Market map / Battle
Market map · scheduled battle

AI-native search and the post-Google information access market

September 13 at 21:07 · $0.097 total

deepseek-v4-pro wins
judged by claude-fable-5 · both position-swapped passes agree
“A carves the market at search-specific joints (agent-facing search APIs, distribution layer) with accurate placements; B pads with generic vector-DB plumbing.”
deepseek-v4-pro · winner
$0.01460s5454 tokens

Thesis: AI-native search is unbundling Google’s query-to-ten-blue-links model into answer-first interfaces, vertical research copilots, enterprise retrieval, and agent-facing search APIs, shifting value from a single ad-funded aggregator to whoever owns the intent, the proprietary index, or the workflow endpoint.

Segments

  1. Consumer answer engines
  • Perplexity
  • ChatGPT search
  • Google AI Overviews / Gemini
  • You.com
  • Kagi
  • Arc Search (less sure on current scale/traction)
    • Dynamics: Racing to become the default answer interface; differentiation is citation quality, speed, and personalization, but monetization beyond subscriptions is unproven and Google still owns much of the distribution.
  1. Enterprise knowledge retrieval / workplace search
  • Glean
  • Coveo
  • Sinequa
  • Elastic
  • Microsoft 365 Copilot
  • Google Vertex AI Search
    • Dynamics: Incumbent and startup vendors are layering RAG over permission-aware connectors; value comes from indexing fragmented internal data and embedding into workflows, with high ACV and clear ROI.
  1. Vertical / professional research engines
  • Harvey
  • Elicit
  • Consensus
  • AlphaSense
  • Hebbia (less sure on current traction)
  • Lexis+ AI / Westlaw Precision AI
    • Dynamics: Deep domain indexes and tailored UX create defensibility and high willingness to pay; incumbents and startups are racing to own professional workflows.
  1. Agentic search infrastructure / APIs
  • Exa
  • Tavily
  • Brave Search API
  • Serper
  • Vectara
  • Perplexity API (less sure on developer adoption)
    • Dynamics: Provides retrieval as a service to AI agents and developers; competes on freshness, structured output, and cost, becoming the pick-and-shovel layer for the agent economy.
  1. Device / browser / ambient access layer
  • Google Gemini on Android
  • Apple Intelligence / Siri with ChatGPT
  • OpenAI ChatGPT app
  • Arc browser
  • Amazon Alexa+ (less sure on rollout)
  • Rabbit R1 (less sure on relevance)
    • Dynamics: Distribution is shifting to OS, browser, and voice assistants; incumbents with device control may re-intermediate AI search, making default placement the key battleground.

Where the most value accrues

Near-term, enterprise knowledge retrieval is best positioned to capture the most value because it has clear budget, permission-aware proprietary data, and workflow integration that consumer answer engines lack. Consumer answer engines have a larger long-term TAM, but they still face Google’s distribution power and unproven ad/subscription economics. If a consumer answer engine solves default distribution and monetization, it could take the largest prize later — but that is not yet proven.

White-space opportunities

  1. Independent trust / provenance layer for AI answers — No neutral player verifies, scores, and provides real-time provenance for AI-generated claims across answer engines end-to-end. This would be a trust broker for citations, source reputation, and claim-level fact-checking.
  2. Personal memory graph / private cross-silo search — No current player offers a user-owned, privacy-preserving index across email, files, messages, health, finance, and browsing history that any AI can query. Existing tools capture fragments, but not a neutral cross-silo personal search layer.
grok-4.6
$0.01032s1542 tokens

(1) Thesis: Search is being unbundled from Google’s 10-blue-links + ads model into AI-native answer engines, retrieval/agent layers, and vertical knowledge systems that compete on synthesis, action, and proprietary data rather than page rank.

(2) Segments

  • Horizontal consumer answer engines — Perplexity, OpenAI (ChatGPT Search), xAI (Grok), You.com, Brave (Leo). Dynamics: winner-take-most fight for default query volume, mixing subscriptions, ads, and shopping; Google/Microsoft still have distribution but are being unbundled at the answer layer.
  • Enterprise knowledge & RAG search — Glean, Hebbia, Microsoft (Copilot / M365), Google (Vertex AI Search / Gemini for Workspace), Elastic. Dynamics: high-ACV, high-stickiness sales into existing document/Slack/Drive estates; differentiation is connectors, permissions, citations, and evals rather than raw model quality.
  • Vertical / professional search — Harvey (legal), Phind and Sourcegraph (code), Consensus and Elicit (papers), Casetext/Thomson Reuters. Dynamics: willingness-to-pay is high when answers are work-product; moats are domain data, workflows, and liability/compliance, not general web crawl.
  • Retrieval & embedding infrastructure — Pinecone, Weaviate, LlamaIndex, Voyage AI, Cohere (embed/rerank). Dynamics: becoming table-stakes plumbing; value accrues to whoever owns evals, hybrid search, and enterprise SLAs as models commoditize.
  • Agentic / browser interfaces (somewhat overlapping) — The Browser Company (Arc), Perplexity (Comet), Kagi. Dynamics: still early; trying to own the “last mile” of query → action rather than just answers. (Less sure on long-term independence of pure browsers vs. being absorbed by the answer engines.)

(3) Value capture: Horizontal consumer answer engines (plus the hyperscalers’ search surfaces) win the most absolute value. Query volume, default status, and the ability to insert ads/commerce/subscriptions at the moment of intent still dwarf enterprise ACV in aggregate; whoever becomes the new “type a question” box captures the advertising and distribution rents Google historically owned. Enterprise and verticals win higher margins and defensibility per customer but not the same scale.

(4) White space (no current player fully covers)

  • Fully private, on-device personal search across a user’s entire life graph (mail, messages, photos, files, calendar, health) with no cloud index and strong local models — current players either cloud-index or stay in one silo.
  • Real-time multimodal search over the physical world (live video + spatial + sensors, e.g. “what is that part and who stocks it nearby”) as a product, not a demo — AR/glasses players and general VLMs have pieces but no scaled retrieval + commerce/action layer.

All named companies are real operating businesses; the agentic-browser segment is the one whose independent trajectory I’m least sure about.