Most businesses treat "AI search" as one channel. Citation studies and visible answers show that ChatGPT, Gemini, and Perplexity can cite different mixes of websites, directories, review platforms, and vertical sources.
Those mixes also change by query, category, location, and product version. The practical job is to inspect what each answer cites, not memorize a fixed ranking recipe.
If your team needs the operating workflow behind this source mix, use the industry AI source audit guide to capture cited URLs, classify source types, and assign the next fix by market.
Yext's October 2025 study analyzed 6.8 million citations across 1.6 million AI-generated responses. It reports the sources observed in that dataset; it does not expose every source or internal selection rule used by each product.
The local source map in one minute
For local businesses, the cited research suggests a useful audit order: inspect third-party and Bing-visible pages around ChatGPT answers, inspect owned-page citations around Gemini answers, inspect the visible vertical sources in Perplexity, and keep Google Search fundamentals healthy for Google AI Search. Treat these as starting hypotheses, not provider rules.
ChatGPT
The cited sample and separate local testing observed both owned pages and third-party sources around ChatGPT answers. Examples included Bing-indexed pages, Yelp, BBB, Foursquare, niche directories, local guides, and review profiles.
What you can maintain: crawlable owned pages, accurate Bing-visible listings, legitimate review profiles, and service pages specific enough to answer a buyer's question.
Gemini and Google AI Search
The cited Yext study found Gemini had the highest owned-website citation share among the engines in its sample. Google AI Search guidance also points back to normal Search foundations. Together, those findings make owned location and service pages worth auditing, but they do not establish a special weight for schema, reviews, or profile completeness.
What you can maintain: useful local pages, accurate Google Business Profile fields, LocalBusiness schema that matches visible content, and legitimate customer reviews.
Perplexity
Perplexity presents inline citations that make its visible source set easier to inspect. In the cited Yext sample, some query groups showed a larger share of niche and industry-specific sources. Operators should record the actual directories, review platforms, local proof pages, and owned content shown for their category.
What you can maintain: accurate profiles on relevant vertical sources, evidence-backed owned content, and a record of the source URLs each answer cites.
For a multi-location service brand, Cheers AI visibility tracking connects visible sources back to prompts, markets, competitors, and the next fix. The Sierra case study shows the operating model: review growth, market-level AI checks, and a public proof page. Whether a product cites that page remains a separate measurement question.
What Yext meant by brand-managed sources
The Yext study found that 86% of citations in its sample came from brand-managed sources. Yext's category included owned websites as well as directory listings and review profiles a brand can claim or maintain.
Brand-managed does not mean brand-owned or fully controlled. A business can correct its directory facts and respond to reviews, but it does not control the platform, customer contributions, or which source an AI product retrieves. The finding supports maintaining public business information; it does not establish a universal ranking formula.
The observed mix is a prompt for investigation, not a permanent engine-by-engine playbook.
Where ChatGPT gets its data
First Page Sage's live standalone-chatbot tracker currently ranks ChatGPT first, but the page's historical percentages can change without an archived snapshot. Its visible citations are worth recording because they show which pages supported a particular answer.
48.73% of ChatGPT citations in Yext's sample came from third-party sites. Yelp, TripAdvisor, MapQuest, and similar pages appeared in that category. Google properties accounted for 465,000 citations in the study, but that count should not be read as direct access to Google reviews.
The cited studies do not establish whether or how ChatGPT directly retrieves Google review data.
Search Engine Land's 2025 local search testing observed Bing-indexed web results in the answers it tested and did not observe Google as a live local source in that sample. That is evidence from a bounded test, not a complete map of ChatGPT's retrieval architecture. Visible sources can include business websites, directories, review platforms, and local guides.
For subjective queries like "best Italian restaurant near me" or "top-rated plumber in Austin," the pattern shifts even more. Directory sources spike to 46.3% of all ChatGPT citations on subjective queries, per the Yext data. These are exactly the queries where a potential customer is making a buying decision.
Sources observed in the cited research and local testing included:
- Yelp reviews and business pages
- TripAdvisor (for hospitality and food)
- MapQuest and Foursquare (for location data)
- BBB profiles
- Your website (especially pages with structured data)
- Bing-indexed web pages observed in the cited local testing
If your reputation program exists only on Google, it may not cover the third-party and Bing-indexed pages that appeared in this sample. Audit the sources shown for your own prompts before adding another platform.
