SOURCE

How does ChatGPT conduct local searches?

Properties1
tagsseo, chatgpt, bing, local-seo, geo-aeo, bing-places
AUTHOR
Damian Rollison
PUBLISHED
2025-05-02
ORIGINAL
view original

5 min read

How does ChatGPT conduct local searches?

Author: Damian Rollison | Published: 2025-05-02 | Source: Search Engine Land · article

Summary

Damian Rollison’s Search Engine Land piece is an independent check of the industry claim that ChatGPT uses Bing for local search. Grade of that claim: partly true, importantly incomplete. ChatGPT’s local results improved after it began pulling Bing web results (fall before May 2025). It does not merely repackage Bing local packs, and it does not see Bing Places profile fields, partner reviews (Facebook/Yelp on Places), or Bing’s local map.

Method: side-by-side queries (dry cleaners in Red Hook, Brooklyn, open Sundays) plus a long interview with ChatGPT about its own process. ChatGPT returned a map-like pack of three. The #1 overlapped with Bing (The Cobble Hill Cleaners and Laundry). ChatGPT’s #2 (Bridgestone Cleaners) was Bing #14 under a longer name. ChatGPT’s #3 (Brooklyn Cleaner & Tailor) was not in Bing’s local listings at all. Overlap is small; ChatGPT is applying its own ranking/filter logic to Bing web results.

The retrieval sketch ChatGPT gave: run a Bing search, take the top 20–30 web results, scan for store sites, review aggregators, niche blogs/local guides, and directories; deep-dive 5–8 “most promising” sources; keep 3–5 that show an average star rating, are not paywalled, and are “verifiable and linkable.” “Promising” means relevant titles/snippets, trusted domains (retailer sites, Yelp, local blogs, BBB, Facebook, news), rich metadata in the snippet (hours, ratings, addresses), and unique content — skip duplicative aggregators and generic directories unless nothing else exists. It does not have direct Google Search or Maps access; Google ratings appear only when a Bing-indexed page cites them (e.g. a business site with Google review structured data). The pack map is OpenAI’s frontend plotting names/addresses ChatGPT supplies, not the model generating a map and not Bing Places.

Tactics ChatGPT recommended: detailed website with schema, hours, testimonials; claim Bing Places because it influences Bing web ranking, not because ChatGPT reads the Places record; Yelp/Facebook profiles; reviews; links from local magazines, “best of” posts, chamber directories; NAP consistency; unique pages that can be cited; monitor the first 20–30 Bing organic results. One-line summary: “Make your business easy for Bing to understand and easy for ChatGPT to cite.”

Caveat, from the author: ChatGPT helped write the article and can be wrong; the process description is kept because it matched the live results.

Key Claims

  • ChatGPT uses Bing web results, not Bing Places. It does not see Places profile details, partner reviews, or Bing’s local map. (independent test + ChatGPT self-report)
  • Not a Bing local-pack mirror. Side-by-side listings diverge after #1; ChatGPT can surface businesses Bing local does not list. (demonstrated)
  • Process: Bing query → 20–30 web results → 5–8 deep-dives → 3–5 citable sources with visible ratings. (ChatGPT self-report, correlated with results)
  • Prefers structured, rated, non-paywalled pages on trusted domains; skips generic directory dupes. (self-report)
  • No direct Google. Google ratings/Maps links only via Bing-indexed third-party citations. (self-report)
  • The map is UI, not model-generated and not Bing Places. (self-report after contradiction)
  • Bing Places still matters indirectly if it lifts the website in Bing web results ChatGPT actually reads. (self-report)
  • Visibility recipe: website + schema, reviews on Yelp/Facebook, local press/chambers, NAP consistency, unique cite-able content, watch Bing top 20–30. (self-report / author summary)

Notable quotes

“While it now pulls web content from Bing, it applies its own algorithm to organize results.”

Why citable: The accurate one-line grade of “ChatGPT uses Bing.” Web index yes; Bing local ranking no.

“Does not see the local profile data managed by businesses via the Bing Places for Business platform.”

Why citable: Direct falsification of “optimize Bing Places so ChatGPT reads your listing.” Places is an indirect Bing-web lever, not a ChatGPT database.

“Make your business easy for Bing to understand and easy for ChatGPT to cite.”

Why citable: Operational GEO line that matches retrieval (cite-able web pages) rather than training-memory implant.

Connections

Entities mentioned: ChatGPT, Google, Reddit · Bing Places (NEW: bing-places), Microsoft Bing (minor), Yelp (minor), OpenAI (minor) Concepts referenced: GEO / AEO (Getting Recommended by AI), Local SEO, Super Citations, Schema Markup (Structured Data), Review & Reputation Management, GBP–Website Alignment (Consistency Signals), 250 Authority Protocol

Contradictions / Tensions

  • Grades Caleb’s “ChatGPT uses Bing” / Bing-as-super-citation-for-voice-AI claim. Caleb Ulku groups Bing (for Business) with Apple Maps and voice AI (ChatGPT) as Super Citations that feed AI recommendations. Partly right, oversold. ChatGPT does retrieve via Bing’s web index, so Bing visibility can matter. It does not read the Bing Places profile, maps, or Facebook/Yelp reviews attached to Places. Claiming a verified Bing listing the way you claim a GBP — as a record the AI “sees” — is false for ChatGPT local answers. The useful remainder: get the website into Bing’s top 20–30 web results and onto cite-able review/press pages.
  • Vs. 250 Authority Protocol training-memory implant. This article describes real-time Bing retrieval + citation filters, which matches the defensible GEO lane on GEO / AEO (Getting Recommended by AI) (retrieval/consensus), not “250 pieces in the weights.”
  • Vs. Google-surface GEO (Google's Guide to Optimizing for Generative AI Features on Google Search). ChatGPT local ≠ AI Overviews. Google: people-first pages in Google’s index. ChatGPT local: Bing-web cite-ability, ratings, non-paywalled structured pages. Do not copy one playbook onto the other.
  • Method limit: process steps are ChatGPT talking about itself. Author flags this. Weight the SERP comparison (demonstrated) above the 20–30 / 5–8 / 3–5 pipeline (self-report).
  • Reddit appears as a cited source in the example pack; the filter text also says it skips unlinked forum fluff. Do not treat “ChatGPT loves Reddit” as unconditional.

Notes

Independent (non-Caleb) local-GEO source. Author works at SOCi; SEL is Semrush-owned; editorial disclaimer is in the extract. Extract filename is sel-chatgpt-local.txt.