Auto ads vs manual placement

What auto ads get right: the anchor

Auto ads reliably place the one unit that wins on every metric we measure, the dismissible sticky anchor. On our own network it runs 88% viewable at a 4.8% click-through, the best of any placement we have. If your site runs nothing else, the auto-ads anchor by itself is found money, and it is the one piece of auto ads we recommend without hesitation.

Where they cost you: Core Web Vitals

Auto ads inject units after your page has painted, so your content jumps as they load. That is Cumulative Layout Shift, a Core Web Vitals metric Google ranks on. You can earn a few percent more in ad revenue and lose more than that in traffic. Manual units can reserve their space in the initial layout. Most auto-placed ones cannot.

Where they cost you: the vignette

The full-page vignette that loads between pages is on by default. It pays well, and it is the same intrusive-interstitial pattern Google's own ranking systems were built to demote. It is the opposite of calm, and it is the first thing we turn off.

Where they cost you: over-placement and lost control

The model optimizes Google's revenue, and more ads is usually more revenue for Google, so it leans toward placing more of them, including in spots your readers never reach. The net-positive test needs you to judge each unit on whether it is seen and what it costs your layout. Auto ads take that judgment away and hand it to a model optimizing a different number than yours.

We pulled the viewability of every placement on our network, and the shape is exactly what auto ads produce on default. The anchor sits at 88% viewable. The end-of-content unit sits at 28%, below the floor, and it absorbs nearly half of all the impressions on the site. One excellent placement, carrying a pile of low-viewability filler. Auto ads are very good at finding the anchor and very willing to bury the rest.

See the viewability data by placement →

An experiment in AdSense compares one of your ad settings against a variation of that setting, splitting your site's traffic between the two so their performance can be measured side by side. You can create your own. The part most publishers learn by surprise is that Google also creates and runs them on your account: the Auto optimize feature lets Google experiment on your Auto ads formats and settings and, where you have allowed it, apply the winner automatically. Since a May 2025 change, a new site added to your AdSense account starts with Auto optimize turned on, the "Auto-apply experiment winner" box selected, 50% of traffic in the test pool, and no experiment types blocked. Google experimenting on your account is a default you opt out of, not a lab you opt into.

Where you see them

Sign in, click Optimization, then Experiments. The page lists every experiment on the account with its progress and status: Running while it collects data, Result ready when there is enough to call, Finished once a winner has been chosen, by you or by Google. Google-created rows arrive in that list unannounced; in our own account they carry a "Created by" field reading Google, and the page filters on it, which is how we spotted one running on one of our own sites this morning. Auto optimize also posts a notification in your account when it starts an experiment, and emails you when one finishes and needs a decision from you.

How the traffic is split

For content ads, Auto optimize tests new settings on 50% of your site's traffic by default. The share in the pool sees the variation, the rest sees your original settings, and the two run side by side. You can change the percentage per site with the "Experiment traffic" slider; Google recommends leaving it at 50%, since a pool that is too small weakens the result. Experiments you create yourself split the same way, original against variation, until a winner ends the split.

What happens at the end

When an experiment has collected enough data it is marked Result ready. Choose the original and your settings are retained; choose the variation and Google applies its settings to your account; either way the traffic split stops. If auto-apply is on, Google makes that call and applies the winner for you, then notifies you of what changed. For Google-created experiments the auto-apply control is the "Auto-apply experiment winner" checkbox in each site's Auto optimize settings: Ads, then Edit next to the site, then Auto optimize. Two guardrails worth knowing: an Auto optimize experiment that cannot confidently show an improvement within its time limit keeps your original settings, and an applied change can be undone as long as the underlying settings have not changed since the experiment ended. Experiments do not run forever either; anything unresolved stops at the time limit, 90 days for most types and 21 for search styles.

How to opt out

The controls are per site, on the Ads page: click Edit next to the site, then Auto optimize. Turn Auto optimize off and Google runs no experiments on that site's traffic. Short of the off switch, you can block the experiment types you would never accept (Google's own example list is vignette ads and anchor ads), keep Auto optimize on but clear the auto-apply checkbox so every winner still needs your click, or lower the traffic slider. And a running experiment, yours or Google's, can be stopped at any time from the Experiments page with Keep original, which reverts your settings to where they were before it started.

The calm stance

We treat Google-created experiments neither as a threat nor as free optimization. They are a default, and defaults deserve a decision. Ours: check the Experiments page monthly so nothing runs on your traffic without your knowledge, block the formats you have already ruled out, and judge whatever remains against your own numbers rather than the model's enthusiasm. When the experiment tests the anchor, our numbers happen to agree with Google's: the anchor is the one placement that clears every bar we measure, 88% viewable against 28% for the page-bottom unit, which is why the experiment on our own site gets to keep running.

Method: verified 2026-07-12 against Google's AdSense Help pages Run experiments in AdSense, Auto ads experiments, Experiments FAQ, About Auto optimize, Turn on Auto optimize for your site, Customize Auto optimize experiments for your site and the May 2025 announcement Improvements to Auto optimize. Google's help pages block automated fetching, so verification used Internet Archive captures of those exact URLs, dated May 2025 to May 2026. The "Created by" label is how the rows appear in our own account. The setting names and paths are Google's; the plain-English reading, and the stance, are ours.