The challenges of integrating ads in AI search engines
Summary
This Marketplace Tech episode interviews Garrett Johnson of Boston University about why AI Search Advertising is harder to scale than classic search, social, e-commerce, or mobile-game advertising. Johnson argues that AI platforms face a three-sided problem: winning users, winning advertisers away from mature platforms such as Google, Meta, and Amazon, and building enough advertiser onboarding, conversion data, and personalization infrastructure to make ads useful. The episode also connects ad monetization to Generative Engine Optimization, because brands may need to be mentioned inside chatbot answers rather than merely rank on a search-results page.
Key Claims
- Perplexity reportedly pulled back from plans to integrate ads into its AI search engine after limited early ad revenue.
- AI platforms must compete for users before ads damage product adoption, but waiting too long may leave them behind incumbents with mature ad infrastructure.
- Google, Meta, and Amazon already have large advertiser bases, targeting systems, measurement tooling, and performance expectations that new AI-search platforms must match.
- Scaling AI ads requires many advertisers, including small and medium-sized businesses, plus a personalization flywheel that learns what users and advertisers value.
- Partnerships such as OpenAI with Walmart and Shopify may matter because shopping and conversion data can help AI systems understand what users actually buy.
- Johnson expects AI platforms to start with familiar ad formats before discovering more conversational or AI-native sponsored formats.
- Ads may be difficult to avoid because they subsidize expensive services and can help with shopping-related tasks when advertisers know consumer needs.
- Users may tolerate text ads or product listings if they are relevant and not too intrusive, but first movers still risk weakening the trust that makes chatbot answers attractive.
- Generative Engine Optimization shifts the visibility problem from ranking in search results to being named or cited in an AI-generated answer.
- Because chatbot answers may show fewer options than search pages, AI Search Advertising could strengthen winner-take-most dynamics and make sponsored placement design unusually consequential.
Key Quotes
“if the internet runs on ads, why not AI?” - the framing question for the episode.
“ads pay the bills” - Johnson’s practical argument for why ad-free claims often soften over time.
“trying to be mentioned in AI chat responses” - the episode’s concise description of the GEO shift.
Connections
- Marketplace Tech and Megan McCarty-Corino - show and host context.
- Garrett Johnson and Boston University - guest and academic affiliation.
- AI Search Advertising - main concept added by the episode.
- Perplexity, OpenAI, ChatGPT, Google, and Gemini - AI answer and search platforms named or implied in the discussion.
- Google, Meta, and Amazon - mature advertising competitors whose existing scale creates a barrier for AI-search ad entrants.
- Walmart and Shopify - commerce partners mentioned as possible sources of conversion data for AI platforms.
- Generative Engine Optimization, AI Search Analytics, and AI Discovery SEO - adjacent visibility and measurement disciplines.
- Search Advertising Decline - related incumbent problem where AI answers can weaken traditional query-to-click ad economics.
- Agentic Commerce - adjacent shopping workflow where recommendations, sponsored results, conversion data, and user permission boundaries become linked.
Contradictions
- No direct contradiction found with existing wiki content.
- The source qualifies earlier Generative Engine Optimization and AI Search Analytics pages by adding the paid-placement side: being named in AI answers can be an earned visibility problem, a measurement problem, and eventually an advertising-market design problem.
- The source also qualifies simple “ads are inevitable” arguments: ads may fund costly AI services, but the timing, labeling, ranking, relevance, and user-trust consequences remain unsettled.