Your weekly guide on everything AI marketing. Covering everything from AI marketing news, tips, deep dives, events, podcasts, jobs and much more. Never miss a beat with our 5-minute newsletter.
In today’s email:
Roku launches a 24-hour channel where even the ads are AI-made
DTC brands point AI at pricing and operations, not just copy
AI ad buyers start behaving like television buyers
Plus: everything Claude writes for you now carries an invisible watermark, a chief AI officer on why discernment beats execution, and get in early as a founding member of GoHalves.
The Top 3 Stories
1. Roku now runs a channel where the ads are AI too
Roku has launched Fairground AI, a free ad-supported channel running 24 hours a day of entirely AI-generated programming. It bills itself as the first channel devoted to AI Cinema, featuring "over 100 of the best AI filmmakers from around the world". The advertising running against that programming is AI-generated as well.
Why it matters: An entire ad environment now exists in which nothing, including the spots, was ever filmed. 📰 The A.V. Club
2. DTC brands are pointing AI at pricing, not just copy
At eTail Boston, Murphy Door chief executive Jeremy Barker described running AI across a catalogue of 300 million design configurations. "We currently turn a product roughly every six minutes. And the prices are actually going down, not up," he said. His stated goal is zero big-box presence and 100% direct to consumer.
Why it matters: The clearest margin case for AI right now is being made in pricing and operations, not in content output. 📰 Modern Retail
3. AI ad buyers are starting to behave like TV buyers
DataBeat found that buyers using AI agents took part in 86% fewer auctions than conventional demand, while clearing prices 13.4% lower. Fewer, larger, negotiated commitments look far more like traditional television buying than open programmatic. Research in April put agentic AI inside 46% of surveyed agencies, but only 22.1% used it for media buying strategy.
Why it matters: If agents consolidate spend into fewer paths, your inventory either makes the shortlist or it disappears. 📰 PPC Land
More trending AI marketing news from last week
LLM referral traffic converts at 20%, beating paid search by 61%.
42% of US millennials would let AI agents spend up to $250.
Databricks closed a $5bn round at a $190bn valuation.
Nvidia lined up over $500bn of outside capital for AI infrastructure.
A draft US letter tells 35 countries to pick an AI side.
California's SB 903 would bar chatbots from posing as therapists.
Together with GoHalves
Helping brands grow and scale faster through AI-driven, co-marketing matchmaking.
Ads are a bid war and cold outreach is dead. The channel that still converts is a warm recommendation from a brand your buyers already trust, and industry studies show partnerships can cut acquisition cost by around a third. The problem is the graft: researching who to even partner with, weeks of emails and chasing, and no real way of knowing who is actually right for you.
GoHalves' AI does that research for you. Its matching engine analyses each side's audience data to find verified, non-competing brands that genuinely fit, then opens the conversation with a campaign drafted for both, so nobody makes the first move. Either side can tweak it before anything runs.
Joining the waitlist is free and commits you to nothing. Matching opens when the first cohort fills, so joining now puts you in it.
Quote of the week
We have spent years rewarding execution. The next decade rewards discernment.
Trending AI tools for marketers
Here are 4 trending tools to explore this week. Quick, practical upgrades to your AI marketing arsenal:
Outcome turns your existing content into personalised audit reports that convert cold leads.
BrowserAct pulls clean competitor and market data from any site by prompt.
Human Behavior spots rage clicks and drop-off, then acts without you digging through dashboards.
isolate.video polishes rough screen recordings into launch-ready product demo videos.
TV, podcasts & streaming
Better Offline, "AI Creativity Doesn't Exist With Chloe Radcliffe and Ed Ongweso Jr."
Half of US consumers now prefer brands that avoid generative AI in their advertising and content, Axios reported last week. Ed Zitron and his guests push the point further, arguing the penalty lands on the person rather than the work, because audiences withdraw respect from anyone who writes with AI even when the writing is good, listen to it here.
AI training
How to build a personalised lead magnet using Outcome
The overview:
Outcome turns an existing blog post or webinar into an interactive assessment that hands every visitor a bespoke action plan.
Step-by-step:
Pick your best-performing asset
Choose one post or webinar that already ranks or converts.
Depth beats freshness, because the tool needs real substance to generate useful questions.Upload it and set the verdict
Paste the URL or transcript and say what conclusion the reader should reach.
Vague framing produces generic reports that nobody forwards to a colleague.Trim the question set
Cut the generated questions down to roughly five.
Completion rates fall off sharply past that, and every extra question costs you leads.Write the recommendation logic
Map each answer path to one specific next action.
Good output reads like advice from a consultant, not like a score out of ten.Embed it and capture the answers
Drop it on the matching landing page and pipe responses into your CRM.
Those answers are first-party segmentation data, so treat them as fields, not as a PDF.
Pro tip: Run the same asset again with a different buyer in mind and you have a second lead magnet without writing a word.
The weekly deep dive
Your AI-written copy can now be detected
Transparency obligations under Article 50 of the EU AI Act became applicable on 2 August, and Anthropic answered by marking Claude's output for every user in the world rather than just the European ones.
What actually shipped
Text from Claude models launched on or after 2 August now carries an invisible watermark built on Google DeepMind's SynthID-Text, which nudges word choices to leave a statistical pattern a reader cannot see. Files are handled differently, with a cryptographically signed C2PA credential tucked into the metadata. Around 190 organisations signed the same EU code, Microsoft, Google, Meta and OpenAI among them.
The net has large holes
Short passages offer too few word choices to detect anything. A heavy paraphrase or a retype strips the signal, and screenshotting a file removes the metadata outright. Anthropic concedes that light editing probably will not remove the watermark while a full rewrite will, which means the crudest workaround is also the most effective one.
Detection is arriving before due process
The detection API is promised but not shipped, and no accuracy threshold or dispute procedure has been published alongside it. That ordering is the part marketers should care about, because the first person to run a check on your copy is more likely to be a client or a journalist than you. Right now there is no published route to contest a result.
Our takeaway: Assume anything Claude touches is traceable from here, and settle your disclosure position with clients before somebody else settles it for you.
📖 Read the full article at Politico
