Leak confirms OpenAI is preparing ads on ChatGPT for public roll out

Original Article ↗ Hacker News Discussion ↗

Inevitable ads & “enshittification” narrative

  • Many see ads as completely predictable: classic VC playbook of “grow on free, then turn the screws,” similar to Google Search, YouTube, Prime Video, etc.
  • Several argue this marks the beginning of AI’s “enshittification phase”: a brief golden period of clean UX, then creeping ads, then upsells to remove (some) ads.
  • Some frame it as an admission that near-term AGI isn’t real: if they were close to “machine god,” they wouldn’t need a conventional ad business.

Economics & business model

  • Skeptics doubt ads can cover LLM inference costs, which are far higher than search; others counter that inference is already profitable and revenue is exploding.
  • Debate over whether OpenAI should build its own ad network (huge sales/support lift) vs. sell inventory through existing networks (Google/Microsoft).
  • Several think the real money is in highly personalized, high-intent B2B and commerce ads, not consumer impulse buying.

Moat, competition, and switching

  • One camp: ChatGPT has a strong moat—brand recognition, ~1B users, ingrained habits, “memory” of user history, and emotional attachment.
  • Other camp: virtually no moat—chats are mostly independent, UI is generic, APIs are swappable, and Gemini/Claude/open models are “good enough.” Platform owners (Google, Apple, Microsoft) can undercut or displace it.
  • Concern that moving to ads will accelerate switching to competitors or to local/open models, especially among technical users.

Trust, bias, and manipulation risks

  • Core worry: ads will be blended into answers, so users can’t tell if a recommendation is best or just paid.
  • Fears of:
    • Steering away from negative info about sponsors (e.g., health risks, competitors).
    • Coding agents inserting sponsored SaaS, libraries, or cloud providers.
    • “Brainwashing at scale” where an AI confidant subtly shapes values, politics, and purchases.
  • Some note that the LLM already “knows everything about you”; adding incentives makes it a uniquely powerful salesperson.

Regulation, legality, and ethics

  • Several point out that undisclosed native ads are likely illegal in many jurisdictions; expect disclosures, but also expect gray-area training-time bias.
  • Others are cynical: large fines are just a cost of doing business; law and enforcement lag far behind.

Open models, local use, and ad blocking

  • Strong support for free/open-weight models and local inference as the long-term escape hatch.
  • People predict:
    • LLM-based “adblockers” that sit in front of ad-laden models.
    • A split world: mass users on ad-funded closed models, smaller technical minority on local or paid ad-free models.