Transparency in AI is undergoing a massive redesign. You might’ve recently caught the news that Gemini is giving users the option to remove the physical, visible watermarks from their images in favor of embedded metadata. But that’s just the tip of the iceberg.
The story that’s absolutely dominating forums, developer Slacks, and tech circles right now is about Claude. As Forbes recently highlighted, Anthropic is drawing a hard line in the sand: moving forward, Claude will leave an invisible, algorithmic watermark on everything it writes.
For the copywriter drafting brand messaging, the developer generating boilerplate code, and the everyday consumer polishing a cover letter, the blank page is no longer truly blank. There is now an indelible, digital fingerprint attached to the output.
It raises a fascinating question about the future of work, specifically when it comes to "Shadow AI" and corporate compliance.
What happens six months from now? What happens when your company runs a routine compliance audit using detection software that specifically looks for those embedded AI watermarks? Suddenly, a harmless shortcut becomes a massive corporate governance breach.
We are entering an era where you can no longer assume that what you generate in a silo stays in a silo. The breadcrumbs will be there.
Here’s a question: Think about your own daily workflows. Think about the quick tasks you hand off to Claude or the drafts you rely on LLMs to smooth out. If every single one of those outputs carried a permanent, traceable watermark, would your current habits hold up under a compliance microscope? Or could these new transparency measures eventually get in the way of how you operate?
Take a moment to reflect on that as we get into this week's news...
Patrick Patterson
CEO at Level Agency
What's interesting this week
Your tROAS target may be trained on conversion values that were never real. Many ad accounts suffer from inflated conversion values caused by overlapping tracking issues. While account performance may look healthy, the bidding algorithm optimizes against inaccurate data. Eventually, reported ROAS diverges from real revenue, creating serious discrepancies. The essential fix is to reconcile platform conversion values against actual closed revenue recorded in your CRM.
Key takeaway: Run that reconciliation before you trust a tROAS target, since a target set on inflated data has already sent budget chasing conversions that were never real.
LinkedIn wants your target audience between 50,000 and 500,000 people. The number comes from a new four-step ad playbook the platform published, covered by Social Media Today, spanning audience sizing, funnel-stage messaging, format selection, timing, and measurement. Most of it will read as familiar to anyone running LinkedIn at scale. The audience range is the piece worth pressure-testing, because LinkedIn's system needs room to find the right people inside a target.
Key takeaway: Check the audience counts on your live campaigns, since tight targeting is the most common reason B2B delivery stalls before the creative gets a fair test.
Nearly nine in 10 B2B marketing decision-makers expect their investment to rise over the next 12 months.
Forrester principal analyst Mark Ogne, writing in MarTech, argues that a bigger 2027 budget flowing into a planning model built for an older buying environment produces motion instead of impact. His sharpest point lands on optimization, which assumes the underlying system still fits the market and can quietly preserve the complexity that blocks adaptation. He puts the real leadership work in deciding what to stop funding.
Key takeaway: Decide what gets cut before you decide where the increase goes, because capacity to adapt comes from divestment rather than from more money.
Signal, Speed and the New Enrollment Playbook: What $100M+ in Spend Actually Reveals August 20 at 10 AM PT / 1 PM ET, plus a live Q&A.
Brad Stephenson, Level's SVP of Marketing, walks through the 2026 Career Education Performance Benchmarks Report. He’s joined by Christopher Mercado, Director of Marketing at Northbridge University, Joy McClure, Chief Admissions Officer at Tricoci University, and Drew Horne, Partnerships Manager at LeadSquared.
The report is built on more than $100M in managed media paired with LeadSquared admissions data across hundreds of institutions. The session covers benchmarks for cost per inquiry, cost per enrollment, and counselor performance, the shift toward Google's AI Max and Performance Max, and how AI search is reshaping the way students discover programs.
The panel also gets into the framework for sending better enrollment signals back to Google and Meta. That’s the piece most teams haven’t built yet, and it decides whether the platforms optimize toward students who enroll.
Conversion signal is the last real lever you control
Note: this post is specific to higher education, but the approach applies across industries.
Every year the platforms take another decision off your desk. Google's AI Max and Performance Max, Meta's Advantage+, and TikTok's Smart+ now handle audience, bidding, placement, and creative. The manual controls that used to separate a good media team from an average one keep shrinking.
You still choose the outcome those systems chase. That single choice now carries the weight a dozen manual controls used to share.
A form fill is the default signal, and a form fill is a hand raise. An algorithm told to produce hand raises goes and finds the people most inclined to raise their hands. Whether any of them turned into customers never reaches the platform, and the algorithm keeps optimizing on a proxy for months.
Google now recommends against this directly. Its guidance points lead-gen advertisers toward downstream goals and value-based bidding when leads differ in worth.
Ernie DeCoite, our SVP of Media, walks through what closing that gap takes, the timing problem that makes it harder than it sounds, and the results one institution saw after connecting enrollment outcomes back to its campaigns.
Google just made the AI label on Gemini content optional
This one is worth a deeper look.
Until last week, anything Gemini generated came stamped with a visible watermark. That stamp told anyone looking at the image, video, or clip of music that a machine made it. On Friday, Google gave users the option to remove it. Josh Woodward, who runs Google Labs and the Gemini app, framed the change as balancing creative control against safety.
As I mentioned above, the stamp was only ever the visible layer. Underneath it, Google embeds an invisible marker called SynthID plus metadata that records where a file came from. That layer stays in every file whether or not the watermark shows. Anyone can hand a file to Gemini or run it through Search and get an answer about whether AI made it.
So the change is cosmetic in one direction and meaningful in another. Creative teams can now ship AI-assisted work without a badge on it. The file still knows what it is, and anyone curious can ask.
That puts disclosure back on your desk. Decide what your team tells clients and audiences about AI-assisted work, and write it down, because Google stopped making that decision for you.
I mentioned Higgsfield in this newsletter a while back. They’ve since released Supercomputer, and it’s worth a second look. You describe the output you want in a chat, and it plans the job, picks the models, and renders it. The skills library covers faceless explainer video, UGC product video, brand kits, and full website builds.
For teams producing volume against fixed headcount, it cuts the time between an idea and a rough cut someone can actually react to.
Your AI visibility, scored.
When did you last audit what AI thinks of your brand?
Don't rely on an unreviewed algorithm to manage your brand’s reputation. Our scorecard pulls back the curtain on how AI platforms frame your brand sentiment and which sources they use to vet your authority.
Inside your custom report:
Visibility score and share of voice: The percentage of AI prompts your brand appears in, benchmarked against your top competitors across all major models.
Sentiment and positioning: Whether AI describes your brand favorably, neutrally, or negatively, and whether you're showing up as the primary recommendation or an afterthought.
Optimization roadmap: The specific content structure and authority signals you need to improve your score.
