Aggressively Imagining Funding Models for Generative Web Search

While diving into Zelch et al. (2023), a line in the abstract caught my attention:

How will generative AI pay for itself? Unless charging users for access, selling advertising is the only alternative. [emphasis added]

While I do not think we should completely reject, or refuse¹, generative search (also called conversational search or chat-first search), I think this present time of disruption is an opportunity to reject a narrowing of options. We should use this opportunistic to rethink what search could be.

How might we pay for generative AI in web search?

There are several potential models for funding the development and ongoing operations of generative AI systems for web search, including individually-owned systems, usage fees, public funding, philanthropic gifting, crowdsourced models, routing funding through other tools, and advertising-based models. Each approach has tradeoffs around factors like accessibility, incentives, transparency, and alignment with the public interest. The optimal funding approach remains an open question.

• Individually-owned (common pool of open source tooling and datasets)
• General-use supported by API fees
• Publicly-funded
• Gifted (perhaps in common) by companies who benefit from better search (or good will)
• Wikipedia model (grants and gifts from foundations, companies, individuals)
• Tool-routed funding for search (see a discussion of browsers in Berjon’s “Fixing Search” (2023))
• Advertising Models:
  • Banners
  • Generated separately from the search response
  • Incorporated into the search response itself

Is there a new design space for search?

People have proposed and tested many different search models and features over the years. This disruptive moment provides a chance to look back at what was tried & failed and what was proposed & ignored, and to look afresh. In addition to looking at alternative funding models and pursuing SERPs very different from the old ten-blue-links (and entangled with both), the disruption in search from generative AI has led people to newly imagine or reimagine other possibilities in the design of search.

1. Search Routing:
   • Consider dynamic search routing, allowing for queries to be directed to specialized tools. We can question if diverse query types—exploratory, navigational, transactional—should be processed by the same back-end mechanism.
2. Redistribution of Search Tasks:
   • The potential decentralization of search tasks into various software environments could lead to embedded search capabilities in tools ranging from text editors to messaging applications. This may distribute costs and possibly offer nuanced search experiences.
3. Individual Search:
   • Individual search tools, agents, and engines, rooted in open-source frameworks and decentralized crawls, could distribute certain costs across user and producer populations.
4. Public Voice in Search:
   • There is much opportunity for more voice in search interfaces and goals. See, for example, SearchRights.org.
5. Extending Search Models:
   • There’s potential in examining alternative models for search operations, including crawling, indexing, and filtering. Opportunities might also exist in the moderation, ranking, and data storage mechanisms within search.
6. Redefining Relevance:
   • A shift from traditional relevance metrics to more custom-designed SERPs might be considered. The design space includes SERPs oriented (even ad hoc) towards a range of specific values or objectives, suggesting a nuanced approach to personalization.