The Geo Fence You Drew Around a Location Does Not Contain the Audience Your Report Says It Reached

Location-based targeting carries an appealing simplicity: draw a boundary on a map, serve ads to people who enter it, and report on how many you reached. Media plans built on this premise are common, and the confidence they project is understandable. The problem is that the boundary on your planning screen and the signal your DSP used to bid on an impression are two different things, separated by several layers of inference that most post-campaign reports do not surface.

How Device Location Signals Actually Work

When a mobile device emits a location signal, that signal travels through an app's SDK, gets packaged by a data aggregator, and eventually reaches the bid stream as a latitude-longitude coordinate attached to a bid request. Each handoff in that chain introduces potential for signal degradation. GPS precision varies by device hardware, operating system permission settings, and whether the app requested foreground or background location. Horizontal accuracy can range from a few meters to several hundred meters depending on those conditions.

By the time a coordinate arrives in the bid stream, it may represent where the device was when the app last polled for location, not where the device is at bid time. That lag is often seconds, but it can stretch to minutes or longer in low-connectivity environments. For a boundary drawn around a conference center, a retail location, or a trade show floor, that temporal gap and spatial uncertainty translate into impressions served to people who were near the target area, not confirmed to have been inside it.

What Your Targeting Platform Accepts as a Match

DSPs and location data vendors set their own tolerance thresholds for what counts as a match to your drawn boundary. A common approach is point-in-polygon matching: if the coordinate falls within the polygon, the bid qualifies. But because GPS accuracy is imperfect, vendors often apply a buffer zone around the polygon before testing containment. That buffer is a practical engineering decision designed to reduce false negatives, and it is reasonable on its own terms. What it means for your campaign is that the effective targeting zone is larger than the boundary you drew, and you rarely see how much larger.

Some platforms also use panel-based or modeled location data to fill gaps where live GPS signals are sparse. If a device does not emit a recent coordinate, the platform may infer location from historical visit patterns, Wi-Fi network associations, or cell tower triangulation. These methods produce usable signals for broad geographic targeting, but they are not equivalent to a confirmed, timestamped GPS fix inside your fence.

Why Your Audience Report Looks More Precise Than the Delivery Was

Post-campaign location reports typically show impression counts attributed to the fence, a reach number, and sometimes a frequency distribution. What they do not show is the distribution of signal quality underlying those impressions. Impressions served on a modeled location signal, a stale GPS coordinate, or a coordinate that fell just outside the buffer look identical in reporting to impressions served on a fresh, high-accuracy fix from a device confirmed inside your fence at bid time.

This is not a disclosure failure unique to any one platform. It reflects the structure of how location data moves through the ecosystem. The report you receive summarizes what the platform accepted as qualifying impressions. It does not reconstruct the signal quality of each individual bid.

Practical Calibration for Location-Based Buys

None of this means location targeting is unreliable as a category. It means that the precision you assume when drawing a boundary should be calibrated against how location signals actually behave in the bid stream. A few practical orientations can help.

First, treat your drawn boundary as a targeting intent, not a confirmed reach perimeter. The impressions your campaign delivers will represent people who were, with varying degrees of certainty, somewhere in the vicinity of your target location during the campaign window. That is a useful audience for many objectives. It is not a confirmed visitor list.

Second, ask your location data vendor how it classifies signal quality in its bid-stream data. Some vendors segment their inventory by GPS accuracy tier or signal freshness and can tell you what share of your impressions came from high-confidence signals versus modeled or inferred ones. If that breakdown is not available, treat reported reach numbers as an upper bound rather than a confirmed count.

Third, consider whether your campaign objective requires high location precision or whether broad proximity is sufficient. Targeting people who work in a specific office building requires tighter signal quality than targeting people who shop in a retail corridor. Adjusting your expectations and your measurement approach to match the actual precision your signals support will produce more useful post-campaign analysis.

Fourth, if you are using location audiences for retargeting or for building a segment of confirmed visitors, ask specifically how your vendor defines a visit for segment inclusion. A single matching bid request is a different standard than multiple confirmed signals across a defined dwell time. The latter is more likely to exclude people who simply drove past the location boundary, and some vendors offer dwell-time filters that you can apply at the segment-build stage.

The Measurement Implication

Location-based campaigns are often used in attribution workflows: a person seen near a location is later observed converting, and that conversion is credited to the campaign. The identity resolution step in that attribution join carries its own uncertainty, as covered elsewhere in this publication. But upstream of that join, the location signal that qualified the impression for your fence also needs to be understood as a probabilistic signal, not a confirmed observation.

When you evaluate the performance of a location-based campaign, the most useful framing is to ask what confidence level your location signals supported, and whether your measurement setup was designed to account for that. A campaign built on high-quality, dwell-filtered GPS signals and a campaign built on broad modeled proximity data can report similar reach numbers while describing very different levels of certainty about who was actually reached.

Understanding that gap does not require rejecting location targeting. It requires building a clear internal picture of what your signals actually represent before you make planning decisions or draw conclusions from the results.