Leaked deck reveals how OpenAI is pitching publisher partnerships
A leaked presentation on OpenAI’s “Preferred Publisher Program” outlines paid partnerships where news and media outlets license their content for training and get priority placement and enhanced branding inside ChatGPT. Commenters see this as a step toward covert, pay-to-play influence over AI answers—akin to ad-driven search results—raising concerns about hidden bias, disclosure requirements, copyright liability, and the erosion of user trust. Many argue this will accelerate the “enshittification” of AI assistants and strengthen the case for open-source or ad-averse models as an alternative.
Program purpose & data access
- Deck describes a “Preferred Publisher Program” (PPP) where OpenAI pays publishers “guaranteed” and “variable” value for access to archives and fresh content.
- Commenters note this is mainly about ongoing scraping/browsing for up‑to‑date answers, not initial model training, which many assume already used publishers’ back catalogs.
- Some see this as a step toward an AI search product, with PPP content getting richer link treatments and priority placement in answers.
Copyright, fair use & data ownership
- Strong disagreement over whether prior training on publisher content is “copyright theft” or fair use; posters stress the legal status is unresolved and will take years to settle.
- Others argue that even if big firms lose, enforcement against countless open‑source and hobby models is infeasible: “the cat is out of the bag.”
- Concern that PPP lets large publishers monetize and wall off data, while smaller sites’ content remains un‑ or under‑compensated.
Advertising, bias & “enshittification”
- Many interpret PPP as the start of pay‑to‑play influence over model outputs, akin to search ads or product placement.
- Fears that answers will be skewed toward paying partners and that such bias will be hidden in model weights and phrasing.
- Several note laws requiring ads to be clearly labeled; others worry OpenAI will comply minimally or rely on fine‑print EULAs.
- Some recall earlier statements that paid subscriptions meant “you’re not the product” and see this as a reversal.
User experience & trust
- Strong anxiety about undisclosed or semi‑disclosed ads inside conversational answers, especially in sensitive contexts (e.g., mental health).
- Comparisons to the web’s and streaming’s gradual ad creep; expectation that even paid tiers may eventually include ads.
- A minority argue that if ad‑like integration remains clearly marked links or separate “sponsored” blocks, users may accept it, as with search.
Business model, competition & open source
- Many see this as inevitable given high training/inference costs and pressure for revenue growth.
- Others warn that ad‑tainted outputs could push developers toward open‑source models (e.g., local LLMs) explicitly tuned to resist marketing influence.
- Some note publishers gain leverage and OpenAI gains a competitive data moat, potentially starving smaller AI competitors of high‑value content.