Stop the Bleed: A Data Scientist's Guide to Advanced Geo-Optimization in Google Ads
Geographic waste is one of the most expensive and least diagnosed inefficiencies in Google Ads. Most operators look at a green campaign with acceptable ROAS and conclude everything is working. They are not looking at where the money is actually going.
Google's default location targeting setting does not target locations. It targets people who are in locations or who have expressed interest in them. These are fundamentally different data types, and conflating them introduces what I call geographic bleed: systematic budget leakage to users who are physically in locations you cannot serve, feeding low-quality signals to your automated bidding models and progressively degrading your Target CPA and Target ROAS performance.
This is not just a waste problem. It is an algorithmic contamination problem. Every irrelevant click from a geographically mismatched user feeds your Smart Bidding model with a data point that says "this type of user, in this context, did not convert." The algorithm learns from that. It adjusts bids. It changes audience predictions. Over time, geographic signal pollution degrades the entire account's machine learning performance, not just the affected campaigns.
The fix starts with understanding how the default settings create the leak, and then building the architecture to close it.
For the signal quality foundation that makes geographic governance meaningful, The $1,127 Algorithmic Tax covers how any form of low-quality data contamination in your conversion signals degrades Smart Bidding accuracy at the account level.
The Presence or Interest Fallacy
Google Ads defaults every new campaign to a location targeting setting called "People in, regularly in, or who've shown interest in your targeted locations."
To a data scientist, this setting conflates two completely different data types into a single targeting parameter:
Physical presence data is derived from verified IP addresses and GPS signals. It represents users who are actually in the location you are targeting right now or regularly. This is the traffic you want.
Interest and intent data is derived from search history and behavioral signals. It represents users who have searched for content related to your target location, regardless of where they physically are. A user in Mumbai searching for "SaaS companies in New York" will be matched against your New York campaigns under this setting. They are a statistical outlier for actual conversion. But they count in your data, and they cost money.
This conflation creates three systemic risks:
Qualified lead dilution. Users matched on intent rather than presence convert at dramatically lower rates than physically present users. Their sessions inflate your impression and click counts without proportional conversion contribution, which degrades your optimization signals.
Algorithmic poisoning. Smart Bidding trains on every conversion event and every non-conversion event. Low-intent, geographically mismatched sessions teach the algorithm that certain query patterns and user profiles produce no conversions. The algorithm bids less aggressively on those profiles in future auctions, which may include legitimate in-territory users who share similar behavioral signals.
OFAC compliance exposure. Google must comply with U.S. Office of Foreign Assets Control sanctions. Territories including Crimea, Cuba, Iran, North Korea, Syria, and the Donetsk and Luhansk regions are under strict restrictions. Because these regions cannot be directly targeted, they often cannot be directly excluded through the standard UI. A user in a sanctioned territory who searches for content related to your service area can trigger your ads under the interest-based matching default. The standard UI provides limited visibility into this exposure.
The User Location Report: What the Standard Interface Hides
Google's simplified campaign interface removed the Location Type segment from the standard location view several updates ago. This means the default reporting shows you matched locations, not user locations. You see where Google thought your ad was relevant, not where the users who clicked it actually were.
To find the truth, you must move into the Report Editor.
Navigate to Insights and Reports, then Report Editor. Create a custom table and add the User Location dimension. This gives you five levels of geographic precision: Country or Territory, Region, Metro (DMA), City, and Most Specific Location at the zip code level.
Cross-reference this data against your Google Analytics bounce rate and conversion data by location. The patterns that indicate geographic bleed are consistent:
High spend combined with near-zero conversions from a specific location. This is the primary signal. A city or region appearing in your matched location data with significant spend but no conversions over a 30-day period is wasting budget on users who are not converting, feeding zero-value data to your bidding models.
High CTR combined with zero conversions. This is the behavioral signature of accidental clicks: users who clicked an ad that appeared while they were doing something entirely different (common in mobile gaming and app environments) or geographic confusion where the user's intent was topically related but geographically irrelevant.
Matched location in a Tier-1 market, user location overseas. This is the interest-based matching failure explicitly visible in the data. Your campaign matched on intent for a high-value market but served to a user physically elsewhere. The click cost you a premium CPM. The conversion probability was near zero.
Build a weekly User Location Report review into your audit cadence. Sort by spend descending and filter for locations with zero conversions over 30 days. These are your exclusion candidates.
Defensive Architecture: Building Geographic Guardrails
The structural fix for geographic bleed operates at two levels: the targeting setting and the exclusion architecture.
Switch to Presence Only targeting.
In every campaign, change the location targeting option from "People in, regularly in, or who've shown interest in your targeted locations" to "People in or regularly in your targeted locations."
