The Myth of “99% Brand Safe”

Brand safety tools claim platforms they measure are ‘99% safe.' Here's why those numbers are probably not real.

Finally, YouTube A/B Thumbnail Testing Is (Almost) Here

One of the changes marketers have most wanted of YouTube was the ability to test thumbnails in a proper multivariate test.

They’ve been testing it for a year now with a small handful of creators, and now they say they’re ready to roll it out to everyone in the next one or two weeks.

How it works

You’ll give YouTube three images to use as thumbnail tests.

YouTube will show those thumbnails evenly across your video's audience, and will select a winning thumbnail based on which one generates the most Watch Time. That’s an interesting design choice since, at first, you’d think they would test based on clicks, but YouTube has always used Watch Time as its north star metric so I suppose it’s not surprising. There also is no way to have it measure any other numbers.

You can check the status of your test by going to the video details page and clicking View Test Report. From there, you'll see how much time is left for the test and how they're performing.

You can also stop the test and manually set one of the thumbnails as the winner.

If you let the test run, the results may take a few hours or up to a couple weeks to be finalized. The more thumbnail impressions that your video receives, the faster your test will finish.

If there's a statistically significant result, you'll see a winner label and YouTube will automatically update your video thumbnail to this winner when the test is done.

You might also see a “preferred” label, which means the results were not statistically significant, but there was a thumbnail that performed slightly better.

If there's no clear winner, the first thumbnail will be selected and shown to your audience.

You can also manually select the video thumbnail you want to use.

Where it doesn’t work

This new test will not work on Shorts or Live videos, and it also won’t be available for videos intended for children or mature adults.

Other updates

YouTube also had a few other updates including a way to use AI to skip to the good parts of a video. That’s running now on the Android YouTube app for Premium subscribers.

Reddit/DoubleVerify — Is “99% Safe” Even Real?

Reddit this week announced a new partnership with DoubleVerify to provide brand safety and suitability measurement for advertisers.

And while Reddit said that more than 99% of impressions appeared next to content it determined was brand-safe, some analysts question that 99%.

So while the specific safety number might be a little less trustworthy than we’d like, even just having this partnership will likely help Reddit continue to grow through its adolescence into a more mature platform.

TikTok Testing Photo-Search in TikTok Shop

TikTok is testing letting users to take or upload a photo to find similar products in TikTok Shop.

It shows up as a camera icon in the search bar once you’re in the Shop UI.

This tech is already in Google’s Lens tool and Amazon’s app.

TikTok Shop now has more than 15 million sellers worldwide — more than a half-million in the U.S.

Google AI’s Möbius Strip From Hell

We all had a bit of fun with Google’s AI advising people to add glue to their pizza to keep the cheese in place.

Journalists, myself included, made fun of it all. Katie Notopoulos from Business Insider actually made it and ate it. 

And many people watched her video.

Including, apparently Google’s AI training bot.

Because now, when The Verge asked Gemini how much glue to add to pizza, rather than having been corrected by the AI engineers with a hard-coded “don’t do that” answer, Gemini now uses Katie’s post as a source for its answer.

– 30 –