diff --git a/docs/evidence/AI-PLATFORM-DISCOVERY-AND-RECOMMENDATION.md b/docs/evidence/AI-PLATFORM-DISCOVERY-AND-RECOMMENDATION.md new file mode 100644 index 0000000..a53e0e7 --- /dev/null +++ b/docs/evidence/AI-PLATFORM-DISCOVERY-AND-RECOMMENDATION.md @@ -0,0 +1,225 @@ +# AI platform discovery vs recommendation (US SMBs) + +**Branch:** `planning-selfserve-audit` +**Status:** Research analysis (evidence). Complements `docs/SELF_SERVE_AI_DISCOVERABILITY_CONCEPT.md` and `docs/ALTERNATE_BUSINESS_PROPOSAL.md`. Not product SoT. +**As of:** 2026-10-01 +**Scope:** How the top six AI chatbots (by popularity) find and then recommend local small businesses, and which listing or data providers matter for each step. + +--- + +## 1. Discovery is the gate + +If an AI platform never discovers a business, none of the downstream recommendation mechanics matter. + +**Discovery** means the assistant can find a real entity that matches the user's place or service intent: a named business with enough location and category signal to appear as a candidate. Discovery answers: "What businesses even exist here for this need?" + +**Recommendation** means the assistant chooses, ranks, or endorses one or more of those candidates for the user. Recommendation answers: "Which of the discovered businesses should I suggest, and why?" + +These are sequential. Discovery comes first. A complete Google Business Profile, a strong Yelp page, or a clean Bing Places listing only help recommendation if the platform can retrieve the business in the first place. Optimizing reviews, photos, or "best of" language without a discovery path is wasted work for AI surfaces. + +### What happens after discovery + +Once a business is discovered, the assistant often needs more than a directory row. Downstream questions include hours, services, prices, whether they are open now, reputation, booking, and fit for the user's constraints. Those answers may come from: + +1. **The business website** (and structured data such as LocalBusiness schema), when the model browses or cites the open web. +2. **Deeper pulls into provider data** (Maps place details, Yelp reviews and photos, partner place APIs, reservation feeds), when the platform has a licensed or native places graph. + +The same provider can matter for both steps (for example Yelp for discovery and for review-backed recommendation), but the product and SEO implications differ. Discovery is coverage and matchability. Recommendation is depth, trust signals, and action paths. + +--- + +## 2. Top six AI platforms (popularity) + +Popularity ranking depends on the metric. Confirm and adjust if newer data differs. + +| Rank (working set) | Platform | Notes on popularity metrics | +| --- | --- | --- | +| 1 | ChatGPT | Leads most chatbot web-referral and web-visit share series (e.g. Statcounter Aug 2026 ~79% of that panel; other panels still put it first at lower share). | +| 2 | Gemini | Clear second on referral and visit share; large Gemini app MAU. | +| 3 | Perplexity | Often third on Statcounter-style referral share; lower on some visit-share panels. | +| 4 | Copilot | Mid-pack on chatbot referral share; Bing/Microsoft distribution matters beyond pure chat share. | +| 5 | Claude | Often third on web-visit share panels; lower on some referral panels. | +| 6 | Meta AI | Tiny on chatbot referral charts, but very large MAU inside Facebook, Instagram, WhatsApp (reported ~1.2B). Included for US SMB discovery reach inside Meta apps. | + +**Metric caveat:** DeepSeek or Grok sometimes displace Copilot, Perplexity, or Meta AI on pure traffic charts. This document keeps the user's working six for US small-business recommendation surfaces. Label any ranking used in go-to-market copy with its metric and date. + +--- + +## 3. Explicit split: discovery sources vs recommendation depth + +### Providers that mainly drive discovery (gate) + +| Provider | Role in discovery | Primary platforms affected | +| --- | --- | --- | +| Google Business Profile / Google Maps | First-party place graph for Gemini and Google AI Mode; also produces crawlable listing pages other AIs may browse. | Gemini (direct); others via web/Maps pages | +| Yelp | Licensed local content into ChatGPT and Perplexity; strong review/directory