A/B testingadsanalytics

How to Use Placement Exclusions to Protect A/B Test Validity

UUnknown

2026-02-18

10 min read

Use account-level placement exclusions to cut noisy traffic and keep your landing page A/B tests valid — a 2026 playbook for creators and publishers.

Stop noisy traffic from wrecking your conversion experiments — fast

If your landing page A/B tests feel like they're drowning in randomness, you're not alone. Creators and publishers running conversion experiments in 2026 face a new class of noise: AI-driven ad placements, more opaque automation formats (Performance Max, Demand Gen), and fragmented inventory that funnels low-intent or bot traffic onto specific placements. The result: biased results, underpowered tests, and wasted decisions.

The single most effective quick-win

Use account-level placement exclusions to remove repeat sources of noisy traffic before you launch an experiment. In January 2026 Google Ads added a centralized setting to block placements across an entire account, and that makes exclusions a practical, scalable guardrail for maintaining A/B test validity.

Google Ads has introduced account-level placement exclusions, allowing advertisers to block unwanted inventory from a single, centralized setting. — Google Ads rollout, Jan 15, 2026

Why placement noise kills A/B test validity

Two things break conversion experiments: bias and variance. Noisy placements amplify both.

Bias: If a particular publisher or app sends largely non-converting traffic — or users who never see your page long enough to decide — and that placement feeds disproportionately into one variant, the experiment’s point estimate will be skewed.
Variance: High-volume but low-quality placements inflate variance. That reduces statistical power and makes it hard to reach significance without huge samples.

In 2026 that problem is worse because Google’s expanded automation formats route traffic dynamically. Account-level exclusions are now one of the few practical levers that work across Performance Max, Demand Gen, YouTube, and Display campaigns to consistently stop bad placements from contaminating your experiments.

How account-level placement exclusions protect tests (the logic)

Prevents predictable contamination: If certain placements repeatedly contribute low-quality clicks, excluding them removes that predictable source of bias.
Reduces unexplained variance: Removing outlier sources compresses variance, meaning smaller sample sizes reach statistical significance.
Centralizes guardrails: With account-level rules you don’t need to remember to add exclusions per campaign — all new campaigns inherit the same quality baseline.

Before you exclude: audit traffic quality (5-minute checklist)

Don’t blindly block placements. Exclusions are a surgical tool — use them after you identify the offenders.

Export last 90 days of placement reports from Google Ads and Display & Video 360 (if used).
In your analytics (GA4, server events, or your CDP), group performance by placement domain, app bundle, and YouTube channel ID.
Calculate these per-placement signals: CTR, session duration, pages/session, conversion rate, and return-on-ad-spend (ROAS) or cost-per-acquisition (CPA).
Flag placements with low session duration (< 3s median), CTR unusually high but conversion near 0, or extremely high bounce rates.
Cross-reference with fraud/bot signals (high event rates from single IP ranges, impossible device/browser combos).

Example BigQuery sanity check

Run a quick query to find placements with high spend but low conversions. Replace table names for your setup:

SELECT
  placement_domain,
  SUM(cost) AS total_cost,
  SUM(conversions) AS total_conv,
  SAFE_DIVIDE(SUM(conversions), SUM(clicks)) AS conv_rate
FROM `project.analytics.ads_traffic`
WHERE date BETWEEN DATE_SUB(CURRENT_DATE(), INTERVAL 90 DAY) AND CURRENT_DATE()
GROUP BY placement_domain
ORDER BY total_cost DESC
LIMIT 100;

How to build an experiment setup that avoids contamination

Follow a structure that treats traffic quality as an experimental factor, not an afterthought. This four-part setup reduces leakage, keeps randomization honest, and makes results auditable.

1) Centralize placement controls before you launch

Create and apply an account-level placement exclusion list in Google Ads to stop low-quality placements from ever reaching test variants.

Go to Tools > Shared library > Placement exclusions (Google Ads), create an exclusion list, add domains, app IDs, and YouTube channel IDs.
Apply the list at the account level so Performance Max and Demand Gen campaigns inherit the same guardrails.
Document the list in your experiment plan so reviewers can audit it.

2) Use deterministic user-level randomization

Randomize at the user ID or hashed client ID, not at the session or ad click level. That prevents the same user from being exposed to both variants across different channels — a common contamination vector.

Server-side experiment assignment is ideal: hash a stable ID and allocate to groups.
If you must randomize client-side, use a persistent cookie for the duration of the test.
Log assignment and exposures in analytics so you can trace cross-channel contamination.

3) Stratify and block by traffic source

Some traffic sources differ dramatically in behavior. Stratify your randomization by high-level traffic channels (paid search, display, video, organic) and by device type so each variant receives equivalent mixes.

Create strata before sampling — e.g., Mobile-Paid Display, Desktop-Organic Search.
Within each stratum, randomize users equally to variants.
Monitor conversion lifts per stratum as well as overall.

4) Add a QA holdback and contamination detectors

Include a small holdback group (e.g., 1–5%) that receives no paid traffic or receives a canonical control; this helps detect external shocks and attribution shifts. Implement automated detectors that monitor imbalance in source distribution across variants.

Use an automated job to compare variant attribution by placement every hour during the test.
Set alerts for drift: if any placement contributes >X% difference between variants, pause the test and inspect.
Audit UTM parameters — inconsistent or missing UTM tags are a common cause of misattribution.

Applying Google Ads account-level exclusions: practical steps

Here’s a practical playbook for using the new Google Ads feature to protect experiments.

