TP502 – Don’t Write Toward the Machine
Pew Research Center published the first clean count of how much of the internet is now written with AI. Roughly one page in ten shows significant signs of it. Filter to pages published since ChatGPT launched and it climbs past a third.
Chris Boyer and Reed Smith spend the first half of the episode on what that measurement actually captures. Pew names four markers that have climbed since 2023: em dashes, Oxford commas, a vocabulary that includes delve, testament, tapestry and enhance, and the construction where a sentence tells you a thing is not just one thing but another. Chris walks through the tells Pew did not count, which turn out to be the more useful ones for anybody responsible for a website. The paragraph that ends by restating itself. Every paragraph running the same length. Content so generic it could belong to any health system in the country, because the model has no lived experience to draw on and never will.
Reed pushes back in the direction the argument needs. Patients have been navigating vague human-written hospital pages for fifteen years. The model did not invent that problem. It removed the barrier holding the volume down, the way digital cameras ended the discipline of thirty-six exposures. Chris names where that leads: everything converging toward the middle, because averaging is how the models were built.
The second half is for people who own content. Google Search Central says generative AI is useful for research and for adding structure to original content, and that using it to generate many pages without adding value may violate the scaled content abuse policy. Google publishes no threshold. No word count, no acceptable percentage, no safe number of pages. Chris and Reed get specific about where AI earns its place in content production and where it costs more than it saves, then land on the trap waiting at the end of it, which is teams running detection scores and editing toward a machine instead of toward a patient.
Lacey Reichwald of AHA Media Group closes the episode with the finding that reframes the conversation. Content creation is the most popular use of AI in marketing departments and it is not where the efficiency lives. She has surveyed health system marketing teams for years, and her argument is that organizations keep backing a process into a tool rather than choosing a tool for a process. She also makes the case for putting the AI skeptics on the working group on purpose.
If your content team is measured on pages published, you have already told them which of these two halves matters.
MENTIONS FROM THE SHOW
- Healthcare’s Long Walk Toward the Patient, the free eBook marking 500 episodes: https://www.touchpointpodcastbook.com
- Pew Research Center, How Much of the Internet Is Written With AI?, August 20 2026. Roughly 490,000 English-language pages from Common Crawl going back to 2021, scored with Open Pangram: https://www.pewresearch.org/data-labs/2026/08/20/how-much-of-the-internet-is-written-with-ai/
- Google Search Central, Guidance on generative AI content, including the scaled content abuse spam policy introduced in the March 2024 spam update: https://developers.google.com/search/docs/fundamentals/using-gen-ai-content
- Google Search Central, Spam policies for Google web search, scaled content abuse definition: https://developers.google.com/search/docs/essentials/spam-policies
- Open Pangram, the open-weight detection model Pew used, from Pangram Labs: https://www.pangram.com/
- LinkedIn “Seems like AI slop” reporting button, launched end of July 2026. Chief Product Officer Hari Srinivasan reported more than one million uses in the first weeks and roughly 40 percent fewer views on flagged posts: https://techcrunch.com/2026/07/30/linkedin-adds-a-button-to-report-ai-generated-slop/
- Social Media Today, LinkedIn says 1M people have reported AI slop, August 20 2026: https://www.socialmediatoday.com/news/linkedin-says-1m-people-have-reported-ai-slop/828465/
- Harvard Business Review, AI-Generated “Workslop” Is Destroying Productivity, September 2025. BetterUp Labs with the Stanford Social Media Lab, 1,150 full-time US desk workers. BetterUp is a commercial platform and the figures are self-reported: https://hbr.org/2025/09/ai-generated-workslop-is-destroying-productivity
- Harvard Business Review, June 2026 follow-on from Matthias Holweg and Thomas Davenport extending the argument into organizational knowledge decay
- AHA Media Group: https://ahamediagroup.com/
- Lacey Reichwald on LinkedIn, where she publishes her tool tip newsletter: https://www.linkedin.com/in/laceyreichwald/
- TP500, “Healthcare’s Long Walk Toward the Patient,” where the hosts describe AI drafts of the book getting smoother and less true at the same time
- Reed Smith on LinkedIn: https://www.linkedin.com/in/reedtsmith/
- Chris Boyer on LinkedIn: https://www.linkedin.com/in/chrisboyer/
- Chris Boyer website: http://www.christopherboyer.com/
- Chris Boyer on BlueSky: https://bsky.app/profile/chrisboyer.bsky.social
- Reed Smith on BlueSky: https://bsky.app/profile/reedsmith.bsky.social
Recommendations from this episode:
- Chris: Effin’ Birds, the daily comic by Aaron Reynolds pairing Audubon-style bird illustrations with profane captions, plus the merchandise and playing card decks. Not family friendly: https://effinbirds.com/
- Reed: Reacher, season four, on Prime Video. Eight episodes, premiered August 12 2026 with the first three, new episodes weekly on Wednesdays through September 16
Find Us Online
Reed Smith
VP @jarrardinc, Advisor @MayoClinic Social Media Network and @SXSW Health & MedTech. @SocialHealthIns and @tpmnetwork founder
Chris Boyer
Principal, @chrisboyer LLC Using digital marketing, experience and UCD/CX to transform healthcare. Co-host of the touch point podcast and founder of the Touch Point Media network.