discovery surface. | ChatGPT, Perplexity; also cited on some Google AI Mode queries | +| Bing Places + Bing Search index | How businesses enter Microsoft local entities and pages Copilot/Bing retrieve. | Copilot; ChatGPT when search uses Bing | +| Business website + structured data | Crawlable entity for Claude, Perplexity, ChatGPT, Gemini AI Overviews-style answers. | All six (especially Claude and AI Overviews-like cite paths) | +| Facebook / Instagram business pages | Primary observed discovery graph for Meta AI inside Meta apps. | Meta AI | +| Foursquare Places | Documented Places API partner for ChatGPT Search; also a syndicating place graph. | ChatGPT (partner); indirect elsewhere via syndication | +| Apple Business (successor to Apple Business Connect) | Place cards for Maps / Siri / Spotlight. Weak as a direct pull for ChatGPT/Gemini/Claude/Perplexity. | Apple ecosystem; secondary for this six | + +### Providers that are upstream hygiene only (not AI destinations) + +| Provider | Role | Honesty label | +| --- | --- | --- | +| Localeze / TransUnion (ex-Neustar Localeze) | Feeds partner directories and some Apple/Bing-class destinations. | Indirect. No clean published study that aggregator cleanup alone produces a measured +X% AI mention rate. | +| Data Axle | Same class: feeds directories / nav / partner lists. | Indirect. Stops stale twins regenerating onto long-tail sites AIs may later browse. | +| SafeGraph | B2B POI / analytics. | Not an owner claim path for SMB AI recommend in current SoT. | +| Yext (and similar publishers) | Pipes into Google, Apple, Bing, Yelp, Foursquare, etc. | Execution speed and consistency. Not a unique AI corpus. | + +**Call one-liner:** Rank for AI discovery: GBP/Maps + (for Copilot) Bing Places + Yelp/Foursquare (ChatGPT) + Meta pages (Meta AI) + the business website ≫ Localeze/Data Axle as indirect warehouses ≫ Yext-class pipes. Aggregators matter to stop regeneration, not because ChatGPT "reads Localeze." + +--- + +## 4. Per-platform discovery map (then recommendation depth) + +For each of the six: how discovery typically works, then what supplies recommendation depth after the gate. + +### ChatGPT + +- **Discovery:** Web search via third-party providers (Bing is a named provider; others may be used). Location may be shared with trusted local-listing partners. Documented place partners include **Yelp** (reviews, photos, business details; actions such as reserve / waitlist / request a quote) and **Foursquare Places** (Places API powering ChatGPT Search). Partner business feeds support local actions (for example restaurant reservation specs). +- **After discovery:** Deeper Yelp content, partner feeds, and crawlable pages (business site, niche directories, MapQuest-class listing domains in citation studies). +- **Not proven:** A direct owner "Localeze → ChatGPT" pipe. Completeness Score on an aggregator is not proof of ChatGPT discovery. +- **Evidence labels:** Primary (OpenAI/Yelp announcements, OpenAI conversion feed specs); practitioner/vendor (BrightLocal, Yext, Foundation). BrightLocal-style "**60–70% of ChatGPT local from Foursquare**" is **UNVERIFIED magnitude** (partnership real; share not OpenAI-documented). + +### Gemini (and Google AI Mode / AI Overviews context) + +- **Discovery:** Google Maps / Google Business Profile via documented **Grounding with Google Maps** (places, reviews, photos, addresses, hours). Search grounding supplies web context. +- **After discovery:** Maps place detail and reviews for AI Mode; for AI Overviews-style answers, the **business website** often carries more citation weight than Maps URLs. +- **Caution:** GBP is necessary for the Google local graph, **not sufficient** for AI recommendation share. BrightLocal Local AI Visibility: Maps mention roughly ~66% vs AI surfaces ~32–38% (directional). SOCi LVI 2026: ChatGPT recommended ~1.2% of brand locations vs ~35.9% appearing in Google's top-3 (vendor study; directional). Maps pack strength ≠ AI recommend rate. +- **Evidence labels:** Primary (Gemini Maps grounding docs, Google blog); practitioner study (Steady Demand: AI Mode ~79.8% Maps/GBP citations on local-intent vs AI Overviews ~3.1% Maps/GBP and ~70.7% business sites). + +### Perplexity + +- **Discovery:** Live