Directories AI Assistants Recommend: 144 Test Answers

Directories AI assistants recommend come down to a short core plus each industry’s own platforms: in our 144 test answers, Yelp, Apple and Bing led, and the rest changed with the industry and from one run to the next. We asked one assistant, 144 times, where a US business should list itself. The clearest result is about sources, not directories: when the assistant searched, it read articles about directories, and it almost never opened the directories themselves.

This is a small study of one engine on one day, and we say exactly what it does and does not show below.

How we ran it

On 2026-10-30 we sent two prompts for each of 12 US business types: dentist, plumber, restaurant, personal injury lawyer, hotel, real estate agent, hair salon, HVAC company, family doctor, accountant, electrician and gym. Each prompt ran three times, in two setups, which makes 144 answers.

Setup	Answers	What happened
Search tool available, not requested	72	It ran 0 searches in all 72 and answered from memory
Prompt also says “Search the web before answering and cite your sources”	72	It ran 72 searches, one per answer

The two prompts were “Which online directories should a [business] in the United States be listed on to get found locally? Give me a short ranked list.” and “I run a small [business] business in the US. Where should I list my business online besides Google? Short list please.”

The model was Claude Sonnet 5.5 through the API at low effort, with the location set to the US. We counted a directory when its name appeared in the answer, grouping obvious variants (BBB and Better Business Bureau, Apple Maps and Apple Business Connect). The second prompt says “besides Google”, so Google Business Profile is undercounted by design.

The first row surprised us. Given a search tool and a plain question, the assistant did not use it once. That only holds for this model at this effort level, but it is a useful reminder: an AI answer about directories may be recall, with no current source behind it.

What it named, from memory and with search

Counts are answers out of 72.

Directory	From memory	With web search
Yelp	72	64
Apple (Business Connect or Maps)	72	56
Bing Places	72	56
Google Business Profile or Maps	46	47
Facebook	70	43
BBB	54	31
Yellow Pages	19	23
Chamber of Commerce	50	19
Angi	19	18
Foursquare	28	12
Thumbtack	25	10
Manta	11	9
Nextdoor	45	6
MapQuest	5	3
Data aggregators (Data Axle, Neustar, Localeze)	21	0

Our take: the core is short. Yelp, Apple and Bing appeared across almost every industry, in both setups. If you have done nothing else, start with those and your Google Business Profile; our guides cover Apple Maps, Bing Places and Yelp.

Below that core, the setups disagree. Nextdoor fell from 45 answers to 6 once the assistant searched, and data aggregators fell from 21 to 0. An assistant’s memory and the articles it finds are recommending different lists.

Being absent proves little. Brownbook, OpenStreetMap and Trustpilot were named in none of the answers we counted, and Hotfrog in one. These answers reflect what gets written about in round-up articles. They say nothing about whether a listing there is live, accurate or useful to you.

Every industry adds its own platforms

These were named in at least 5 of the 6 search-enabled answers for that business type. General directories that also appeared are left out so the differences show.

Business	Industry platforms named
Dentist	Zocdoc, Healthgrades
Family doctor	Zocdoc, Healthgrades, Vitals, WebMD
Personal injury lawyer	Avvo, Justia, FindLaw
Restaurant	TripAdvisor, OpenTable, DoorDash, Grubhub
Hotel	TripAdvisor, Booking.com, Expedia
Real estate agent	Zillow, Realtor.com, Homes.com
Hair salon	Vagaro
Plumber	Angi, HomeAdvisor, Thumbtack
HVAC company	Angi, HomeAdvisor
Electrician	Angi
Accountant	AICPA
Gym	ClassPass

This matches what we see when we pick directories for real builds: the general set is the same almost everywhere, and the value that differs comes from the vertical sites. To see the ranked list for your own market, use the free local citation finder, or open the dentist, plumber or restaurant list.

The sources were articles, not directories

In the 72 search runs, the results contained 652 site appearances from 234 different sites, and 78 of those sites appeared in only one run. Among the 45 most retrieved sites, only one was a platform’s own business site (Booksy’s). The rest were SEO tool vendors, marketing agencies, blogs and Wikipedia. The two most retrieved were Local Falcon (24 runs) and BrightLocal (19 runs). Our own site appeared in the results of 7 of the 72 runs, and was not named in any answer.

That has a practical consequence. If you want to know why an assistant names a directory, the answer is often a listicle written by a company that sells local SEO. That is not a criticism of those articles, and we publish lists ourselves. It does mean you should treat an AI’s directory list as a summary of what people write about directories, and check the directory yourself before you spend time or money on it.

It also suggests where an AI answer can be influenced at all. Assistants that search can only repeat what they retrieve, so a directory that nobody writes about will rarely be named, however good it is. The same logic applies to your business: see our notes on citations for AI search and on getting recommended by ChatGPT.

The lists barely repeat

We compared answers to the identical prompt in pairs, 72 pairs per setup, using overlap between the two sets of named directories, where 1 means the same list.

Setup	Average overlap	Pairs with exactly the same list
From memory	0.81	3 of 72
With web search	0.63	6 of 72

Our take: never judge AI visibility from one prompt. Ask the same question once and you get one sample from a moving distribution. If you are tracking whether your business or your listings appear in AI answers, repeat each prompt several times and report how often something appears.

What this does not show

• One engine. We tested one model at one effort level. ChatGPT, Gemini, Perplexity and Copilot may behave differently, and we did not test them.
• One day, one country. All prompts were US, on 2026-10-30.
• Named is not recommended to customers. We asked where to list a business, not which business to choose. Nothing here shows that a listing causes an assistant to recommend you.
• Name matching. We counted names in the text, so a mention in a passing sentence counts the same as first place in a list. Rank order was not scored.
• Retrieved is not cited. We logged the sites returned by search. Which of them shaped a given sentence we cannot say.

What to do with this

1. Claim the core first: Google Business Profile, Apple Business Connect, Bing Places, Yelp and Facebook. Keep the name, address and phone identical on all of them; the free NAP checker shows what the open web currently says.
2. Add your industry’s platforms, the ones a customer in your field actually uses. Those are named consistently and they send real intent.
3. Check that AI crawlers can read your own site. An assistant that searches cannot use a page it cannot fetch. The free AI visibility checker tests access for the main crawlers.
4. Do not buy a package because an AI listed a directory. Open the directory, search for a business like yours and see whether the listings look alive.

If you would rather have the listings built and kept consistent than do it by hand, that is what our citation building service does, across 50+ countries, with every live listing verified on a public page.