THE SUNDAE TUESDAE ☀️

Creator marketing intelligence for brand builders

August 18, 2026

The Scariest Thing in This Business Is Watching It Fail in Public

Watching Tate McRae’s Neutrogena video get picked apart, frame by frame, by strangers online, is about as close as this industry gets to a horror movie. Not because a face wipe didn’t wipe. Because for a few hours, a lot of people got to watch a major brand’s approval chain fail in real time, in public, with nowhere to hide. That’s the actual fear in creator marketing. Not “does this creator have brand safety problems.” Whether the thing you handed off will hold up once it’s out of your hands.

Neutrogena had every layer a brand that size is supposed to have: talent, agency, legal, brand. None of it caught a video where the product never touches anyone’s face. The easy version of this story says nobody had the standing to flag it. The harder version is that the moment to catch it probably wasn’t on set at all. She shot it on her own jet, and there’s a good chance nobody from the brand was even there. The real catch point was an inbox: someone watching the finished cut, a year of relationship-building banked on it going well, deciding whether to go back to a major ambassador and her team and say the video she’d already delivered wasn’t good enough. I’d like to think I’d have gotten it right, but I’ve been in the room, and I know the stakes.

That’s the real premise of this business: giving up some control on purpose, to someone whose judgment you’re betting on. It works when the person with the least leverage in the room still has enough standing to say “this won’t hold up.” The brands that got burned this week weren’t the ones who gave up control. They were the ones who never built a room where anyone actually could.

Four more stories below, and in every one of them, somebody let a default setting, a famous face, or a friendly metric make a call that was actually theirs to make.

1. Twitch Auto-Enrolls Streamers to Train Amazon’s AI, and Its Own Product Chief Explains Why

TechCrunch, August 12, 2026

As of August 12, Twitch automatically enrolled every streamer’s content into training for Amazon’s generative AI models; opting out means finding a toggle in account settings most creators didn’t know existed. Chief Product Officer Mike Minton explained the choice on a live Patch Notes stream: “If this was opt-in, nobody would opt in. That’s honestly the answer.” His own Head of Community, Mary Kish, visibly winced on camera, then spent the rest of the stream doing damage control in chat. Minton also couldn’t confirm whether Amazon had already used creator content for training before the policy was announced. The backlash carries real weight: the opt-in request on Twitch’s own feature-request forum has more than 13,000 upvotes, against roughly 250 for other top asks.

Sundae take: AI-training rights are becoming a real, negotiated term in creator agreements, not a hypothetical one, and this is the clearest evidence yet that platforms will default to whichever setting benefits them unless someone pushes back in writing. What’s actually notable here is how well the pushback worked: 13,000 upvotes moved Twitch to ship an opt-out toggle within days, which says the leverage in these fights is more available, and louder, than most brands assume.

2. Phia Knew Its App Was Taking Credit for Sales It Didn’t Drive

Bloomberg, August 11, 2026

Phia, the AI shopping app co-founded by Phoebe Gates and Sophia Kianni, ran a “cookie stuffing” feature that dropped affiliate cookies onto browsers so the app collected commission credit for purchases it had nothing to do with. Bloomberg reports the founders knew about the practice for months before publicly denying wrongdoing. When Phia turned the feature off in July, its average daily revenue fell from roughly $80,000 to between $10,000 and $28,000.

Sundae take: Every affiliate and creator program runs on a chain of attribution that mostly goes unaudited, and Phia is a preview of what that number can look like once someone finally checks. Somewhere between two-thirds and seven-eighths of what looked like performance was attribution theater, not results.

3. Creator Income Still Tracks Reach. That’s Not the Contradiction It Looks Like.

CreatorIQ, “State of Creators 2026,” August 11, 2026

CreatorIQ’s survey of 5,055 creators across 10 regions found brands rank follower count last among partnership criteria, yet creator income still tracks audience size more closely than any other factor, and views outpredict engagement across every platform measured. Most creators are still earning very little: 67% make under $10,000 a year, and only 4% clear $100,000. The survey also found a split by tier: creators overall prefer TikTok for branded content 52% to 43%, but creators earning over $250,000 a year favor Instagram 60% to 30%.

Sundae take: This isn’t really evidence of brands saying one thing and doing another. Paying more for a bigger audience is just math, not hypocrisy — an 18,000-follower creator with a strong engagement rate is still reaching a fraction of the people a 500,000-follower creator reaches, engaged or not. CreatorIQ calls this a fixable “authenticity gap,” worth noting since CreatorIQ also sells the discovery tools built to find the creators that framing says reach-based pricing is missing. The more useful read is the tier split: brands chasing “authentic” engagement from smaller creators to beat the market on price aren’t finding an inefficiency nobody else noticed. They’re buying something different, not something cheaper.

4. Meta Goes to Trial in Oakland While Its Own Numbers Say the Australia Ban Isn’t Working

Courthouse News / Social Media Today, August 10–13, 2026

On August 10, the Ninth Circuit dismissed Meta and TikTok’s appeals, ruling that Section 230 is “a defense against liability, not blanket immunity from being sued” and leaving roughly 2,400 youth-addiction lawsuits intact. Two days later, jury selection began in Oakland in a case brought by 29 states against Meta, the largest test yet of youth social media litigation. In between, Meta disclosed it has removed 750,000 under-16 accounts in Australia since June 2025, up from 550,000 in January, crediting better AI detection. Australia’s eSafety Commission says 81.5% of underage users are still getting into restricted apps, and half say no platform ever asked them to verify their age. Meta’s own recommendation is age verification enforced once, at the operating-system level, instead of platform by platform.

Sundae take: The exposure here is that a platform can be sued over youth-harm claims and simultaneously publish numbers showing its own enforcement doesn’t work. This is not a position any brand wants to be standing near when a media plan gets scrutinized.

Worth Knowing

  • TikTok Shop’s US gross merchandise value nearly doubled to $11.8 billion in H1 2026, up 103% year over year — Momentum Works

  • Whatnot closed a $545 million Series G at a $20 billion valuation, nearly double its October 2025 mark — TechCrunch

  • Only 6% of more than 15,000 branded creator assets analyzed delivered both strong platform engagement and strong brand-building potential — Kantar

  • Spotify introduced a label for AI-generated artists and will exclude them from algorithmic and editorial recommendations — Spotify

❝

From Sundae Collective

We build creator programs with vetting and measurement designed in from the start, not added after something breaks. Take a look →

Working on something in creator marketing and want a second set of eyes? Get in touch.