This single setting change eliminates interest-based geographic matching entirely. Your ads will only serve to users the system has verified are physically in your target locations. For B2B campaigns, local services, and any business with geographic service constraints, this is not optional. It is the minimum viable configuration.
Build account-level geographic exclusions.
Account-level location exclusions apply as an absolute boundary across every campaign in your account. A campaign-level targeting setting cannot override an account-level exclusion. This makes account-level exclusions the correct mechanism for geographic constraints that should never be violated under any circumstances.
Exclude at account level: countries you do not service, regions where regulatory or licensing constraints prevent you from transacting, and international territories that consistently appear in your User Location Report with zero conversions.
For Performance Max and Demand Gen campaigns specifically, account-level geographic exclusions are the only reliable mechanism. These campaigns operate with significant automation and find the cheapest available inventory across all Google properties. Without hard geographic boundaries, they will systematically discover low-cost inventory in irrelevant markets and allocate spend there. Campaign-level suggestions and recommendations do not constrain PMax the way hard account-level exclusions do.
Geo-Targeting Precision: Choosing the Right Geographic Unit
The geometry of your target definition determines the resolution of your data and the quality of your IP-to-location matching.
| Location Type | Precision | Risk Factor | Best Use Case |
|---|---|---|---|
| State | Low | Low: Clear boundaries, limited nuance | National brands with state-level distribution |
| DMA | Moderate | High: DMAs often extend far beyond metro cores | TV-aligned campaigns in major markets |
| City | High | Moderate: Misses high-value users outside municipal lines | Businesses with strict municipal licensing requirements |
| Radius | Highest | Low: Best IP-to-location matching accuracy | Local services, high-intent retail, service area businesses |
The zip code patchwork approach that many operators use for local targeting is statistically erratic. Google's IP-to-zip matching fails frequently in high-density metro areas, producing both false positives (users outside your targets triggering your ads) and false negatives (users inside your targets not seeing your ads). The boundary complexity of overlapping zip codes in urban markets compounds this problem.
Radius targeting produces better results for most local service use cases. Setting a pin at your primary service location with a 5 to 10 mile radius is significantly easier for the system to match accurately than a patchwork of 30 zip codes. It also produces cleaner geographic data in your User Location Report, making it easier to identify and act on the waste patterns you find.
Proximity Bidding: The Conversion Multiplier
Once your geographic targeting is clean and your exclusions are in place, proximity bidding amplifies the performance of your highest-value geographic segments.
Users physically close to your location or service area convert at higher rates than users at the boundary of your target geography. Proximity bidding applies bid multipliers to reflect this conversion rate differential.
A concentric radius approach works well for most local service businesses: apply a +75% bid modifier for users within 3 miles, a +50% modifier for users within 5 miles, and no modifier or a slight negative modifier for users at the outer boundary of your serviceable area. This ensures you are bidding most aggressively for the users with the highest conversion probability while maintaining presence throughout your service area.
For B2B campaigns targeting specific office districts or commercial zones, proximity bidding to business district pins rather than city-wide targeting can significantly improve lead quality by concentrating spend on users physically in decision-making environments.
The Microsoft Ads pressure valve.
When Google CPCs in premium metros become unsustainable for B2B targeting, Microsoft Advertising offers a meaningful geographic budget extension. CPCs on Bing in most US metros run 20 to 35% lower than equivalent Google campaigns. More importantly, Microsoft Ads supports LinkedIn Profile targeting layers: job function, industry seniority, and company size can be layered onto geographic targeting to create a precision B2B audience that Google's geographic targeting alone cannot produce.
If you are being priced out of Google's competitive auctions in your target geography for B2B queries, Microsoft Ads with LinkedIn targeting is not a fallback. It is a precision instrument for the audience segment that geographic targeting alone cannot isolate.
Validating Geographic Changes: The Experiment Protocol
Never deploy a major geographic targeting change, particularly a shift from Presence or Interest to Presence Only, without a controlled experiment. The change can affect impression volume, CPC, and conversion rate in ways that are difficult to attribute without a clean control condition.
Create a geographic experiment directly in the Google Ads Experiments interface. Google removed the Draft-and-Experiment workflow in favor of direct experiment creation. Build a 50/50 cookie-based split: users are assigned to either the existing configuration or the new Presence Only configuration and remain in that group for the duration of the test.
Cookie-based assignment is mandatory for geographic experiments. Query-based splits allow the same user to encounter both configurations, which confounds the attribution and makes it impossible to measure the true effect of the targeting change.
Run the experiment for 4 to 6 weeks. Monitor weekly for cannibalization effects where the experiment version disproportionately pulls traffic from the base campaign. If cannibalization is occurring, the 50/50 split is working correctly but you need longer runtime to accumulate statistically significant conversion data at each split level. Do not read results before sufficient conversions have accumulated in both variants.