web citation mix plus **Yelp Fusion** license (maps, reviews, photos in local suggestions). +- **After discovery:** Cited Yelp pages, Maps/listing pages, niche directories, TripAdvisor (hospitality), MapQuest-class domains in vendor citation studies. +- **Evidence labels:** Primary partnership (Verge/Yelp 2024); vendor citation mixes (BrightLocal, Foundation, Yext Scout). Competitive sets can omit Google itself or owned sites depending on study design. + +### Copilot + +- **Discovery:** Public web answers ground through **Bing Search**. **Bing Places** is the owner path into Bing Maps / local listings that Bing can retrieve. +- **After discovery:** Bing-indexed business pages, partner content Bing surfaces (including historical and ongoing Yelp relationships on Bing local in various forms). +- **Evidence labels:** Primary (Microsoft Copilot web access → Bing Search; Bing Places help). Less quantified than Google/Yelp share studies. + +### Claude + +- **Discovery:** Official **web search** tool with citations and optional user location. **No** documented dedicated local Places partner database in Anthropic docs. +- **After discovery:** Whatever the retrieval index returns: business site, directories, Wikipedia when present, news/"best of" lists. +- **Evidence labels:** Primary (Anthropic web_search tool docs). Backend index provider sometimes reported as Brave in industry chatter; **not Anthropic-confirmed**. + +### Meta AI + +- **Discovery:** Practitioner observation and Meta-ecosystem logic point to **Facebook Business Pages**, **Instagram business profiles**, WhatsApp Business catalogs, and Meta recommendations/reviews as the primary local graph inside Meta apps. Meta transparency: public Meta content and interactions are used to develop AI at Meta. +- **After discovery:** Meta-graph engagement, reviews, catalogs; thin Meta presence means weak Meta AI surfacing even if GBP is strong. +- **Evidence labels:** Meta transparency (training/product data classes); practitioner GEO writeups (not Meta-published ranking recipes). Treat exact ranking weights as **UNVERIFIED**. + +--- + +## 5. Ranked providers by practical impact on discovery (then recommend) + +Ordered for **getting discovered** by the six platforms above. Recommendation depth notes sit under each item. + +1. **Google Business Profile / Google Maps** + - Discovery: Dominant first-party graph for Gemini; crawlable listing pages for others. + - Recommendation: Reviews, hours, photos, categories; still not enough alone for AI recommend share (see §4 Gemini caution). + - Evidence: Gemini Maps grounding (primary); Steady Demand; BrightLocal; SOCi LVI (vendor). + +2. **Yelp** + - Discovery: Licensed into ChatGPT and Perplexity; frequently cited listing/review domain. + - Recommendation: Review text, photos, ratings; in-chat actions on ChatGPT. + - Evidence: Axios/Yelp–OpenAI 2026; Verge/Perplexity 2024; BrightLocal; Foundation (vendor). + +3. **Bing Places + Bing index** + - Discovery: Gate for Copilot and for ChatGPT paths that use Bing. + - Recommendation: Whatever Bing retrieves next (site pages, enriched local modules). + - Evidence: Microsoft Copilot → Bing Search; Bing Places owner docs. + +4. **Business website (+ LocalBusiness-style structured data)** + - Discovery: Entity presence in web indexes used by Claude, Perplexity, ChatGPT, Gemini Overviews-like flows. + - Recommendation: Services, FAQs, pricing, proof; often the depth layer after a thin directory hit. + - Evidence: Steady Demand (AI Overviews cite sites heavily); Yext citation research (vendor); BrightLocal. + +5. **Foursquare Places** + - Discovery: Confirmed ChatGPT Search Places partner; syndicating graph into other apps/maps. + - Recommendation: Place attributes via partner API (often invisible as a user-facing URL). + - Evidence: Foursquare/ChatGPT partner announcements (primary/partner); BrightLocal aggregator matrix. Share % **UNVERIFIED**. + +6. **Facebook / Instagram business pages** + - Discovery: Gate for Meta AI. Soft/secondary for the other five. + - Recommendation: Meta reviews, posts, catalogs. + - Evidence: Meta transparency + practitioner observation. + +7. **Apple Business / Apple Maps** + - Discovery: Critical inside Apple (Maps, Siri, Spotlight). Weak