Where Gemini gets its data
Gemini is Google's AI engine. It remains the second-largest standalone AI search tool in First Page Sage's May 2026 tracker, and Google is pushing it deeper into Search. At I/O 2026, Google said AI Mode had passed 1 billion monthly users and that AI Overviews can now flow into AI Mode conversations across desktop and mobile.
52.15% of Gemini citations in Yext's sample came from brand websites. That was the highest website share among the engines in that study.
Google's products can use Google's search and local data systems, but the exact retrieval and ranking mix is not public. The safe conclusion is narrower: useful, crawlable pages and accurate Business Profile information are important inputs to maintain.
The website share makes owned pages worth auditing. Detailed service pages, location pages, and accurate structured data can make a business easier to understand, but the study did not prove that any one format causes Gemini recommendations.
The Princeton GEO research (Aggarwal et al., KDD 2024) tested content interventions on a general generative-engine benchmark and reported gains of up to 40% in that environment. It was not a local-business study and did not test Gemini's local ranking system.
Sources to audit for Gemini:
- Your website (service pages, location pages, about pages)
- JSON-LD schema markup (LocalBusiness, Service, FAQPage)
- Google Business Profile data (hours, categories, attributes)
- Google Maps data (location accuracy, service areas)
- Structured content with statistics and citations
For Gemini, inspect whether useful owned pages appear in citations and whether those pages accurately answer the local query.
Where Perplexity gets its data
Perplexity takes a different approach than both ChatGPT and Gemini. It positions itself as an "answer engine" and cites sources inline, which means its citation patterns are more visible and more specific than the others.
The Yext study found a larger niche and industry-specific source share for some Perplexity query groups than for the other engines it tested. Zocdoc appeared prominently in healthcare and TripAdvisor in hospitality. The study does not establish one permanent source hierarchy for every home-services query.
For subjective queries, 24% of Perplexity's citations come from niche industry sources. That's a higher concentration on vertical-specific content than either ChatGPT or Gemini.
Perplexity visibly cites academic papers, industry reports, and specialized publications for some queries. The E-GEO paper from 2025 studied citation behavior in an e-commerce context; it should not be treated as a local-business ranking model.
The useful action is to inspect the cited pages in your own category and improve factual gaps where you have first-hand evidence.
Sources observed or worth checking for Perplexity:
- Industry-specific directories (Zocdoc, Angi, Houzz, Avvo, TripAdvisor)
- Specialized review platforms relevant to your vertical
- Content with inline citations and data points
- Academic and research sources
- Niche publications and industry blogs
If an accurate, relevant industry directory repeatedly appears in your market's answers, it belongs in the source audit.
Why one strategy doesn't fit all three
Say you run a dental practice.
For ChatGPT, check whether Yelp, industry directories, the business website, or other Bing-visible pages appear in the citations. Keep the relevant profiles accurate, but do not assume every directory belongs in the plan.
For Gemini, inspect whether owned service pages appear and whether they accurately state treatments, accepted insurance, service area, and other patient-facing facts. Use applicable schema to match the visible page, not as a ranking shortcut.
For Perplexity, inspect whether Zocdoc, Healthgrades, Vitals, or a state association page appears in the answer. The relevant source will depend on the query and market.
Same business, different source hypotheses to test depending on which product the patient uses.
The research on LLM product visibility (Kumar et al., 2024) provides useful evidence that page content can affect model outputs in a product benchmark. It did not test local service pages, so it should not be treated as proof of a local ranking tactic.
One source is not a complete visibility audit
We built the Cheers AI Visibility Grader to test how businesses actually show up across AI engines. Some of the results caught us off guard.
Chick-fil-A, one of the most recognizable restaurant brands in America, scored just 65/100 with only 2 out of 12 AI mentions across ChatGPT, Gemini, and Perplexity.
That single grader result does not prove why the brand appeared or failed to appear. It does show why operators should inspect prompt-level answers and sources instead of assuming Google visibility transfers unchanged to every AI product.
BrightLocal's 2025 consumer search behavior study found that 40% of consumers in its survey used generative AI when searching. Another BrightLocal survey found that 3% considered an AI platform their default for local searches. These are different measures, so use them as snapshots rather than one adoption trend line.
BrightLocal's 2026 LCRS data also found that 88% of AI users in its survey fact-checked results by verifying sources and checking reviews. That shows a generated answer may not end the research journey; it does not establish one universal funnel or prove that reviews close every AI-assisted purchase.
