Predicting OpenAI's ad strategy

Original Article ↗ Hacker News Discussion ↗

Ad-based LLMs as “inevitable” vs. counterexamples

  • Many assume ads will extend to all ChatGPT tiers, including expensive plans, because high-income users are the most valuable ad targets.
  • Others argue double-billing (subscriptions + ads) angers users, but skeptics point to cable TV, news sites, and streaming platforms that already do this with little long-term backlash.
  • Kagi and similar subscription-first products are cited as proof ad-free services can exist, though critics say such models don’t scale to Google/OpenAI size.

How ads are likely to be integrated

  • Strong expectation that LLM ads will be “native” and subtle: biased recommendations embedded in answers, not banner slots.
  • Proposed mechanisms:
    • Fine-tuning or per-advertiser “poisoned” models that always rate one brand best.
    • Emitting abstract tokens like <SODA> that an ad engine resolves in real time.
    • Prepending hidden advertiser text to prompts or running a second model to rewrite answers to align with the winning bid.
  • This is compared to product placement and “organic” word-of-mouth; many note users already casually advertise brands in conversation.

Trust, manipulation, and regulation worries

  • Core concern: once the model’s objectives include ad revenue, users can’t trust advice on purchases, health, or politics.
  • Some fear AI as an extremely powerful behavioral manipulation tool, especially for pharma, political influence, and “LLM-induced brand loyalty.”
  • Ideas floated: banning targeted ads, making ads explicitly opt-in, or even banning many ad formats entirely. Others argue enforcement is hard and regulatory capture is real, though past successes (smoking, lead, asbestos bans) are cited.

Economics: can ads really pay for this?

  • Several doubt the global ad market (already dominated by Google/Meta) can support AI’s massive compute bills and valuations; AI ad spend may simply cannibalize search and social budgets.
  • Some call the current AI boom a bubble; others argue it’s an “unbubble” where we’re still underestimating long-term revenue and productivity gains.
  • Debate over whether AI’s productivity guarantees profits: intense competition, local models, and commoditization could squeeze margins.

User responses and alternatives

  • Suggested defenses: disconnect from ad-driven platforms, return to books/vinyl, use local/open models, adblockers, or AI-based ad filters.
  • Others note most people won’t bother; mobile and DRM-like mechanisms may further limit blocking.
  • Some say they’ll quit any paid AI service the moment inline ads appear; others are largely unconcerned and view relevant ads as a fair trade.

Ads, AGI, and what it signals

  • One camp: turning to ads shows AGI / “machine god” profits aren’t near; if AGI were imminent they wouldn’t pivot to a “scummy” business model.
  • Another: powerful models already look like AGI for many tasks; ads are just bridge revenue while scaling continues.
  • Philosophical paradox raised: in a world where AGI destroys most jobs, who would have money to buy what the ads promote?