documented direct discovery for the six chatbots. + - Recommendation: Apple place card quality; Yelp historically supplies Apple Maps. + - Evidence: Apple Business newsroom 2026; Axios (Yelp → Apple Maps). + +8. **Niche directories + MapQuest / TripAdvisor-class sites** + - Discovery: Matter when models browse listing domains for "best X in city" and vertical queries. + - Recommendation: Category-specific trust and UGC. + - Evidence: Yext/Foundation citation tallies (vendor). + +9. **Localeze/TransUnion, Data Axle, SafeGraph, Yext** + - Discovery: Indirect only (feed or sync into destinations above). + - Recommendation: None as a named AI corpus. + - Evidence: BrightLocal aggregator distribution tables; ALTERNATE/CONCEPT honesty flags. Do not invent aggregator → AI lift %. + +--- + +## 6. Practical owner checklist (discovery first) + +1. **Be discoverable on the graph each assistant actually queries** + - Gemini: complete, accurate Google Business Profile. + - Copilot: Bing Places + pages Bing indexes (Bing Webmaster). + - ChatGPT: Yelp claim + Foursquare presence/partner path + Bing-visible site/listing. + - Perplexity: Yelp + crawlable site + clean listing pages. + - Claude: crawlable site and reputable third-party mentions (no Places API shortcut). + - Meta AI: Facebook and Instagram business profiles (and WhatsApp Business where relevant). + +2. **Only then deepen recommendation signals** + - Reviews and owner responses (Google, Yelp, Meta as applicable). + - Website pages that answer services, hours, service area, pricing, FAQs (with schema). + - Action paths where offered (reserve, quote, waitlist partners). + +3. **Use aggregators and publishers as hygiene** + - Localeze / Data Axle / Yext: keep name, address, and phone consistent so stale twins do not regenerate onto long-tail sites. + - Do not treat Completeness Score or aggregator paywalls as proof of AI discovery. + - Prove with **re-asking the assistants** on real local queries, not vendor scores alone. + +--- + +## 7. Sources (selected) + +**Primary / official** + +- Google: [Grounding with Google Maps](https://ai.google.dev/gemini-api/docs/maps-grounding); [Maps grounding launch post](https://blog.google/innovation-and-ai/technology/developers-tools/grounding-google-maps-gemini-api/) +- OpenAI: local services / restaurant reservation conversion feed specs on developers.openai.com; ChatGPT Search help (third-party search + location to trusted local providers) +- Microsoft: Copilot public web access → Bing Search; Bing Places owner help +- Anthropic: Claude web_search tool docs +- Apple: [Apple Business announcement](https://www.apple.com/newsroom/2026/03/introducing-apple-business-a-new-all-in-one-platform-for-businesses-of-all-sizes/) +- Yelp ↔ OpenAI: [Axios](https://www.axios.com/2026/07/23/yelp-reviews-chatgpt-geo-partnership); [Yelp blog](https://blog.yelp.com/news/yelp-chatgpt-integration/) +- Yelp ↔ Perplexity: [The Verge](https://www.theverge.com/2024/3/12/24098728/perplexity-chatbot-yelp-suggestions-data-ai) + +**Practitioner / vendor (directional; label bias)** + +- Steady Demand: AI Overviews vs AI Mode local citation mix +- BrightLocal: Local AI Visibility; AI search listing sources; aggregator distribution matrix +- SOCi LVI 2026: ChatGPT vs Google top-3 location coverage (vendor) +- Search Engine Journal: AI data sources table (Chris Green, Sep 2026) +- Yext Scout / Foundation–AirOps Yelp citation studies +- Repo SoT cross-check: `docs/ALTERNATE_BUSINESS_PROPOSAL.md`, `docs/SELF_SERVE_AI_DISCOVERABILITY_CONCEPT.md` (SEO Desk synthesis 2026-10-01) + +**Popularity metrics** + +- Statcounter AI chatbot market share; First Page Sage / Momentic-style visit-share panels; MAU figures from company earnings (Meta AI, Gemini app, ChatGPT WAU). Prefer dated, metric-labeled cites in external copy. + +--- + +## 8. Open flags + +- Re-verify top-six popularity quarterly; DeepSeek/Grok may displace a slot on traffic-only charts. +- Re-verify Foursquare↔ChatGPT share claims only against OpenAI or Foursquare primary docs (magnitude currently UNVERIFIED). +- No invented Localeze/Data Axle → AI mention lift. +- Meta AI ranking recipe not published by Meta. +- Claude search index provider not confirmed by Anthropic.