HOW GOOGLE BUILDS IN-MARKET AUDIENCES: THE SIGNALS THE SYSTEM USES

Most advertisers treat in-market audiences like a checkbox: tick the segment that matches the product, run the campaign, hope for the best. But the interesting question isn't which segment to pick, it's how Google decides who belongs in it in the first place. If you understand which signals the system uses to build an in-market audience, you stop treating it as a black box and start predicting how it will behave on your account, where reach will be thin, where intent will be strong, and why the same segment performs differently across two campaigns.

This article is built around that mechanism, the actual purchase-intent signals behind the scenes, cross-checked against Google's own documentation and against what I've watched these segments do on real accounts. The goal is not to recite definitions. It's to give you a working model of how the audience is assembled, so the inference you draw about its behaviour is grounded in how it's actually constructed.

What an in-market audience really represents

An in-market audience is a group of users that Google believes are actively researching and considering buying a product or service in a specific category, right now. The emphasis is on now: this is a target audience defined by what people are doing this week, not by who they are in general. Google's own reference describes these as users reached based on their recent purchase intent, and that single word, recent, is what makes the segment behave the way it does. It's documented directly in Google's overview of About audience segments.

Intent, not interest

The cleanest way to internalise in-market audiences is to contrast them with affinity audiences. Affinity describes who a person is, a long-term pattern of interest, a cooking enthusiast, a frequent traveler. In-market describes what a person is doing, actively shopping for a category in a short window. In digital marketing terms, these are audiences based on present behaviour rather than enduring identity. The same individual can be a lifelong "Auto Enthusiast" (affinity) for a decade while only entering the "in-market for new vehicles" segment for the few weeks they are genuinely comparing models. That distinction is the entire point: in-market is built on transient behaviour, affinity on durable behaviour.

Why membership is short-lived

Because an in-market segment corresponds to a single purchasing decision, membership naturally expires. Once the signals that put someone in the segment fade, or a purchase is detected, the system removes them. From operational experience, this is why in-market reach is smaller and more volatile than affinity reach: you are looking at a rolling, recency-bound population, not a stable one. The inference matters operationally, you can't expect an in-market segment to deliver flat, predictable volume the way a broad demographic does.

The signals the system actually reads

Here is the core of it. Google doesn't place a user into an in-market audience on the strength of a single click. It correlates several purchase-intent signals into a pattern that reads as active research. From the documented behaviour and from what's observable in account data, the system leans on a recurring set of inputs, and it's this stack that lets you target people who are genuinely close to buying rather than merely curious.

The behavioural inputs

The signals that consistently surface are: the search queries a user runs, the clicks on ads and the conversion rates that follow those clicks, the specific pages and content consumed during a session, the frequency of sessions within a category, and app usage. Comparing competing products within the same category, or researching offers from rival brands, is a particularly strong indicator, it's the digital equivalent of standing in a shop weighing two options. Google's reference notes that the underlying data, such as page-visit history and past searches, can also feed the bidding and accuracy of audience campaigns, which tells you these aren't passive labels, they're live inputs to the model.

Why it's a correlation, not a checkbox

The reason this matters: any one of those signals, taken alone, is noise. Someone who searches "best running shoes" once might be a runner, a gift-buyer, or a curious reader. The system only treats it as in-market intent when the signals stack, repeated sessions, product comparisons, content depth, and a recency that says this is happening now. From operational experience, this is exactly why in-market segments capture searches that keyword targeting misses: the user's history qualifies them even when their immediate query is too generic to reveal intent on its own.

The inference for your account

If signals stack into membership, then the segments most useful to you are the ones where your category produces clear, repeated research behaviour, considered purchases, comparison-heavy categories, higher-ticket goods. Conversely, impulse or commodity categories generate weaker intent signals, so the in-market segment will be thinner and less differentiated. You can predict the usefulness of a segment before you ever run it, simply by asking how much deliberate research your buyers actually do.

Where the activity behind the signals is collected

A detail most guides skip, and it changes how you read performance, is where the signals come from. Google is explicit that the source of the activity is tied to where the ad will eventually be shown.

Google products vs third-party properties

According to Google, users are allocated into audience categories on Google's own products based primarily on their activity on those products, while allocation for third-party websites and apps is based primarily on activity across those third-party properties. In Display & Video 360 the same split is spelled out: in-market lists are populated from web activity, and you target users as they browse pages across the web. This is documented in Google's page on In-market audience targeting.

The inference: signal context follows placement

The practical takeaway is that the same named in-market segment is not built from identical data everywhere. On Search and YouTube the membership leans on Google-product activity, queries, watch behaviour, Google-surface engagement; on the open web and the Display Network it leans on third-party browsing. From operational experience, this is part of why a segment that performs beautifully in search campaigns can underwhelm on Display: the population is qualified by different signals, so the depth of intent differs even though the label is the same. Reading the segment as "one audience" hides that nuance; reading it as "a label populated by context-specific signals" explains the gap.

How to actually use in-market audiences in 2026

Knowing how the audience is built only pays off if it changes what you do. In 2026 you can apply in-market segments in three modes, and the right one depends on how much you trust the signals for your specific case. Treat each mode as a distinct targeting option rather than assuming one audience type behaves the same way everywhere.

Targeting, observation, and audience signals

In targeting mode your ads show only to the selected in-market audience, useful when you want to restrict reach to high-intent users. In observation mode the ads run normally while Google reports how the segment performs, which is the safer starting point because it lets you read the signals without throttling reach. And in newer surfaces the segment becomes an audience signal rather than hard targeting, you hand it to the system as a hint about who is likely to convert and let automation expand from there. Which of these is available depends on the campaign type you're running. Google describes this signal-based, segment-driven construction in its Demand Gen audiences overview.

What experience says about each mode

From operational experience, observation first is almost always the correct move on Search. Because in-market reach is recency-bound and thin, hard-targeting it on day one often starves the campaign of volume before you've learned anything. Running it as an observation layer lets the underlying signals reveal which segments correlate with your conversions, and only then do you tighten to targeting or fold the segment in as a signal for automated bidding.

The adjacent-segment trick

When no in-market segment maps cleanly to your product, the signal model tells you what to do: pick the adjacent segment whose research behaviour overlaps with your buyer's. If there's no segment for your exact niche but a broader category attracts the same comparison and research signals, target that and let your creative pre-qualify. You're effectively borrowing the intent signals of a neighbouring category, which only works because you understand that the audience is built from behaviour, not from a product taxonomy.

The bottom line

The value of knowing how Google builds in-market audiences is that it turns a guessing game into a prediction. Membership is a stack of recent purchase-intent signals, search behaviour, ad clicks and their conversion rates, content depth, session frequency, comparison activity, collected from Google products or third-party properties depending on where the ad runs. Read the segment that way and its quirks stop being mysterious: the thin reach, the Search-versus-Display gap, the short membership window all follow logically from the signals the system uses. Pick segments where your buyers genuinely research, start in observation, and treat the audience as what it is, a live read of intent, not a static list.