Proof of Person

If you tap a button in the Substack app, you can now see how much of a newsletter was written by AI. The detector is made by Pangram, and Substack CEO Chris Best said, “When content made by no one takes over parts of the internet that are supposed to be human, it pollutes the commons and makes it hard to discover and hear human voices.” Substack is the second major platform this summer to position itself as human-first. Reddit got there a month earlier with “People are the Best,” a brand campaign that positions real conversation against an internet filling up with AI-generated video and paid influencers.

Social media is facing a classic supply and demand problem. AI lets users pump out more content, so platforms are pulling levers like AI detection and limiting reach so they can maintain revenue from advertising. The reach changes are particularly visible on LinkedIn, where everyone and their mother is now a LinkedInfluencer.

Pangram’s own scan of just over a million posts this spring found that more than 40% of long-form LinkedIn posts were fully AI-generated, and that LinkedIn accounted for 62% of all the AI content flagged across five platforms. Over roughly the same stretch, Richard van der Blom’s algorithm tracking, cited in Forbes, has average post views down about half year over year.

Labeling may come with a trust cost. A University of Arizona team ran thirteen experiments on AI disclosure and found that admitting you used AI erodes trust regardless of the task or the audience, and that owning up to it yourself scores worse than getting caught.

By labeling, the platforms are betting that social pressure will bring back an equilibrium to the content market.

Signals This Week

1. Substack’s detector applies to everything over 100 characters

The scan runs on a reader’s demand against anything published after 4:30pm ET on July 21, and it reaches notes, replies, and comments, not just posts, per TechCrunch. Writers can run Pangram on a draft before publishing, dispute a flagged result, and attach an optional author’s note disclosing how they used AI. Pushback arrived within a day. One writer told Press Gazette the feature is “nothing but an imperfect purity test for writers,” and essayist Monica Herbet argued that “the presence of AI does not prove the absence of a human.” Pangram’s own July research puts Substack’s corpus at 21.9% AI-generated or AI-assisted.

2. A new federal bill would drop the requirement that a fake ad be damaging

Sen. Adam Schiff and Rep. Ro Khanna reintroduced the AI Ads Act on July 27. It extends federal fraud prohibitions to content generated wholly or partly by AI, covers anyone purporting to act for a candidate or committee, and strikes the existing requirement that a misrepresentation be “damaging” before it breaks the law. A companion bill would put disclaimers on paid influencer posts, aimed at the millions Tom Steyer’s campaign spent on unlabeled posts from news influencers.

The bill was reintroduced after Steve Hilton released his first campaign spot, “The People vs The Machine,” casting the California Republican gubernatorial candidate as a bulwark against a robot and caricatures of prominent Democrats. The spot was created by Charles Curran who created the viral AI videos in support of Spencer Pratt’s Los Angeles mayoral run. Schiff told Semafor that “California’s primary this year made a clear case that our current guardrails around campaign advertisements are not keeping up.”

3. Meta will certify that you are a person

Facebook Verified launched July 24, a checkmark that can be earned with a short video selfie matched against profile photos. It rolls out on Marketplace, Dating, Groups, and Profiles before it reaches Feed. Meta’s stated reason names the problem plainly: “AI makes it easier to generate content, profiles, and messages.”

4. OpenAI values the answer box at $100 billion

Ads have run inside the free version of ChatGPT since February, in seven countries, against roughly a billion monthly users. Business Insider reports the internal projections: $100 million in annual recurring revenue as of March, $2.4 billion for 2026, $100 billion by 2030. Google’s entire global ad business did $294 billion last year. EMARKETER puts the whole US chatbot ad market under $1 billion for 2026, and analyst Nate Elliott said of the 2030 target: “I don’t think there’s any chance they hit that goal.”

5. Five publishers are weighing whether to cut Google off

USA Today, Politico, Reuters, The Economist, and People are all evaluating whether to block Google’s crawler over AI Overviews losses, per the Wall Street Journal’s reporting. Semrush data for June 2025 to June 2026 shows USA Today’s organic Google traffic down almost half, CNN down about 25%, Politico down 23%, and Business Insider down more than 85%, while the Guardian and the BBC gained. USA Today CEO Mike Reed: “It’s time to take a stand and say enough is enough.” The same report says Reddit’s roughly $60-million-a-year data licensing deal with Google hit renewal turbulence; Reddit shares fell as much as 5.8%.

6. AP rewrote its AI rules for everyone who pitches it

The updated standards, published July 23, require AI-generated or manipulated material to be “clearly identified and presented in context when it is included in AP journalism,” with disclosure whenever generative AI “plays a material role in published content.” AP does not define “material,” leaving that call to its editors. The photography rule—no generative AI to create, alter, or enhance a news photo—stays.

7. Thirty-one states now regulate political deepfakes

Maryland, Minnesota, Oregon, and Texas ban deceptive synthetic political content outright, per MultiState’s tracker. Most of the other laws apply only inside a 90-day pre-election window. Colorado requires metadata naming the generation tool and the creation timestamp. Courts are moving the other direction: California’s and Hawaii’s laws are blocked on First Amendment grounds, with a federal judge finding California’s AB 2839 likely “impermissibly restrictive of protected political expression.”

8. Politically charged AI outrage bait is now a sales funnel

Instagram and TikTok accounts with hundreds of thousands of followers are posting AI-generated videos of teens being bullied for selling Christian apparel at Pride events with links to the merchandise attached, per Semafor.

9. China already has a market rate for your face

On platforms including ActID, New Claw, and ByteDance’s Hongguo, ordinary people license their likenesses into AI microdramas, commercials, and games for $15 to $700 per use, the platform keeping 10%, per Rest of World. ActID counts about 800 registered users, some 300 of whom have agreed to license their image. Over 95% of the 128,000 microdramas released in China in the first quarter used AI, and the Guangzhou Internet Court has heard roughly 700 face-theft cases in three years.

Each week our AI reads thousands of articles to bring you the most important stories in AI and influence. I find AI is extremely strong at separating signal from noise and summarizing the results. It’s still not a great writer.

If this edition resonated, please share The Influence Model.

Best,

Ben