Why ChatGPT, Gemini, Claude, and Perplexity Recommend Different Businesses

Say you run a roofing company in Atascadero. You ask ChatGPT for a good roofer in Atascadero, and your company is on the list. Then you ask Gemini the same question and get three other names. Perplexity gives you a different mix again, and Claude doesn't name anyone at all. It suggests checking a directory instead.

Nothing is broken. These engines don't read the same parts of the web, and they don't behave the same way once they've read it. Some lean on directories and review sites. Some lean on business websites and Google's own map data. Some decline to shortlist local businesses at all.

This guide covers that gap between engines: what published studies have observed about each assistant's sources, how to check your own results fairly, and what not to read into a mismatch.

Between engines vs. the same prompt twice

There are two different kinds of "the AI gave me a different answer," and it helps to keep them apart.

The first is asking one assistant the same question twice and getting two different results. That's normal, and we cover it in our guide to same-prompt AI variance. The second is asking four different assistants the same question and getting four different shortlists. That's what this page is about.

Both happen at once, so a fair comparison needs a method. Our guide to checking whether AI recommends your business across engines walks through it, and we won't repeat those steps here.

A single blended "AI visibility score" hides exactly what you need to see. If ChatGPT names you, Gemini doesn't, and Claude declines to name anyone, an average of those results describes none of them. Keep each engine's result separate.

Different engines lean on different sources

None of these companies publishes how its assistant decides which businesses to name. What researchers can do is ask a lot of local questions, record which sources each engine cited, and look for patterns.

The figures below come from those published studies. They are not Honest Visibility's own research. They describe tendencies in specific samples at specific times, and source mixes do shift from month to month.

ChatGPT leans on directories in several local studies

Cheers ran 32 home-services prompts across 100 US and Canadian metro areas for the 28 days ending September 2, 2026, and classified 356,019 citations (Cheers, updated September 2, 2026). ChatGPT split roughly evenly, with 48.8% of its citations going to contractor websites and 44.7% going to directories and review platforms.

Steady Demand saw the same general lean when it compared Gemini and ChatGPT on matched local-service queries. In its Citation Ledger, Gemini leaned toward businesses' own websites and ChatGPT leaned toward directories (Steady Demand, updated September 9, 2026).

That points to a practical clue, not a guarantee. Don't assume your Google Business Profile covers you on ChatGPT. Accurate third-party listings are worth checking, and so is the Bing side of things. Our guide to Bing Places and ChatGPT visibility covers that path.

Gemini and Google AI Mode lean on business websites and Google's own data

In the same Cheers baseline, Gemini pointed to contractor websites in 80.7% of its home-services citations. Google AI Overviews and AI Mode were also site-heavy at 70.3%, with another 11.4% of citations going to google.com itself.

SOCi's 2026 Local Visibility Index looked at a very different group: multi-location brands, such as chains and franchises. In that sample, Gemini recommended a higher share of locations (11.0%) than ChatGPT or Perplexity. SOCi also measured Gemini's profile accuracy at 100.0%, noting that Gemini is grounded in Google Maps, which the study used as its source of truth (SOCi, July 29, 2026).

That's a chain and franchise sample. It tells you something about direction. It doesn't tell you what will happen to a single-location salon in San Luis Obispo.

The practical clue here is to keep your Google Business Profile and your website saying the same things: name, hours, services, service area. And test AI Mode on its own. Showing up in the local map results on Google doesn't mean you'll show up in an AI Mode answer.

Perplexity cites widely from the live web

Cheers found that 72.3% of Perplexity's home-services citations went to contractor websites. In a qualitative comparison, growth-onomics described Perplexity as the most citation-dense of the assistants it looked at, pulling from a wide spread of domains with sources attached throughout the answer (growth-onomics, August 19, 2026). That's an observed description, not a measurement.

The clue is to have owned pages with plain facts that are easy to quote, plus some visible corroboration from other sites. Because Perplexity shows so many sources, you may spot your site in the source list without your name in the answer. Track those two things separately.

Claude is selective and sometimes declines to name anyone

Nico Digital put 15 local buyer questions to ChatGPT, Gemini, Perplexity, and Claude on September 1, 2026. Claude declined to name businesses on 5 of the 15 and pointed the buyer to directories instead (Nico Digital, September 1, 2026). Most of those prompts were about Indian cities, with three international mirrors, and each ran once. It's a snapshot, not a rule.

Formative Digital looked at Claude from another angle. In a May 2026 scrape of 1,732 Claude citations for local questions in Ontario, curated directories led the way. ThreeBestRated alone was cited 116 times, and Google Maps never appeared as a standalone citation (Formative Digital, May 2026). growth-onomics described Claude as more selective, with fewer sources and more hedging.

The clue is directory and corroboration hygiene. If Claude sends a customer to a directory, you want your listing there to be complete and correct. Don't panic when it declines, and measure Claude separately from the others.

Tendencies from specific samples at specific times, not rules. Source mixes shift month to month. Treat what you find as a clue, not the proven cause.

Sources: Cheers, 32 home-services prompts across 100 US and Canadian metros, 28 days ending Sep 2, 2026 (356,019 citations). Nico Digital, Sep 1, 2026. Formative Digital, May 2026 (Ontario). growth-onomics, Aug 19, 2026. Third-party studies, not Honest Visibility research.

What the cross-engine gap numbers actually say

According to SOCi's 2026 Local Visibility Index, ChatGPT recommended 1.2% of brand locations, Gemini 11.0%, and Perplexity 7.4%. The same brands showed up in Google's local 3-pack (the three map results at the top of a local search) 35.9% of the time. That sample is multi-location brands. It isn't a census of independent shops, and the magnitudes may not carry over to a five-person business.

