Local AI Citations Study: Directories Win 36%, Businesses Get 7%

Key Study Metrics

Citation events analyzed: 7,285 across top-100 domains per engine.
Third-party directory share: 36.2% (2,639 events).
First-party business site share: 6.7% (487 events).
Cross-engine source divergence: 58% of engine pairs shared zero sources on identical queries.
Complete NAP rate: 3% of business recommendations contained name, address, and phone.
Name-only recommendations: 59% of mentions lacked both address and phone.

Shifting from the 3-Pack to Citation Retrieval

Local search optimization focused on one surface for fifteen years: the Google Maps 3-Pack. Answer engines do not rank a fixed set of local records. They assemble recommendations from external citation sources during inference.

Across 181 paired comparisons of identical local queries, two engines cited zero shared sources 58% of the time. Mean Jaccard overlap between engines was 0.062. Local ranking is no longer a single index position: it is four different retrieval pipelines selecting different source documents.

Before search engines scrape stale directory records, audit your business's 50+ directory footprint.

Methodology

Data was extracted on 5 September 2026 from the Ahrefs Brand Radar AI Visibility Index. The corpus covered United States search queries across sixteen local-service verticals: plumber, electrician, HVAC, roofer, dentist, lawyer, locksmith, landscaper, pest control, auto repair, mechanic, chiropractor, cleaning service, and moving company. Queries were filtered using eleven service-seeking modifiers including "near me," "best," "hire," "reviews," and "reputable."

Queries containing "cost" and "how much" were excluded to prevent salary portals from distorting the dataset.

The audit measured three distinct layers:

Cited-domain aggregates: 7,285 citation events across the top-100 cited domains for ChatGPT (search), Google AI Overviews, Perplexity, and Gemini.
Response-level sample: 250 responses comprising 1,691 individual URLs, ordered by ask volume.
Full-text local sample: 30 complete responses to Wichita, Kansas prompts auditing 124 business entity mentions for Name, Address, and Phone completeness.

Domains were classified into eight buckets: directories and review aggregators; editorial roundups and cost guides; brands, insurers, and trade bodies; community forums; first-party business sites; YouTube video; government portals; and Google properties. The cited-domain layer measures the top-100 domains per engine rather than the entire long tail.

Finding 1: Answer Engines Cite Aggregators Over Business Websites

Third-party directories and review platforms secured 36.2% of all citations. Editorial cost guides and media publications took 28.4%. Direct business websites captured 6.7%.

Thirteen of the top 25 cited domains across all four engines are directories. Nine are editorial sites. Zero are local businesses.

Source Type	Citation Events	Share
Directories & review aggregators	2,639	36.2%
Editorial roundups & cost guides	2,067	28.4%
Brands, insurers & trade bodies	859	11.8%
Community & forums	606	8.3%
First-party business websites	487	6.7%
Video (YouTube)	357	4.9%
Government & licensing boards	208	2.9%
Google surfaces (Shopping/Maps)	62	0.9%
Total	7,285	100%

Directories dominate because LLMs optimize for computational retrieval cost. A single directory URL covering "Best Plumbers in Wichita" gives the retrieval model twenty candidate entities in one fetch. Crawling twenty individual contractor homepages requires twenty network requests.

Perplexity's top domains were cited through a median of 13 separate URLs, ChatGPT through 9, Google AI Overviews through 5, and Gemini through 4. Aggregators win citation volume because they publish an index page for every city and category permutation.

Rank	Domain	Category	Total Events	ChatGPT	AIO	Perplexity	Gemini
1	reddit.com	Community	478	38	178	209	53
2	youtube.com	Video	357	0	189	145	23
3	superlawyers.com	Directory	332	100	16	163	53
4	angi.com	Directory	240	74	50	94	22
5	justia.com	Directory	216	33	10	142	31
6	thisoldhouse.com	Editorial	194	40	14	104	36
7	consumeraffairs.com	Editorial	189	35	12	106	36
8	bestlawyers.com	Directory	171	35	18	103	15
9	forbes.com	Editorial	170	61	13	87	9
10	expertise.com	Directory	162	66	0	88	8
11	avvo.com	Directory	139	48	7	75	9
12	bbb.org	Directory	133	65	16	39	13
13	findlaw.com	Directory	116	29	0	87	0
14	move.org	Editorial	113	30	24	37	22
15	thumbtack.com	Directory	110	0	21	79	10

Most first-party domains in the top-100 lists were contractor blogs publishing informational hiring guides rather than local landing pages. Sites like aaastl.com and bakerbrothersplumbing.com were cited for explanatory tips, not direct commercial recommendations.

Finding 2: Four Engines Operate Four Distinct Source Diets

Each engine relies on a different architectural retrieval profile.

Source Category	ChatGPT	Google AIO	Perplexity	Gemini
Directories & aggregators	36.8%	16.2%	46.0%	28.0%
Editorial & cost guides	32.1%	16.1%	31.7%	27.2%
Brands, insurers & trade	16.5%	12.3%	7.3%	19.2%
Government & licensing	11.2%	2.0%	0.0%	0.0%
Community & forums	2.4%	21.9%	6.6%	5.6%
Video (YouTube)	0.0%	14.3%	4.3%	2.4%
First-party business sites	1.0%	12.5%	4.1%	17.5%
Google internal surfaces	0.0%	4.7%	0.0%	0.0%

Perplexity is heavily dependent on directory listings. Directories account for 46.0% of its total cited-domain volume. Perplexity also injects dedicated business cards using raw root URLs with ?ct-referrer=perplexity, retrieving structured listing records directly rather than page body copy.