Export placement performance, identify the worst offenders (low conv rate, high cost).
Create a named exclusion list in Google Ads: “Experiment Quality Guardrail — Jan 2026”.
Add domains, app bundle IDs, and YouTube channel IDs to that list.
Apply at the account level and document the change in your experiment runbook.
During the live test, re-check the placement report weekly and update the list if new offenders appear.

What to exclude (and what to think twice about)

Exclude: low-viewability networks, known ad-farm domains, high-CTR but zero-conversion apps, and certain YouTube channels that drive accidental clicks.
Consider: niche publishers that convert well for specific verticals — exclude only after checking conversions by audience segment.
Don’t rush to exclude every high-traffic placement: some can look noisy but actually drive valuable micro-conversions or assists.

Measuring validity: statistical steps and sample-size guidance

Even with exclusions, you still need solid statistical procedures. Here’s a quick guide to power, sample size, and guarding against false positives in 2026’s fast-paced experiments.

Sample size basics

Run a power calculation using your baseline conversion rate, the minimum detectable effect (MDE) you care about, and your chosen power (usually 80–90%). Exclusions typically reduce variance — which lowers required sample sizes — but account for that conservatively.

Example: baseline conversion 3%, MDE 10% relative (0.3% absolute), power 80%, alpha 0.05 → you'll need N per variant. Use an online calculator or your stats library (R, Python) to compute exact numbers.

Statistical significance and adaptive tests

Many teams use sequential or adaptive testing to get quicker results. If you do, correct for multiple looks (alpha spending) and ensure your stopping rule is pre-registered. Exclusions don’t change the math — they only improve the data quality that your tests rely on.

Post-hoc contamination checks

Compare the distribution of placements across variants using a chi-square test to detect imbalance.
Check metrics like time-on-page, bounce rate, and device mix by variant — large disparities indicate contamination.
Use the holdback group to measure external shocks.

Detection and remediation: what to do if contamination appears

Even with the best plan, contamination can occur. Here's a triage playbook.

Pause the experiment if your alerts flag a major imbalance.
Identify the offending placement(s) via placement and UTM reports.
Exclude them at the account level (adds immediate guardrail across campaigns).
Recompute your power/sample-size based on the cleaned traffic and extend the test if needed.
Document the event and the remediation in the test log so future tests can learn from it.

Real-world example (publisher case study, anonymized)

Context: a mid-sized publisher ran a conversion experiment on a new lead-gen landing page. After 10 days, variant B looked worse by 20% but the sample size was small. An audit showed 40% of traffic to variant B came from a syndicated content app with extremely short sessions and near-zero conversions.

Action taken:

Paused the experiment.
Added the app bundle ID to an account-level exclusion list (applied to Performance Max and Display).
Restarted the test and rebalanced randomization to ensure equal strata distribution.

Outcome: After excluding the noisy app, the variance dropped and the expected lift reappeared for variant B. The publisher reached significance 6 days later with a smaller total sample than originally projected.

2026 trends that make exclusions even more important

Automation-first ad products: Google’s auto-bidding and placement algorithms are ubiquitous. These systems can route high volumes of low-intent traffic quickly — account-level exclusions are one of the few universal controls.
Cookieless measurement: As third-party identifiers decline, placement signals become noisier. Excluding known poor placements helps limit measurement leakage.
AI-generated ad inventory: Programmatic formats increasingly include AI-curated inventory where quality varies — manual exclusions are required to enforce brand and experiment safety.
Increased ad fraud sophistication: Fraud rings adapt fast. Maintain exclusion lists and refresh them from fraud detection partners frequently. See a practical case-study template for dealing with fraud signals.

Operational checklist before launching any conversion experiment

Run a 90-day placement quality audit.
Create and document an account-level exclusion list in Google Ads.
Choose deterministic user-level randomization and persist assignment.
Stratify randomization by major traffic channels and device type.
Compute sample size and power; include contingency for exclusions.
Enable automated drift detectors and holdback groups.
Plan remediation steps and document them in the experiment runbook.

Quick reference: sample BigQuery contamination detector

Use this query to compare placement distribution across variants (adapt to your schema):

SELECT
  placement_domain,
  variant,
  COUNT(DISTINCT user_pseudo_id) AS users
FROM `project.analytics.experiment_events`
WHERE experiment_id = 'landing_test_2026'
AND event_date BETWEEN '2026-01-01' AND '2026-01-14'
GROUP BY placement_domain, variant
ORDER BY placement_domain, users DESC;

Final thoughts — treat traffic quality as part of experiment design

In 2026, high-quality experiments require more than clever copy or layout tweaks. They need guardrails at the ad-platform level. Account-level placement exclusions are a powerful, new tool in your toolbox to remove repeat sources of noise — and when combined with deterministic randomization, stratification, and active monitoring, they dramatically improve A/B test validity.

Actionable takeaways

Audit placements before each experiment and build an account-level exclusion list in Google Ads.
Use deterministic user-level randomization and stratify by channel.
Monitor placement distributions during the test and pause if contamination appears.
Recompute power after exclusions — you may reach significance faster with cleaner traffic.

If you want a plug-and-play checklist, downloadable CSV template for exclusion lists, and the sample BigQuery scripts pre-populated for GA4 schemas, get our Experiment Quality Kit — designed specifically for creators and publishers running landing page conversion tests in 2026.

Call to action

Protect your next experiment: download the Experiment Quality Kit, import the exclusion CSV to Google Ads, and follow the runbook to launch cleaner, faster, and more reliable A/B tests. Book a 20-minute audit with our team if you need help implementing account-level exclusions and building a contamination-resistant experimental setup.

Unknown

Contributor

Senior editor and content strategist. Writing about technology, design, and the future of digital media. Follow along for deep dives into the industry's moving parts.