Minimum viable citation stack for a local service brand
If you run a local service business, do not start with every directory on the internet. Start with the sources that answer engines can reconcile.
- Owned website: crawlable home, service, location, proof, and contact pages with visible business facts
- Google source layer: accurate Google Business Profile fields, categories, services, hours, reviews, and photos
- Bing-visible source layer: Bing Places, crawlable website pages, and citation sources that appear in Bing-indexed results
- Review sources: Google plus the third-party review and directory platforms customers actually check in the category
- Vertical sources: Angi, Houzz, Zocdoc, TripAdvisor, Avvo, or similar category-specific sources when they fit the buyer journey
- Proof sources: case studies, photos, service examples, credentials, local press, partner pages, and customer-facing evidence that supports the same claims
The stack is only useful when the facts agree. A location page, Google profile, BBB profile, and Yelp listing that describe different hours, phone numbers, or services make the business harder to trust. A smaller source stack with cleaner facts is usually stronger than a bigger stack full of conflicts.
For the recurring workflow, use Citation Cleanup for Local and AI Search and How to Audit AI Search Visibility Across Locations together. One gives the source map. The other turns the source map into market-level work.
How to build a multi-engine AI visibility strategy
Use the Yext citation data and other research to form testable hypotheses, then verify them against your own prompts and markets.
Start by auditing your current state. Ask ChatGPT, Gemini, and Perplexity about businesses in your category and market. Save each response, then note where you appear, where you do not, and which sources are cited. This produces answer-level evidence that complements an SEO audit; it does not replace crawl, index, ranking, or conversion analysis.
Improve owned pages where the audit finds missing or inaccurate information. Add applicable LocalBusiness JSON-LD that matches the visible page. Build useful service pages with specifics such as pricing context, services, areas served, and response expectations. The Princeton GEO study reported gains from some content interventions in its benchmark, but it did not test local pages or establish a Gemini ranking formula.
Maintain legitimate review and directory profiles that appear in the answers you audit. Yelp, Foursquare, BBB, and industry directories may appear in ChatGPT citations, but the mix varies. Ask every eligible customer neutrally and never script services, keywords, ratings, or outcomes for them to include.
Claim relevant vertical directories that repeatedly appear in your category's results. Healthcare might include Zocdoc or Healthgrades; home services might include Angi or Houzz. Accuracy and buyer relevance matter more than hitting an arbitrary directory count.
Keep your name, address, phone number, website URL, hours, and services accurate across public sources. Harmless formatting differences such as "Street" versus "St." are not the same as a wrong phone number or old address. Prioritize contradictions that could misroute a customer or describe the wrong branch.
Finally, monitor and revisit regularly. AI models update their training data and retrieval sources on their own schedule. What works today might shift in six months. Run your AI visibility audit monthly and track which sources each engine cites for your category.
Measure the market without manufacturing urgency
BrightLocal has reported substantial use of AI in local discovery, while ChatGPT and Google AI Mode have reached large audiences. The surveys and platform metrics use different questions and denominators, so do not combine them into one growth rate.
The reason to monitor now is operational: teams need a baseline before they can tell whether a source correction, page update, or review program changed what the products show.
Visible source mixes can differ by product, prompt, market, and date. Record the sources shown for the answers you test instead of assuming a fixed provider rule or creating profiles everywhere. Turn that evidence into a recurring workflow with How to Audit AI Search Visibility Across Locations, then use Citation Cleanup for Local and AI Search to decide which third-party sources deserve cleanup first.
For tools that turn multi-engine source differences into local work, see Best AI Visibility Tools for Local Businesses. For the operating workflow behind provider, prompt, citation, and market tracking, read How to Audit AI Search Visibility Across Locations.
Check where you stand today with the free Cheers AI Visibility Score.
Sources
- Yext: AI Visibility in 2025 (6.8M Citation Study)
- Yext: AI Citations Press Release
- Princeton GEO Paper (KDD 2024): Generative Engine Optimization
- Harvard: Manipulating LLMs to Increase Product Visibility (2024)
- BrightLocal: Local Consumer Review Survey 2026 / AI Trust
- BrightLocal: Consumer Search Behavior (2025)
- E-GEO: Generative Engine Optimization in E-Commerce (2025)
- First Page Sage: Top Generative AI Chatbots (May 2026)
- Google: A New Era for AI Search (I/O 2026)
- Search Engine Land: How ChatGPT Conducts Local Searches
Dylan Allen-Arnegård is the CEO of Cheers, the local search platform for service businesses.