AdSense experiments: how they run, what you can test, and how to read the results
Key takeaways
- An AdSense experiment compares your current settings (the original) with a variation on a share of your traffic, so you can see the effect on earnings before applying a change.
- Google's AdSense blog described the Experiments page in January 2018 as a place to experiment with ad unit settings and with allowing and blocking ad categories.
- In the same 2018 post, Google said experiments marked Ready to complete would be auto-completed if no winner was chosen after 30 days.
- Auto optimize lets Google run optimization experiments for you and automatically apply improvements; Google notes that high traffic lets experiments run faster.
- Separately, AdSense may run its own limited experiments to optimize the auction, without modifying advertisers' targeting settings.

AdSense experiments let you test a change to your ad settings on part of your traffic before applying it everywhere. AdSense shows your current settings (the original) and a variation side by side, then reports how each one earns. You pick a winner, or let Auto optimize run experiments and apply improvements for you. The hard part is reading results honestly.
What is an AdSense experiment?
An experiment is a controlled comparison inside your AdSense account. Instead of changing a setting for the whole site and comparing this week with last week, AdSense serves two versions at the same time:
| Element | Role |
|---|---|
| Original | Your current settings, used as the baseline |
| Variation | The same settings with one change, such as an ad category allowed instead of blocked |
| Traffic split | Part of your traffic sees the variation, the rest sees the original |
| Results | Earnings and related metrics for each version, over the same period |
| Winner | The version you keep, applied to all traffic when you complete the experiment |
The value of running both at once is that seasonality, news cycles and advertiser budgets affect both versions equally. A before-and-after comparison cannot separate your change from those outside forces, as explained in AdSense seasonality.
What can you test?
Google’s own description of the feature has grown over time. In Our continued investment in AdSense Experiments (January 10, 2018), Google wrote that the Experiments page “has allowed you to experiment with ad unit settings, and allowing and blocking ad categories to see how this affects your earnings,” and announced ad balance experiments: “You can now run experiments to see how different ad fill rates impact revenue and ad session lengths.”
| Experiment area | What the variation changes | Source of the description |
|---|---|---|
| Ad unit settings | How a given ad unit is configured | 2018 AdSense blog |
| Ad categories | Allowing a category you currently block, or the reverse | 2018 AdSense blog |
| Ad balance | The share of ads shown (fill rate) | 2018 AdSense blog |
| Ad format settings via Auto optimize | Combinations of ad settings, tested by Google | 2022 AdSense blog |
| Search ads styling (AdSense for Search) | Custom styling variants | Custom Search Ads developer reference |
The list in your account today may differ from these dated descriptions. Google’s help center also has separate articles on auto ads experiments and blocking controls experiments, which describe current limits such as how many experiments can run at once. Those pages could not be retrieved for this guide, so check them in the AdSense Help Center before you start, rather than relying on third-party summaries.
Search ads styling experiments
If you run AdSense for Search, styling can be tested too. The Custom Search Ads web reference documents an isExperimentVariant parameter passed to the ad loaded callback, “set to true if a custom styling experiment is running and the variant of the experiment is chosen to render the ads.” If your page code reacts to how ads render, it can log this flag so you know which version a visitor saw.
How do you run an experiment step by step?
Interface labels change, so these steps describe the logic rather than exact button names.
- Write the hypothesis first. For example: “Allowing a currently blocked general category will raise earnings without hurting user experience.” One change per experiment.
- Check the baseline. In AdSense reports, note recent page RPM, impressions and earnings for the site or ad unit you will test, so you know what normal variation looks like.
- Create the experiment from the Experiments area of your account, choose the site or settings concerned and define the variation.
- Let it run untouched. Do not change other ad settings on the same site during the test, or you will no longer know what caused a difference.
- Wait for full weeks. Weekday and weekend traffic often behave differently. Cover at least one complete weekly cycle, and more if traffic is low.
- Read the results, as described below, and decide.
- Apply or discard. Choosing the variation as winner applies its settings; choosing the original keeps things as they were.
- Record what you learned. Keep a simple log: date, hypothesis, result, decision. Google warned in 2018 that experiments “completed more than one year ago” would be deleted, and pointed to a “Download Data” button to keep the data.
How do you read the results?
The results page compares the variation with the original. The question is not “which number is higher today?” but “is the difference large and stable enough to trust?”
A worked illustration
Hypothetical numbers, invented for illustration only. Suppose an experiment splits traffic evenly and, after two weeks, shows:
| Version | Page views | Estimated earnings | Page RPM |
|---|---|---|---|
| Original | 40,000 | $120.00 | $3.00 |
| Variation | 40,000 | $126.00 | $3.15 |
Page RPM is earnings divided by page views, times 1,000: $126 / 40,000 x 1,000 = $3.15. The variation is 5% higher. Before applying it, ask:
- Is 5% outside normal noise? If your daily RPM routinely moves by 10% or more, a two-week difference of 5% may be chance. Look at the trend across the whole period, not the last few days, and if the results page shows a status such as “Ready to complete” (the label Google used in its 2018 post), treat it as the minimum, not a reason to stop at the first good day.
- Did both versions get comparable traffic? A split that is very uneven, or a traffic spike that hit one version harder, makes the comparison weaker.
- Is the gain coming from where you expect? Check impressions and clicks for each version. A gain driven by a handful of unusually valuable clicks is less durable than a gain spread across impressions.
Use page RPM or total earnings for the decision, not CTR or CPC alone. A variation can lower CPC and still earn more, as explained in page RPM vs impression RPM.