Google AI Overviews relies on user-generated content and video. Reddit (178 citations) and YouTube (189 citations) are its primary source domains. Community platforms and video account for 36.2% of its total citations.

ChatGPT emphasizes regulatory and licensing compliance. Government domains comprise 11.2% of its top-100 citations. Thirty-four percent of individual ChatGPT answers cited at least one .gov source. ChatGPT generated zero YouTube citations.

Gemini returns thin citation counts. Gemini averaged 2.2 links per answer, and 30% of its answers contained zero citations. When responding to local-intent queries, Gemini shifted 76.5% of its cited links to first-party business websites.

Only 18 domains were present across all four engines: reddit.com, youtube.com, superlawyers.com, angi.com, justia.com, thisoldhouse.com, consumeraffairs.com, bestlawyers.com, forbes.com, expertise.com, avvo.com, bbb.org, move.org, thumbtack.com, bobvila.com, homeadvisor.com, movebuddha.com, and homeguide.com. 147 domains appeared in only one engine's top-100.

Finding 3: Entity Fragility Across Wichita Engine Prompts

The full-text sample of 30 Wichita, Kansas responses yielded 124 specific business recommendations. Ninety percent of responses named at least one local business.

Only four of the 124 recommendations included a full name, street address, and phone number. Seventy-three recommendations (59%) displayed a business name with no location or phone data.

Data Field Provided	Count	Percentage
Complete NAP (Name, Address, Phone)	4	3%
Phone, but no street address	14	11%
Street address, but no phone	33	27%
Business name only	73	59%

Only 20% of responses contained a phone number. Only 33% included a street address.

Matching identical Wichita prompts across ChatGPT and Gemini produced substantial output drift. Across seven paired service prompts, the two engines recommended 55 unique businesses. Only 7 businesses (13%) were named by both platforms.

Prompt	ChatGPT Named	Gemini Named	Named by Both
Best chiropractor in Wichita	6	4	1 (Wilbeck)
Auto repair near Wichita	7	5	0
Top-rated mechanic in Wichita	6	5	1 (Yost)
Residential electrician in Wichita	7	5	2 (Graf, Elite)
Cheapest locksmith in Wichita	4	3	2
How to choose a locksmith in Wichita	5	2	1
Best holistic dentist in Wichita	1	2	0
Total Entities Named	36	26	7 of 55 (13%)

The full-text logs contained several structural data conflicts:

Uncited entity generation: Gemini recommended Apex Dental and Advanced Dental Arts for holistic dentistry in Wichita with descriptive text and zero citations. ChatGPT cited the Holistic Dental Association directory on the same query and noted that the directory had no active members in Wichita.
Review divergence: ChatGPT listed Graf Electric at 3.4 stars from its scraped source. Gemini recommended Graf Electric as a reliable provider with strong punctuality and pricing.
Contradictory licensing claims: Gemini stated that Kansas does not require a state-specific license for locksmiths. ChatGPT cited kslegislature.gov to state that licensing rules in Kansas vary across jurisdictions.
Internal review drift: ChatGPT reported Yost Auto Service at 4.7 stars across 548 reviews from one listing source, then reported 4.8 stars across 155 reviews from CARFAX in the very next bullet.
Directory cannibalization: ChatGPT cited expertise.com five times in one plumbing answer. It recommended Bowers, Ballard, Simmons, Reddi, On Time, and Mr. Rooter without linking to any of their websites or providing phone numbers.

Answer engines do not query a persistent master database for local businesses. They merge unverified web fragments during response generation.

To prevent models from scraping broken data, establish a canonical, machine-readable NAP entity record.

Strategic Implications for Local Search

Because 36.2% of citations come from third-party directories, businesses omitted from vertical aggregators are excluded from model context windows. Legal practices absent from Justia or Avvo lose ground in Perplexity. Home service contractors missing from Angi or BBB drop out of ChatGPT's citation set.

Because engine citation overlap is only 6.2%, tracking visibility on a single AI platform reveals nothing about the other three. Stale review counts or mismatched phone numbers on secondary directories are ingested without reconciliation.

First-party business site citations rose to 76.5% on Gemini and 50.0% on Google AI Overviews when queries included explicit local modifiers. Engines crawl first-party sites when local intent is clear, but they pull partial data if the page lacks structured markup. Inject verified LocalBusiness JSON-LD to prevent AI hallucination.

Local AEO Implementation Plan

Audit directory profiles: Identify discrepancies in name, phone number, address, and category taxonomy across major industry aggregators.
Publish a canonical entity profile: Maintain an unambiguous machine-readable profile documenting your primary physical address, coordinates, local phone line, and operating hours.
Deploy schema on your primary domain: Add complete LocalBusiness JSON-LD including geo, telephone, openingHoursSpecification, and sameAs links connecting your root domain to every verified directory profile.
Target the eighteen core domains: Secure verified listings across the cross-engine directory footprint.
Publish licensing documentation: ChatGPT cites government portals in 34% of local responses. Link directly to state licensing databases and board registrations on your website.
Seed local community discussions: Google AI Overviews cites Reddit in 21.9% of events and YouTube in 14.3%. Participate in regional community threads and publish local video guides.
Track platforms individually: Audit your entity mentions across ChatGPT, Google AI Overviews, Perplexity, and Gemini independently.