes almost none, the second page is contributing nothing and you can act on that without a counterfactual.

Run a control alongside it. Pick a term that only one of the two brands has ever targeted, pull the same two periods, and look at the change. Search demand moves for reasons that have nothing to do with your publishing, and a 40 per cent decline on your contested term looks damning until you notice the control fell 38 per cent over the same window because the category had a quiet year.

Working one shared term across two properties

Take the packaging pair and one term.

In the twelve months before the logistics brand published anything on it, the materials brand's page took 4,100 impressions and 178 clicks, at an average position of 6.8. Clicks over impressions gives a click through rate of 4.3 per cent.

In the twelve months after, the materials page took 3,100 impressions and 74 clicks at position 11.4, a rate of 2.4 per cent. The logistics page took 1,900 impressions and 21 clicks at position 17.2, a rate of 1.1 per cent. Summed, that is 5,000 impressions and 95 clicks, a combined rate of 1.9 per cent.

Impressions rose from 4,100 to 5,000, a gain of 22 per cent. Clicks fell from 178 to 95, a loss of 83 clicks or 47 per cent. More people saw the group and fewer came.

Now the control. A materials term the logistics brand has never touched fell from 2,400 clicks to 2,210 over the same two periods, a decline of 8 per cent. General demand explains eight points of the 47, so roughly 39 points want another explanation and the contested page is the obvious candidate.

That is the shape of a diagnosis. Two periods, two properties, one control, and a number in clicks that a commercial director can weigh against the cost of the work.

Why does a site: search mislead you here?

The quickest diagnostic in circulation is the worst one.

A site search restricted to your domain returns pages that contain the term. It tells you nothing about whether any of them appear in a normal result set for that query, which is the only thing that matters, and it will happily return forty pages on a portfolio that has one page ranking.

There is a second distortion on top. Google groups results from the same host, so a normal search shows you one result from a domain and hides the rest, which makes it impossible to see how your own pages are competing. Hardwick's 2024 guide recommends appending a filter parameter to the search URL to switch that grouping off before drawing conclusions.

Neither fixes the underlying problem, which is that a search box is a sample of one, run from one location, on one day, by somebody whose account has a history. The query export is boring and reproducible, and reproducible is what you need when the recommendation is to delete a page somebody wrote.

There is one more measurement trap worth naming. Google's performance report documentation, current in 2026, defines average position as "The average position of the topmost result from your site", so where two pages on the same property compete, the metric only ever reflects the better of the two. Across two separate brand domains you get two honest figures, which is one of the few ways the cross brand case is easier than the within brand one.

How do you tell a real case from two shows with different readers?

Same term, different intent, is the most common false positive in an event portfolio.

An exhibitor list page and an editorial explainer can both rank for a category term while answering unrelated questions. The buyer typing the term wants an explanation. The supplier typing it wants to know who is exhibiting. If both pages hold their clicks, the term is doing two jobs and the correct action is nothing at all.

A second false positive comes from your own show names. Where two shows in a group carry similar branded terms, the query export will show both properties appearing on each other's brand searches, and that overlap is almost never worth acting on. Somebody searching one show name and landing on the sibling has either mistyped or is looking for the group, and consolidating those pages would remove one of your two shows from search on its own name. Filter branded terms out of the analysis before you start, and count how many of the flagged terms disappear when you do. In portfolios where the shows were named by the same marketing team, it is often most of them.

Read the two pages before you read the data, and write down in one sentence what question each answers. Where the sentences differ, be sceptical of the diagnosis even if the numbers look bad, and check whether something else changed in the period, such as a redesign, a redirect, or a competitor publishing something better.

Where the sentences are the same, you have a real case, and the fix is choosing a winner and folding the loser into it rather than deleting anything. Where the pages are close copies of one another instead of independent takes, the instrument is different again and the canonical routing decision comes first.

Where this stops

The test proves association and stops well short of cause. Something else always changed in the twelve months between your two periods, and a control term reduces that risk without removing it. Treat the result as evidence strong enough to justify a merge, and weak enough that you should record what you did and check the term again two quarters later.

Query data also has a floor. Terms with a handful of monthly impressions produce click counts too small to compare, and a fall from 6 clicks to 3 is noise wearing a percentage. Set a minimum, something like 500 impressions in the earlier period, and put everything below it on a watch list instead of a work list.

The harder limit is organisational. In most portfolios the two pages belong to two brands with two P and Ls, and the merge takes traffic from one and gives it to the other. No amount of query data resolves that, because it is a question about who owns the audience. Settle the ownership rule before you run the diagnosis, or the diagnosis becomes ammunition in an argument nobody scheduled.

Pick the single term you already suspect, export twelve months of query data from both brand properties plus one control term, and add up the clicks. If the summed figure held, close the ticket and tell the agency why. If it fell, you have the beginning of a case, and the next question is which page keeps the traffic and how the content operation stops the pair recurring, which usually comes down to a commissioning check and an anchor text policy between the brands.