What about user experience?
Revenue is not the only output. The 2018 blog introduced an “Ad session length” metric on the results page and noted that “longer ad session lengths are usually a good indicator of a healthy user experience.” Whatever metrics your results page shows today, also watch your own analytics for the test period: pages per session, bounce or engagement, and complaints from readers. A variation that earns 5% more this month but drives visitors away will cost more later.
What is Auto optimize?
Auto optimize hands the testing to Google. In Your AdSense 2022 holiday checklist (October 24, 2022), Google described it as a way to “let Google run optimization experiments for you,” testing “different ad settings combinations” and automatically applying “improvements aimed at increasing revenue.” The same post added that high holiday traffic means “your experiments can run faster.”
| Your own experiments | Auto optimize | |
|---|---|---|
| Who picks the variation | You | |
| Who decides the winner | You | Google, automatically |
| Best for | Testing one specific idea, such as a category or a format | Ongoing tuning without manual work |
| Your effort | Set up, monitor, decide | Review what was applied |
| Risk to watch | Deciding too early | Changes you did not plan, so check your settings periodically |
Auto optimize does not replace judgement about your readers. Review which settings changed after an optimization, and compare engagement in your analytics for the period.
Are there experiments you do not control?
Yes. Google’s help page About the ad auction states that “AdSense may run limited experiments designed to optimize the auction.” These may include modifying the standard auction model or mechanics, simulating ad calls and auctions, adjusting publisher settings, and discounting certain bids or modifying their priority. The page adds that “advertisers’ ad targeting settings will not be modified.”
These auction experiments are not visible as entries on your Experiments page. They are one more reason to read your own experiments as comparisons run over the same period, rather than as precise predictions of future earnings. The auction itself is covered in this section’s guide on CPC and CPM bids.
What are good first experiments?
Start where the decision is genuinely uncertain and the change is easy to reverse.
| Idea | Why it is a good candidate |
|---|---|
| Unblocking a general category you blocked “just in case” | Google’s auction guidance links fewer restrictions to more competition; an experiment shows whether that holds for your site |
| Comparing ad balance levels | Tests whether showing fewer ads loses little revenue, which can improve user experience |
| Trying an auto ads format on one site | Shows the earnings effect before you roll it out elsewhere, see auto ads vs manual ad units |
Keep sensitive categories you block for brand or audience reasons out of testing. Some blocks are editorial decisions, not revenue questions. The options are covered in blocking controls.
Common mistakes
- Ending the test after a good day. Early results swing; wait for full weeks and a stable trend.
- Changing several things at once. You learn nothing about which change mattered.
- Running a test during an unusual period without noting it. Holiday peaks or a viral article can distort both versions, and the result may not hold afterwards.
- Judging by CTR or CPC. Decide on earnings and page RPM, and check user experience alongside.
- Forgetting to log results. Old experiments can be deleted; download or note the data you want to keep.
- Testing to justify invalid behaviour. No experiment makes click encouragement or misleading placements acceptable.
Frequently asked questions
What can I test with AdSense experiments?
Google's AdSense blog described the Experiments page as a place to experiment with ad unit settings and with allowing and blocking ad categories, and later added ad balance experiments that test different ad fill rates. The current list of experiment types, such as auto ads and blocking controls experiments, is shown in your account and in the official Experiments help pages.
How long should an AdSense experiment run?
Long enough to cover full weekly cycles and enough traffic for a stable result. Google noted in 2018 that experiments marked Ready to complete were auto-completed after 30 days if no winner was chosen, and in 2022 that high traffic lets experiments run faster. Check the current limits in the Experiments help pages before starting.
What is the difference between Auto optimize and my own experiments?
With your own experiments you choose the variation, watch the results and pick the winner. With Auto optimize, Google runs optimization experiments for you, tests different settings combinations and automatically applies improvements aimed at increasing revenue. Auto optimize saves time, while manual experiments let you test one specific hypothesis you care about.
Does Google run experiments on my site without asking?
Google's ad auction help page says AdSense may run limited experiments designed to optimize the auction, such as modifying the auction mechanics, simulating ad calls, adjusting publisher settings or discounting certain bids. It also states that advertisers' ad targeting settings will not be modified. These are separate from the experiments you create yourself.
Sources
- Our continued investment in AdSense Experiments · Google (The Keyword blog), January 10, 2018
- Your AdSense 2022 holiday checklist · Google (The Keyword blog), October 24, 2022
- About the ad auction · Google AdSense Help
- Custom Search Ads web reference · Google for Developers
Related guides
- AdSense blocking controls: advertiser URLs, categories, the Ad review center and the cost of blocking
Block advertiser URLs, general and sensitive categories or single ads in the Ad review center: official limits, steps, site-level options and the revenue trade-off.
- AdSense reports: dimensions, metrics, saved reports, CSV exports and finding what changed
How AdSense reports work: breakdowns and metrics with official definitions, saved reports, CSV exports via the Management API, and a method to find what changed.
- Page RPM vs impression RPM: why they differ and which one to track
Why page RPM and impression RPM move differently, the formula that links them, and how to use both to diagnose AdSense revenue changes correctly.
- Auto ads vs manual ad units: which one to use, and how to combine them
Auto ads or manual ad units? Compare control, formats, page and area exclusions, and how AdSense combines both on one page without double work.
- AdSense seasonality: why RPM changes through the year and how to read it without panicking
Why AdSense RPM moves through the year: advertiser budgets, auction competition and the Q4 to January swing, with official retail data and a way to read it calmly.