How Google Changes Your Searches (A Study of 10,000 Queries)

Edited by Emilie Martin

A recent Wired article (since removed) made the controversial claim that Google was replacing broad product category searches with results for specific brands. Specifically, the article cited a search for “kids clothing” being replaced with results for the brand “Nikolai kidswear.” Google quickly rejected any implication that they were commercializing organic search results.

This conclusion seems to have been reached from the following slide (via @adamkovac):

The slide appears to show the results of Google Ads matches, and while Google’s broad matching of ads to searches can be unpredictable and controversial, this does not reflect Google modifying or replacing organic search queries or results.

That said, those of us who closely watch Google results know that Google does occasionally correct and reinterpret searches. How often does this happen, and how far do they go? To answer this, we dug into our daily MozCast research set of 10,000 searches to find situations where Google returned something different than what searchers typed.

I’ll call this category “transparent” because these are cases where Google is clearly telling searchers what they’re doing. Across our 10K keyword set (which deliberately includes misspellings), about 5% of searches resulted in transparent corrections. You’re probably familiar with one of the older forms of this, the “Did you mean” correction:

Google correctly assumes that I meant the brand, which includes an apostrophe. This is so mundane that it barely even qualifies as spell-checking. More common these days (at least in our data set) is the “Showing results for” correction, like this one:

Here, Google is telling us that the common usage is “Verizon Wireless” as a two-word phrase. What’s interesting about this variant is that Google automatically takes you to the corrected results. Still, this is hardly a controversial change, and that’s true for the vast majority of corrections in our data set. Here are a few examples:

All of this is not only harmless but arguably useful to searchers. Let’s move into slightly more controversial territory. What about when Google doesn’t tell us what they’re doing?

To more quickly find modifications, we dug into how often the titles of organic results matched the query keywords (we were forgiving about punctuation, plurals, and word order). Ignoring transparent corrections like “Showing results for,” there were 27 queries in our data set (0.27%) where no titles technically matched the query. Here’s an example:

While Google isn’t alerting us to the change, the modification here is very basic and essentially the same as the previous “verizonwireless” example. Here are a few examples of relatively minor modifications with no alert to the searcher:

These get a bit more interesting as you go down the list. Note that “Target Search Query” here is an assumption based on the results shown and just demonstrates that Google understands close synonyms and equivalencies (including similar entities).

In the “tsa regulations” example, Google recognizes the relevance of the TSA’s website:

Of course, Google can equate “TSA” with “Transportation Security Administration.” Still, in this case, the top 10 results are all from the TSA’s own website and represent various official rules and regulations. The agency/entity has been given preference over the text itself.

The “usairways” case is also interesting because two things are in play. One, Google recognizes that we almost certainly meant “us airways”, but they also understand that US Airways merged with American Airlines:

Interestingly, due to US Airways complicated ownership history, United Airlines also made an appearance in the first few Google results. Here again, Google seems to have some knowledge of the historical and current relationships between brands/entities.

So, what happens when Google gets really wild with modifications? Honestly, the answer isn’t exactly shocking, but let’s take a look at some more interesting examples.

Digging deeper, we found some more compelling cases, ranging from complex synonyms to situations where the results look (on the surface) entirely different from the original query. Take, for example, these modifications for “quads”:

“Quads”, “quad muscles” and “quadriceps” are clearly synonyms, and unless someone intended an uncommon usage of “quads,” like a shortened form of quadruplets or quadrangles, this modification seems to be well-intentioned.

Here are a handful of examples of “medium” (an arbitrary label at best) modifications:

Like “quads,” understanding that “yorkie” refers to Yorkshire terriers is reasonable synonym handling. The next three get a bit more interesting.

For a searcher, returning auto loan calculators for “calculate car payment” doesn’t seem like a wild detour, but it’s undoubtedly disruptive for search marketers relying on exact-match phrases (that’s a topic for another blog post). The search for “quicken loans arena” is interesting because, again, it reveals Google’s potential understanding of entity relationships:

Google seems to recognize that the naming rights to Quicken Loans Arena changed hands in 2019. While swapping corporate sponsors might upset a few Cleveland Cavaliers fans, most searchers are arguably looking for the stadium itself and not a history lesson. This is interesting from a query-matching perspective — every single word has been replaced — but the results clearly match the intent of the original query.

Finally, we have what might be the most interesting example of a modification in our 10K query set. Look at the organic results on a search for “department stores”:

After a Places Pack (local/map results) of department stores, we see organic results for a local mall and an assortment of brand-name department stores. Instead of returning content about the keywords “department stores,” Google is returning specific department stores.

Ultimately, Google tries to disambiguate queries where searcher intent is unclear. Arguably, people searching for “department stores” aren’t looking for a definition of the phrase or a history of department stores — they’re searching for department stores in their local area.

Riffing on the “department stores” example, I came up with an example that superficially matches the controversial claim in the Wired article. Consider these results for “kids clothing stores”:

Again, we lead with a Places Pack (not shown) and the same local mall (to be fair, it’s a nice mall), followed by a number of specific brands selling kid’s clothing. This is about as close to moving from vague commercial intent to specific commercial intent as I’ve found.

Consider two things, though. First, these are product category pages that legitimately cover the topic of kid’s clothing. Second, Google isn’t aggressively pushing any particular products. There is a product listings block (which typically combines paid and free listings) at the end of the page, but the organic results, like the “department stores” example, arguably match the intent of the searcher and don’t particularly benefit Google’s bottom line.

I think it’s fair to be skeptical of Google’s financial incentives, but I genuinely believe that the modifications we uncovered serve most searchers. Google’s goal isn’t to exactly match the words in your query — that’s an artifact of search engines past, which couldn’t understand natural language or real-world entities and brands. Google’s goal is to answer the intent of a search and the question that search represents.