What Sources Does ChatGPT Use to Give Recommendations?

Across the 28 days ending September 2, 2026, the Cheers market baseline collected 70,723 ChatGPT citations from home-services prompts run in 100 metro areas. 48.8% of them pointed at a contractor's own website. 44.7% pointed at a directory or review platform such as Angi, BBB, Yelp, or Expertise. Social and forum sources took 4.3%, media and vendor pages 2.0%, and google.com 0.2%.

So the practical answer is two sources, roughly split: your website and the directories that describe you. Everything else is rounding.

Last verified: September 2, 2026.

OpenAI does not publish a retrieval map, and its help article on searching the web is explicit that nothing is owed to any site:

ChatGPT ranks search results using multiple factors intended to help users find relevant, reliable information. Placement is not guaranteed.

The full panel cuts, including the same split for Gemini, Perplexity, and Google AI Overviews, are in AI visibility statistics, with the sampling rules in the methodology. To see which sources come back for your own brand and markets, run the Cheers AI Visibility Grader.

The short answer

Visible citations and published studies show local-business answers drawing from review platforms, directories, websites, and news articles. The internal source mix varies by product, mode, prompt, and date, and no public source provides a complete ranking or retrieval map.

For the broader strategic frame, read What is Generative Engine Optimization?. For Google's own AI Search advice, read How Local Businesses Can Show Up in Google AI Search.

If ChatGPT is already showing the wrong hours, phone, service, or branch, use the business-information correction workflow to trace the visible source and verify the fix by market.

The longer answer

Modern AI assistants can combine model knowledge with web search or retrieval features. The balance is product- and query-specific.

Model knowledge can contain information learned before the current query, but vendors do not publish a complete business-by-business inventory of that knowledge.

Search and retrieval features can fetch current web information when the product and query use them. A local prompt does not guarantee that every available source will be searched or cited.

Machine-readable business information includes structured data on business websites and platform-managed business records. Public documentation does not establish one universal weighting for those inputs.

OpenAI's help article, updated in late August 2026, describes the retrieval step in plain terms. ChatGPT rewrites your question into one or more targeted queries, sends those to search providers, and shares an approximate location taken from your IP address so that a "near me" question becomes a city-scoped query. OpenAI gives site owners exactly one concrete eligibility instruction: allow OAI-SearchBot to crawl the site, and make sure the host or CDN is not blocking OpenAI's published searchbot IP ranges. That is a robots and firewall check, not a content tactic.

Sources that matter for local businesses

Google Business Profile supplies business information to Google's local systems, and Google's own guidelines warn that duplicate profiles cause display problems on Maps and Search. Whether ChatGPT reads that record directly is unestablished. In May 2025, Search Engine Land's Damian Rollison described ChatGPT running a Bing search for the local query, taking 20 to 30 of the top results, and reordering them with its own logic, while not reading Bing Places profile data directly. Local Falcon's November 2025 teardown observed the same shape and named the platforms it saw most: Yelp, TripAdvisor, Better Business Bureau, MapQuest, and Yellow Pages. Both are dated observations of a system that changes, and OpenAI's current documentation names no provider except in the restaurant-reservation case, where availability may come from OpenTable, Resy, or Yelp.

The Cheers baseline gives the same picture from the citation side. The most cited platform domains across all 100 metros, all four engines, were google.com, expertise.com, angi.com, bbb.org, and reddit.com. Every one of those has a record about your business that you can read today.

Review platforms like Yelp, Facebook, TripAdvisor, and industry-specific sites can appear in search results and citations. They provide customer-authored context, but the products do not publish a universal sentiment or review-ranking formula.

Your website can be retrieved when it is crawlable and relevant to the query. Clear service descriptions and accurate schema help machines interpret the page, but they do not guarantee a citation or recommendation.

Directory listings can supply business facts and appear in retrieved results. Keep relevant listings accurate, especially phone numbers, addresses, hours, URLs, and services.

News and press coverage may appear when it is relevant to the query. Earned coverage should be factual and useful, not created to simulate authority.

For a local operator, conflicting public facts create a customer problem and a source-audit target. If Bing-visible pages, Yelp, BBB, Apple Maps, and the company website disagree on service area or phone number, correct the sources customers and crawlers can reach. Do not claim that consistency alone causes a recommendation.

For the plumbing-specific version of this source map, read How Do Plumbing Companies Get Recommended by ChatGPT?.

What this means for your strategy

You can't control which sources a particular AI query uses. But you can ensure your business is well-represented across all the major sources.

Earn reviews where customers use the platform. Keep relevant profiles accurate and follow each platform's solicitation policy. Do not manufacture a multi-platform review footprint for an assumed AI benefit.

Implement applicable schema markup. It should describe the visible page accurately and should not be treated as a recommendation guarantee.

Maintain business-information accuracy. Correct wrong or stale names, addresses, phone numbers, hours, URLs, and services. Harmless formatting differences are not the same as conflicting facts.

Keep information current. Update profiles and pages when hours, services, locations, contact paths, or policies change. Old data is not always worse than no data, but incorrect data can harm the customer journey.

Build a source stack. Your website should explain services and locations clearly. Your listings should match the same facts. Your reviews should show recent customer experience. Your schema should mark up the business in a machine-readable way. If you need the technical piece, see What Is JSON-LD?. If you need the review piece, see How Reviews Support AI Visibility for Local Businesses.

Sources

Every link below was opened and checked on September 2, 2026.

• Searching the web with ChatGPT. OpenAI Help Center, updated August 2026. Query rewriting, search partners, approximate location, reservation providers, ranking, and OAI-SearchBot eligibility.
• How does ChatGPT conduct local searches?. Search Engine Land, May 2025, by Damian Rollison. The Bing retrieval step and the reordering that follows.
• ChatGPT local search data sources. Local Falcon, November 2025. The review and directory platforms observed most often.
• Guidelines for representing your business on Google. Google Business Profile Help. One profile per business, and what duplicates do to Maps and Search.
• Optimizing your website for generative AI features on Google Search. Google Search Central, last updated July 2026.
• AI Visibility in 2025: How Gemini, ChatGPT, and Perplexity Cite Brands. Yext, October 2025. 6.8 million citations across 1.6 million responses.
• Cheers market baseline, 28 days ending September 2, 2026. 32 home-services prompts, 100 metro areas, four engines, 70,723 ChatGPT citations, aggregate only. Methodology.

Dylan Allen-Arnegård is the CEO of Cheers, the local search platform for service businesses.