Brands showed up in Google's local 3-pack far more often than AI assistants recommended them

Share of brand locations each AI assistant recommended, next to how often the same brands showed up in Google's local 3-pack (the three map results at the top of a local search). Sample: multi-location brands (chains and franchises), not independent single-location businesses.

Claude isn't shown because SOCi's index didn't include it. These rates may not carry over to a five-person business.

Source: SOCi's 2026 Local Visibility Index (SOCi, July 29, 2026). Third-party study, not Honest Visibility research.

Steady Demand found about 8.3% average overlap between the domains Gemini and ChatGPT cited on matched local queries. When both engines named a recognizable business, they put the same business first about 4.9% of the time.

Nico Digital found that on 11 of its 15 questions, no single business was named by all four assistants.

Source mixes also move. Steady Demand's September update showed Gemini's Reddit citation share falling from 13.7% to 1.7% between its July and September 2026 waves, while its directory share rose from 10.4% to 19.1% (Steady Demand, September 9, 2026). One month of movement isn't a new permanent rule, but it's a good reminder that any snapshot has a date on it.

Put together, these studies say the engines are selective and disagree often. They don't say your business has a particular percent chance of being named anywhere.

Cited is not recommended

An engine can read your website and still not recommend you, so it helps to sort each engine's answer into one of five labels.

When you look at an AI answer, label what happened, one engine at a time.

1. Recommended Your business is named as an option in the answer.
2. Mentioned Your business comes up, but not as a recommendation.
3. Cited only Your website or listing appears in the sources, but your name isn't in the answer.

   Example: For a question about agencies in Kolkata, Perplexity pulled Nico Digital's site into its sources and then named other agencies in the answer. (Nico Digital, September 1, 2026)

4. Not present You're nowhere in the answer or the sources.
5. Incorrect facts You appear, but with wrong hours, a wrong phone number, or services you don't offer.

Those labels beat a simple yes or no. The multi-engine check guide has a simple grid for recording them.

A fair diagnosis when engines disagree

If one engine names you and another doesn't, the goal is to understand the pattern, not to force a better answer. A fair check looks like this:

1. Ask the same unbranded question a customer would ask, like "who does kitchen remodels in Paso Robles," in each engine.
2. Start fresh chats, and write down the date, time zone, and any settings like location or web search.
3. Change one thing at a time if you run it again, so you know what made the difference.
4. Save the answer and the source list for each engine.

Then look at the clues by stack. For Google's surfaces, compare your Google Business Profile against your website. For ChatGPT, check your Bing Places listing and confirm your site shows up in Bing at all. For ChatGPT, Claude, and general corroboration, look at your major directory and review profiles. For every engine, make sure your own pages state the operational facts plainly: hours, services, and the area you serve.

These are clues, not causes. A missing Bing listing or a thin services page might explain why one engine skipped you. It might also have nothing to do with it. The engines don't tell you why they chose what they chose.

One thing not to do is keep rerunning the question until your name shows up. That tells you the engine can name you sometimes. It doesn't tell you what a customer is likely to see.

If you'd like a recorded starting point, our free scan records one unbranded AI customer search. It's one snapshot, which is often enough to see whether there's a question worth digging into. When engines disagree and you want the broader four-engine pattern, that's what the complete analysis is for.

What not to conclude from a cross-engine mismatch

A mismatch is useful information, but it's easy to read too much into it:

• Being named on one engine isn't a permanent "AI rank." It's one answer on one day.
• Being missing on one engine doesn't mean you're invisible everywhere, or that you'll stay missing.
• A difference in your directory listings or website isn't proof of why an engine chose someone else.
• Claude declining to name anyone doesn't mean Claude has a problem with your business.
• Even on a single engine, the same question can give different answers on different runs. Our same-prompt variance guide explains why.
• No single fix guarantees a different answer next time, whether it's a Google Business Profile update, a Bing Places listing, or a directory cleanup.

If you're not showing up on any engine, the question is broader than disagreement. Our guide on why AI doesn't recommend your local business covers the wider set of reasons.

When a recorded multi-engine analysis helps

Most owners don't need weekly AI monitoring. They need to know whether there's a real problem and where to look first.

That's why we start with the free scan. It runs one unbranded AI customer search, gives you a private link to the report, and doesn't require an account.

If the engines disagree and one snapshot isn't enough to make a decision, the optional complete analysis is a one-time $29. It runs a broader recorded pass across ChatGPT, Claude, Gemini, and Perplexity, checks the accuracy of what they say about your business, compares you with competitors, and gives you a prioritized plan. It's there if you need it. A free scan with a clear answer is a perfectly good place to stop.

Short FAQ

Why does ChatGPT name my business but Gemini doesn't?

They tend to draw on different sources. In published local studies, ChatGPT has leaned more on directories and review platforms, and Gemini more on business websites and Google's own data. Treat any difference you find in your listings as a clue, not the proven reason.

Is Google AI Mode the same as Gemini?

No. They're separate Google products, and studies like Cheers' report them separately. Test each one on its own.

Why won't Claude recommend any local business sometimes?

In Nico Digital's test, Claude declined on 5 of 15 local questions and pointed buyers to directories. That's how it handled some local questions, not a judgment about your business. Keep your directory listings accurate for when it sends people there.

Does appearing in the Google local pack mean I'll show in AI answers?

Not necessarily. In SOCi's multi-location sample, brands showed up in the local 3-pack far more often than any AI assistant recommended them. The fundamentals overlap, which we cover in AI visibility vs. SEO, but a good local pack position doesn't carry over automatically.

Should I chase a different strategy for every assistant?

Usually not. Accurate listings and a clear website with your real hours and services help across the board. What changes by engine is where you look when something seems off, so measure each one separately.