Google
Rarely. Across about 350,000 U.S. locations SOCi measured in 2026, ChatGPT named 1.2% when asked to recommend a business, while 35.9% of those same locations showed up in Google’s local pack. Hold two things before any of these numbers mean anything: SOCi measures the 2,751 multi-location brands it manages for clients, not independent businesses, and CallRail, the other study here, measures its own customer base.
Does ChatGPT recommend local businesses at all?
Rarely, and it is the stingiest of the three assistants. Each one names a sliver of what Google’s local pack already shows.
SOCi, Local Visibility Index 2026 · about 350,000 U.S. locations across 2,751 multi-location brands SOCi manages, not a sample of independent businesses
View data
| Item | Value |
|---|---|
| Google local pack | 35.9% |
| Gemini | 11% |
| Perplexity | 7.4% |
| ChatGPT | 1.2% |
A location counts as recommended when a standardized prompt puts it in the first five results. The study does not measure what position it landed in.
How many customers actually reach you through an AI assistant?
Fewer than four in a thousand. Of more than 30 million inbound calls CallRail tracks, 0.353% began in an AI search, against 0.073% in January 2026.
CallRail, on its own customer base · nearly 20 million calls in the January 2026 report, more than 30 million in the latest update
View data
| Item | Value |
|---|---|
| January 2026 | 0.073% |
| Prior month | 0.203% |
| Latest update | 0.353% |
That is close to five times the January rate. The multiplication is ours: CallRail publishes the three shares, not a growth figure.
CallRail measures calls to the businesses that use its call tracking. It is not a reading of the U.S. market as a whole.
A call that started in an assistant and arrived without a detectable origin is never counted in that share.
This share is of phone calls. The 1.2% is of locations. Two different universes, so the two never add up.
Of the calls attributed to AI search in September, ChatGPT accounted for 98.65%, Perplexity 0.88% and Claude 0.46%.
Does anyone publish how an assistant picks a business?
No. Neither OpenAI nor Google publishes ranking documentation for local recommendations. That absence is the honest answer, and it is why nobody can sell you a place in one.
How a crawler reaches your pages, and that an assistant goes out and reads what is published on the open web.
The order of the few names it hands back, or why it picked your competitor over you.
How much a rating, a review count or a distance weighs inside that order.
A file that grants access grants access, not standing. See whether your business needs an llms.txt file.
This is the model, not a measurement. No step here has a documented weight.
View data
| Step | Value |
|---|---|
| Somebody asks | 01 (recommend a business of this type in this market) |
| The assistant searches | 02 (it goes out to the open web) |
| It reads what is published | 03 (your profile, your site, your reviews) |
| It names a few | 04 (no published rule orders them) |
Which businesses are already getting AI-driven calls?
Legal and home services. Both run above CallRail’s 0.353% average, and both are measured on that same customer base.
CallRail, September 2026 · its own customer base, detectable origins only
View data
| Item | Value |
|---|---|
| Legal | 0.761% |
| Home services | 0.616% |
| Manufacturing | 0.549% |
| Retail | 0.511% |
SOCi, by category: retail 10.2%, restaurants 5.3% and financial services 3.1% of locations recommended by ChatGPT.
CallRail, by industry: legal 0.761% and home services 0.616% of inbound calls.
One study counts locations, the other counts calls. A good share in one says nothing about the other.
What should you do this quarter?
The same three things that already win local search. Nothing AI-specific pays yet, because nothing AI-specific is documented.
- 01Make the profile complete and correctHours, categories, service area, photos. It is the record the local pack reads first, and 35.9% of SOCi’s locations appeared there. See what makes a profile show up in the map pack.
- 02Ask for real reviews, and answer themAsk while the job is still fresh, and reply to what arrives. Ratings and review counts are visible to anyone reading your listing, human or not.
- 03Make the site readable and fastIf a crawler cannot read the page, nothing downstream can repeat what is on it. Plain text, real headings, a page that loads.
New reviews per month
nobody askedevery finished job
The ask is the step most owners skip.
Appearances in the local pack
unmeasuredtracked monthly
35.9% of SOCi’s locations appeared. Know your own number.
Calls attributed to AI search
unmeasuredtracked monthly
CallRail’s base runs at 0.353%. Yours is probably zero today.
Profile fields left blank
counted oncezero
Hours, categories and service area, checked every quarter.
View data
| Metric | Day zero → target |
|---|---|
| New reviews per month | nobody asked → every finished job |
| Appearances in the local pack | unmeasured → tracked monthly |
| Calls attributed to AI search | unmeasured → tracked monthly |
| Profile fields left blank | counted once → zero |
When somebody asks whether ChatGPT recommends you, the honest answer is almost never, and the honest plan is the one above.
Almost never today, and fewer than four inbound calls in a thousand start there.
Nobody can sell you a ranking that nobody publishes.
The profile, the reviews and the site. See what an audit finds on yours.
Questions owners ask
Will ChatGPT recommend my business?
Rarely, today. Across about 350,000 locations SOCi measured in 2026, ChatGPT named 1.2%, while 35.9% of the same locations appeared in Google’s local pack. Gemini named 11.0%. SOCi measures multi-location brands it manages, not independent businesses. Treat an AI recommendation as a bonus on top of local search.
Are customers finding local businesses through AI assistants?
Barely, but it is moving. Of more than 30 million inbound calls CallRail tracks on its own customer base, 0.353% began in an AI search, against 0.073% in January 2026. That is still fewer than four calls in a thousand, and it only counts calls whose origin can be detected.
How do I get ChatGPT to recommend me?
Nobody can tell you honestly: neither OpenAI nor Google publishes how a local recommendation is ranked. The work that helps is the work that already helps local search, which is a complete profile, real reviews, and a fast and readable site.
Should you pay for AI visibility or GEO services?
Not for a promised ranking. No ranking is documented, so nobody can promise you a place in one. Paying for a complete profile, real reviews and a fast site is the same work, and it pays in Google today, where 35.9% of those locations already appear.
Sources
- SOCi, Local Visibility Index 2026 (about 350,000 U.S. locations across 2,751 multi-location brands SOCi manages; a standardized recommendation prompt, first five results counted, position not measured). Publisher site; the report page could not be opened from our environment, so nothing here is quoted (2026)
- CallRail, AI search attribution across its own customer base (nearly 20 million inbound calls in the January 2026 report, more than 30 million in the latest update; only calls with a detectable origin are counted). Publisher site; the report page could not be opened from our environment, so nothing here is quoted (2026)
The first meeting starts with findings, not promises.The free Business Health Report reads your business and shows you what it found, in days.
Get Your Free Business Health Report