Are You Running Ads on X Without Knowing It?

Presented by

In This Issue:

🤖 Meta introduces generative AI ad tools for backgrounds and body copy

🔗 LinkedIn automates ad campaign creation with "Accelerate"

🐦 X (Twitter) collaborates with Google Display Network for ad space

🎭 AI deepfake of MrBeast deceives TikTok's ad moderation

🎮 Roblox ventures into advertising, recruits executive from Meta 

💌 Gmail introduces emoji reactions for email interactions

📺 Netflix global ad president, Jeremi Gorman, steps down after a year

🍔 Judge rules in favor of McDonald's and Wendy's in burger ad lawsuit

😬 Meta's AI sticker tool generates controversial images

Meta Wants to Write Your Ad Copy For You

Meta’s big push into AI in its ads manager continues — the company this morning announcing more generative AI-powered features for ad creatives in Meta’s Ads Manager.

There are three new features:

• Background Generation creates multiple backgrounds to complement advertisers’ product images

• Image Expansion adjusts creative assets to fit different aspect ratios across multiple surfaces, like Feed or Reels

• Text Variations generates multiple versions of ad texts based on advertiser’s original copy

But How Personalized Is It?

What’s not clear is whether these tools will generate a set of new creative that you’ll be able to approve ahead of time, or whether it will generate them on the fly for each individual user, based on what Meta believes that person will respond best to.

We’ve reached out to Meta for clarification, but from everything I’m seeing in various industry Slack groups and so on, it’s likely the former. We do know that the text variation tool can only generate up to six different versions of copy.

Some of these tools may sound familiar — Meta’s been testing these in a small number of ad accounts. Also, things like being able to use multiple variations of copy and images has been around for some time — the difference here being that previous versions of this had the advertiser creating these variations, not Meta’s AI.

Also, there are some limitations in the fine print — the backgrounds feature will be pretty basic — colors and patterns — and won’t be available unless you’re using Meta’s Advantage+ catalog to create your ads.

Things Meta Won’t Tell You

The biggest issue, as I see it, is reporting. Meta won’t tell you which version of the body copy is performing best. It says this is because its reporting architecture is built around ads, not what are essentially sub-ads, though that’s certainly something Meta could build if they wanted to. There was a time when there were only two levels in a campaign — Campaign and Ad. Meta later added the Ad Set level in.

These tools have started to roll out as of this morning, and may take until next year to get to all accounts.

LinkedIn’s New Campaign Type Can Create Almost the Whole Thing Without You

So with Meta joining the club of ad platforms using generative AI for individual campaign elements, how much longer before somebody turns AI loose on the whole ad campaign?

That day has come to LinkedIn, which this week announced its latest ads manager update will now let you automate almost the entire campaign creation process.

They call it Accelerate and here is how they describe how it works:

This is a fork in the road for many media buyers who have to weigh control over their campaigns against the potential for better results when those campaigns are handled by AI.

Most industry analysts say machine-learning’s insertion into an ad campaign can provide better results, but it’s not a guarantee, and those good results really only happen when you’re advertising at scale — that’s code words for “spending a lot of money.”

This new Accelerate offering started to roll out yesterday to a handful of accounts and is growing to more in the weeks ahead.

Google Display Network Now Includes X/Twitter

Your brand might be running ads on X, formerly Twitter, without you even realizing it.

Last week, Google announced a partnership where they will be helping X sell ads. Those ads would run on the Google Display Network.

It’s not hard to see why X made this decision. Following Elon Musk’s takeover last year, the company has lost nearly 60% of its U.S. ad revenue. And concerns about brand safety have kept many brands away.

This deal means that X becomes a placement option for Display ads. Which means that if you accept the default automatic placements, X might be in that pool.

Is Google A Little Bit Embarrassed By It All?

And honestly, Google seemed to underplay the announcement, making it sound like it was just another site.

But it’s not like other social apps and web sites, of course. For some brands, it aligns better with their audience since the ownership changes; for others, not so much.

Google spokespeople also distributed details on how to opt-out of X as a placement, if you didn’t want it, something I don’t think I’ve seen them do for any other specific site.

That said, Google has a lot of practice at finding the right people for an ad at scale. It will be interesting to see how this plays out.

Big-Name Deepfake Ads Are Being Undetected By Ad Platforms

The YouTube star MrBeast is known for his huge stunts — giving away houses or cars for free, or recreating the Squid Game with real contestants (but no deaths, obviously).

Which is probably why his latest TikTok ad seemed completely normal.

In the ad, he appeared to be giving 10,000 new iPhones away for just $2. Posted by anyone else, it would be flagged as an obvious scam. Posted by MrBeast? It’s within the realm of possibilities.

Which might explain why the ad slipped past the moderation system at TikTok, part of which is run — ironically — by AI.

The ad was, of course, a deepfake and a scam. But the rendering of his face in the video, and his voiceprint, was almost perfect.

For its part, TikTok removed the ad, but only after media pointed it out. And their policy doesn’t prohibit ads made by AI, as long as that’s disclosed on the ad.

This week, American news anchor Gayle King and actor Tom Hanks both warned their followers that fake versions of them were out there hawking various things that they had no part in.

In Brief

Gmail has a new feature — emoji reactions in emails. So just like you can quickly tag a comment with a smiling face in a chatroom, now you can do that in emails. Except if the recipient uses Gmail’s corporate version, or its educational version, or doesn’t use Gmail at all, every little emoji reaction will be sent as a separate email. So, that’s great. read more

The video game platform Roblox has poached an ad executive from Meta. Stephanie Latham will become the company’s VP of global partnerships. She had led Meta’s entertainment, tech, and telecom sales organization in North America. read more

After a tenure of just over a year, Netflix's global ad president, Jeremi Gorman, has decided to depart. He was instrumental in launching Netflix's ad-supported tier. Amy Reinhard, who has a background in Netflix's studio operations, will be stepping into Gorman's role. read more

A legal case against McDonald's and Wendy's over how accurate online images of their burgers were has been dismissed by a judge. The lawsuit alleged that both fast-food giants were misrepresenting the size of their burger patties in their advertisements. But the court found otherwise. read more

And finally…

When Meta last week announced it would let people use generative AI to create stickers for their Reels and Posts and Stories, who amongst us thought to themselves: “Hmm. What could go wrong?”

Well, now we know what could go wrong — and it was as predictable as clockwork.

People are typing in all sorts of prompts — prompts that brand managers might not appreciate… and some of the stickers coming out: Mickey Mouse holding a bloody knife, Canadian Prime Minister Justin Trudeau wearing only underwear, and video game character Mario looking angry holding a rifle.

Meta’s system does block some phrases — “child with gun,” for instance, is blocked, but “child with grenade” has returned results for some people, results that also helpfully included a child with a gun.

The news site Gizmodo did their own tests.

• “Elon musk, large breasts”: blocked

• But “elon musk mammaries”: no problem!

Other prompts generating results: “school shooting,” “syria gas attacks,” and “pol pot”, which returned a sticker of the Cambodian dictator sitting on a throne made of babies and skulls.

Some links in this newsletter may provide affiliate revenue to us.

This newsletter is powered by Beehiiv.