After the experiment concludes, examine three outputs: total conversion volume difference, cost per conversion difference, and conversion rate difference. A move to Presence Only typically reduces impression volume while improving conversion rate and cost per conversion. The net effect on total conversions depends on your specific geographic situation. The experiment tells you the actual numbers for your account rather than requiring you to infer from industry benchmarks.
The Signal Integrity Imperative
Geographic optimization is ultimately a signal integrity problem. Every dollar spent reaching a user who cannot convert is not just a wasted dollar. It is a corrupted data point that degrades your bidding model's ability to find users who can convert.
At scale, geographic bleed creates an account-wide accuracy problem. The algorithm has been trained on a dataset that includes thousands of sessions from users who were never viable customers. Its conversion probability predictions are lower than they should be for legitimate in-territory queries, because those queries are being evaluated alongside the noise from interest-based geographic matches.
Switching to Presence Only targeting, building account-level geographic exclusions, reviewing the User Location Report weekly, and running experiments before major changes does not just stop the waste. It removes the noise from your training data. The algorithm then makes more accurate predictions from a cleaner dataset, which produces better bid decisions, which produces better conversion performance, which produces better training data.
The compounding return on signal integrity is the same principle at work in every other dimension of Google Ads governance. Clean data trains better models. Better models find better customers. Better customers generate better data. The geographic hygiene work pays forward into every other optimization you do.
Frequently Asked Questions
What is the difference between "Presence or Interest" and "Presence Only" in Google Ads location targeting? Presence or Interest serves ads to users who are physically in your target locations AND to users who have shown interest in those locations through their search history, regardless of where they are physically located. Presence Only serves ads exclusively to users the system has verified are physically in your target locations. For most businesses with geographic service constraints, Presence Only is the correct setting. Presence or Interest is the default but it routinely leaks budget to users who cannot be served and feeds low-quality data to Smart Bidding models.
How do I find which locations are actually costing me money in Google Ads? The standard campaign interface shows matched location data, not user location data. To see where users actually were when they clicked your ads, navigate to Insights and Reports, then Report Editor. Create a custom table with the User Location dimension and filter by spend with zero conversions over 30 days. These are your highest-priority exclusion candidates. Cross-reference against Google Analytics bounce rate by location for additional confirmation.
Why should I exclude locations at account level rather than campaign level? Account-level geographic exclusions act as hard boundaries that campaign-level settings cannot override. For locations you never want to serve under any circumstances, account-level exclusions are the correct mechanism. Campaign-level settings can be accidentally changed, overridden by automated recommendations, or simply missed when setting up new campaigns. Account-level exclusions apply automatically to every current and future campaign in the account, including Performance Max and Demand Gen campaigns that otherwise resist campaign-level geographic governance.
Is radius targeting better than zip code targeting in Google Ads? For most local service use cases, yes. Google's IP-to-zip code matching fails frequently in high-density metro areas due to the complexity of overlapping zip code boundaries. Radius targeting sets a simple geometric boundary from a central point that the system can match against IP data more accurately and consistently. It also produces cleaner User Location Report data, making geographic waste easier to identify and exclude. The exception is businesses with hard municipal licensing boundaries where zip code precision is legally required.
How does geographic bleed affect Smart Bidding performance? Smart Bidding trains on every conversion event and non-conversion event in your account. Sessions from interest-matched users in irrelevant locations generate non-conversion signals that the algorithm incorporates into its conversion probability predictions. Over time, this degrades bid accuracy for legitimate in-territory queries because the algorithm has been trained on a dataset contaminated with low-intent geographic noise. Switching to Presence Only targeting and building geographic exclusions removes this noise from the training data, which improves bidding accuracy and conversion performance across the entire account.
What is proximity bidding and how do I set it up? Proximity bidding applies bid multipliers to users based on their physical distance from a specific location. In Google Ads, create bid adjustments under the location settings of your campaign and set a radius target around your primary location. Apply a higher positive modifier for your closest radius (such as +75% for 3 miles) and a lower modifier or no modifier for outer radii. This concentrates your highest bids on the users most likely to convert, who are typically those closest to your physical location or service hub.
Sources
- Google Ads Help — Target Ads to Geographic Locations
- Google Ads Help — About Measuring Geographic Performance
- Google Ads Help — Exclude Ads from Geographic Locations
- Google Ads Help — About Bid Adjustments
- Search Engine Land — What Changed in Google Ads Locations Reporting and Why You Need a Custom Report
- Search Engine Land — Google Adds Location Targeting Controls to Demand Gen Campaigns
- Optmyzr — How Optmyzr's Geo HeatMap Solves Common Google Ads Location Targeting Challenges
- Greenlane — Are You Checking User Location Reports in Google Ads?