Which Data Sources Should You Care About For AI Search?

Tier Typical use Supply Proof standing What the proof says Reference 1 Net & search discovery Google Search Confirmed + present Grounding with Google Search connects Gemini to real-time internet content material, returning inline citations to supply URLs. Google — Gemini API docs, Grounding with Google Search 1 Bing Search Confirmed + present Microsoft paperwork Bing outcomes getting used to boost Copilot responses. Not re-verified on this go. Microsoft Bing 3 Widespread Crawl Confirmed historic GPT-3 used filtered Widespread Crawl as roughly 60% of its sampling combination; LLaMA 1 reported 67%. Widespread Crawl 3 Historic internet corpora (C4 and many others.) Confirmed historic C4 is a cleaned by-product of Widespread Crawl; LLaMA reported C4 at 15% of its pretraining combination. TensorFlow Datasets — C4 4 Net grounding providers Robust proof / doubtless Class inference protecting third-party grounding/retrieval intermediaries. No single canonical supply. — 1 Merchandise & buying Google Service provider Middle Confirmed + present Service provider feed information underpins Google’s buying surfaces. Retained on person instruction; Google Buying eliminated as it’s a floor, not a supply. Google Service provider Middle Assist 2 Service provider / retail feeds (OpenAI) Confirmed + present Retailers share a safe, repeatedly refreshed CSV/JSON feed of identifiers, descriptions, pricing, stock, media and fulfilment so ChatGPT can floor merchandise precisely. Refreshes accepted as typically as each quarter-hour. OpenAI Builders — Agentic Commerce, product feeds 4 Microsoft Service provider Middle Robust proof / doubtless Equal business feed infrastructure; inferred parallel to Google Service provider Middle fairly than individually evidenced. — 4 Market feeds Robust proof / doubtless Class inference. Shopify catalog information is already built-in into ChatGPT, which helps the sample. OpenAI Assist — Buying with ChatGPT Search 1 Native & locations Google Maps Confirmed + present Grounding with Google Maps is a documented instrument alongside Search grounding, giving fashions geospatial context. Google Cloud — Grounding API 1 Google Enterprise Profile Confirmed + present Enterprise profile information feeds Google’s native surfaces. Carried from the supply desk; not individually re-verified. — 1 Yelp Confirmed + present Yelp licenses evaluations, images and enterprise info to OpenAI for real-time native suggestions. Past grounding it additionally drives actions: ChatGPT customers can ebook a desk or be a part of a waitlist, and Request a Quote lets customers contact suppliers in-chat. Yelp’s 10-Q confirms it’s reside. Axios; Yelp weblog; Yelp 10-Q FY2026 4 OpenStreetMap Robust proof / doubtless Broadly used open geospatial corpus; inferred fairly than confirmed for any named mannequin. — 4 Foursquare Robust proof / doubtless Left in Tier 4 intentionally: the OpenAI deal is Yelp’s, and no equal proof exists for Foursquare. — 4 Tripadvisor Robust proof / doubtless Class inference for overview/journey information. No confirmed deal recognized on this go. — 1 Data & reference Wikipedia Confirmed + present Explicitly current in GPT-3’s disclosed combination and LLaMA (June-Aug 2022 dumps, 20 languages); additionally extensively used as a reside reference/RAG corpus. Wikimedia dumps 1 Wikimedia Confirmed + present Similar corpus household as Wikipedia. Licensing is unusually clear: principally CC BY-SA with attribution/share-alike obligations. Wikimedia dumps 4 Wikidata Robust proof / doubtless Structured entity layer; strongly implied by knowledge-graph use however not individually confirmed. — 1 Group / Q&A / social Reddit Confirmed + present The Google deal gave entry to the Reddit Information API for ‘real-time, structured, distinctive content material’, and permits Reddit content material to be displayed throughout Google merchandise — i.e. reside grounding, not solely coaching. Tom’s Information (Google/Reddit deal) 2 Reddit Confirmed + present Similar deal, coaching facet: Google might use Reddit posts to coach its AI fashions and enhance providers similar to Search; reported at roughly $60m/yr. NOTE: Reddit is reportedly weighing whether or not to resume — deal with as unstable. Fortune; Neowin/WSJ on renewal doubt 4 Social platforms Robust proof / doubtless Class inference protecting platform-wide social corpora. — 4 Boards / communities Robust proof / doubtless Class inference. Overlaps Reddit however generalised to non-Reddit boards. — 1 Information & writer content material Dwell writer pages Confirmed + present Reached at inference time through search grounding fairly than pretraining; retrieval choice and crawlability govern inclusion. Google — Grounding with Google Search 2 Licensed writer content material Confirmed + present OpenAI has a number of express licensing partnerships (FT, Axel Springer, AP, Information Corp). Phrases differ per associate on coaching vs grounding vs attribution. OpenAI — FT content material partnership 2 Writer partnerships Confirmed + present Axel Springer’s deal contains in any other case paywalled materials in solutions; AP licensed a part of its textual content archive. OpenAI — Axel Springer partnership 3 Historic information corpora Confirmed historic Archive materials absorbed in pretraining; distinct from reside licensed entry. — 2 Developer / technical GitHub Confirmed + present LLaMA used GitHub’s public BigQuery dataset, restricted to Apache/BSD/MIT tasks; The Pile individually contains GitHub. Public visibility is just not an open licence. GitHub 2 Stack Overflow Confirmed + present Named in licensing-deal mapping alongside Reddit and Shutterstock as a knowledge platform powering a number of consumers. LLM Pulse — AI content material licensing offers mapped 2 Technical docs Confirmed + present Vendor documentation corpora; extensively used however not tied to a single disclosed settlement. — 4 npm / PyPI registries Robust proof / doubtless WEAKEST ENTRY IN THE TABLE. Relabelled from ‘bundle registries’ to call examples. No disclosed settlement or documented retrieval use discovered — think about slicing. — 1 Journey & commerce actions Google Lodge Middle feeds Confirmed + present When Gemini or AI Mode present lodge choices with real-time costs, that information comes from the Google Resorts feed. In Aug 2026 Google added lodge reserving inside AI Mode accomplished with Google Pay, so that is grounding plus actions. TechCrunch — AI Mode journey replace 4 Reserving / associate feeds Robust proof / doubtless Reserving Holdings and IHG are reported as contributors in Google’s agentic reserving pilot, which helps the route however stops wanting a documented feed spec. InfosTourisme (IHG/Reserving pilot) 4 OTA / commerce sources Robust proof / doubtless Class inference. OTAs run their very own charge feeds into these surfaces. — 4 Reservation / stock APIs Robust proof / doubtless Class inference protecting reserving/stock endpoints uncovered to brokers. —