Skip to main content
Cheers

Research

Is ChatGPT Showing the Right Phone Number for Your Business?

ChatGPT prints a phone number in 9 of 10 US home-service answers. We tested 900: 17.6% of the US numbers we could check weren't on the business's site.

Dylan Allen-Arnegard, CEO and Co-Founder of Cheers
Dylan Allen-Arnegård

CEO & Co-Founder, Cheers

14 min readPublished

Last verified .

A plumber's hand in a rust canvas cuff presses a key on a cream corded wall phone in a basement utility room, its coiled cord hanging beside a copper water line

Since Oct. 9, 2026, ChatGPT has printed a phone number in more than 9 in 10 US home-service answers in our research panel. On Nov. 1 we checked 900 of its October numbers against each business's own site, and about 1 in 6 of the US ones we could check weren't there. If you use tracking numbers or several lines per branch, the number under your name may not be the one you'd pick.

Last verified: November 1, 2026.

Start with your five busiest markets and call the number under your name from an outside phone. The website comparison tells you which numbers need attention; the call tells you whether a customer reaches the right branch.

What changed on October 9

In the Cheers Market Baseline, our research panel, 12.6% of US ChatGPT answers carried at least one phone number from Oct. 2 to 8, 2026, about 1 in 8. From Oct. 9 to 31, 92.5% did (5,835 of 6,307).

Sistrix, which samples logged-out ChatGPT queries in six countries including the US, dated ChatGPT's switch to a building-block layout to the night of Oct. 8 to 9, 2026, in a report published Oct. 30. Before it, an answer was running text with a source list. The report doesn't mention phone numbers or local answers, but our panel shows the phone jump on the same night: 14.8% of answers in the US and Canada carried a number on Oct. 8, and 87.7% did on Oct. 9.

Each answer also holds more numbers than it used to: ChatGPT printed about 6.4 times as many phone numbers per answer after Oct. 9 as before (4.53 against 0.71, across all 100 metros). That count includes repeats, so it measures numbers on the screen, not how many businesses ChatGPT recommended.

Google's engines stayed far below ChatGPT. Google AI Mode barely moved, from 8.8% to 11.1% of US answers with a phone number, while Gemini rose from 4.6% to 14.5% as the Gemini answers our panel captured moved from place cards to cited answers.

Cheers Market Baseline, US

ChatGPT went from 1 in 8 answers with a phone number to 9 in 10

Share of US home-service answers with at least one phone number

  • ChatGPT: 1,910 and 6,307 answers
  • Google AI Mode: 1,910 and 6,299 answers
Sistrix dates ChatGPT's switch to a building-block layout to the night of October 8 to 9, 2026 (report of Oct 30, 2026; logged-out queries in six countries). In our panel the phone jump came the same night. The same 32 questions ran in the same 60 US cities in both windows.

Cheers Market Baseline: answers with text to 32 home-service questions in 60 US metros, Oct 2 to 31, 2026; pulled Nov 1, 2026.

US answers with at least one phone number, Oct 2 to 8 vs Oct 9 to 31, 2026
EngineOct 2 to 8Oct 9 to 31
ChatGPT12.6%92.5% (1,910 and 6,307 answers)
Google AI Mode8.8%11.1% (1,910 and 6,299 answers)

How many of those numbers are on the business's own website

We looked for each sampled number on the website of the business it was printed under. Of the 374 US numbers we could check, 17.6% (66) weren't on the site in the pages we checked, and that's after we read every mismatch by hand. Some we couldn't check at all, because the answer linked no business site, the site didn't load or it showed no phone number. Against all 469 sampled US numbers, 62.3% turned up on the business's site; the two shares sit on different bases, so they don't add up to 100.

Canada's sites matched far more often, with 5.9% of 356 checkable numbers missing. Across both countries, 11.9% of the 730 checkable numbers weren't on the site.

Phone number check, ChatGPT, October 2026

About 1 in 6 US numbers we could check weren't on the business's site

17.6%

of US phone numbers we could check weren't on the business's own website (homepage, contact page, the page ChatGPT linked and up to 8 more)

Base: 374 checkable US numbers from ChatGPT answers, Oct 2 to 31, 2026

62.3%

Of all sampled US numbers, confirmed on the business's own site

469 US numbers

5.9%

Canada, same check

356 checkable numbers

11.9%

US and Canada together

730 checkable numbers

Checkable means the site loaded and showed at least one phone number. A number missing from the site can still reach the business, for example a tracking line or another branch. We didn't dial any number.

Cheers Market Baseline ChatGPT answers, Oct 2 to 31, 2026; business websites checked Nov 1, 2026; every mismatch read by hand.

17.6% of 374 checkable US phone numbers printed by ChatGPT in October 2026 weren't on the business's own website in the pages checked. 62.3% of 469 sampled US numbers were confirmed on the site. Canada 5.9% of 356; US and Canada together 11.9% of 730.
FindingValueBase
of US phone numbers we could check weren't on the business's own website (homepage, contact page, the page ChatGPT linked and up to 8 more)17.6%374 checkable US numbers from ChatGPT answers, Oct 2 to 31, 2026
Of all sampled US numbers, confirmed on the business's own site62.3%469 US numbers
Canada, same check5.9%356 checkable numbers
US and Canada together11.9%730 checkable numbers

The share held steady through the change: 17.4% of checkable US numbers weren't on the site in the week before Oct. 9, and 17.8% weren't after. Volume is what moved. With the share flat and each answer carrying more numbers, a customer now runs into far more numbers that the business's site doesn't show.

Our estimate, counting repeats across both countries, is about 7 such numbers per 100 answers before Oct. 9 and between 45 and 59 per 100 after. The low end assumes every number we couldn't check is on the business's site, and the high end assumes those behave like the ones we did check. Either way it counts printed numbers, not calls or lost jobs.

Call tracking is the clearest pattern

When a US site loaded a call-tracking script, 41.0% of the numbers ChatGPT printed for that business weren't on the site (32 of 78), against 12.1% when it didn't (34 of 280). Across the US and Canada it was 34.0% (36 of 106) against 8.5% (51 of 603).

Phone number check, ChatGPT, October 2026

ChatGPT's number was missing from the site far more often when the site ran call tracking

Share of checkable numbers not on the business's own website

Bars run from 0 to 50%.

US

Site runs a call-tracking script41.0%

32 of 78 numbers

No call-tracking script12.1%

34 of 280 numbers

US and Canada

Site runs a call-tracking script34.0%

36 of 106 numbers

No call-tracking script8.5%

51 of 603 numbers

We read each page's HTML before any script swaps the number, so on a tracking site we compared ChatGPT's number with the site's default number. Unresolved cases are left out of these counts.

Cheers Market Baseline ChatGPT answers in 100 metros (60 US, 40 Canadian), Oct 2 to 31, 2026; sites checked Nov 1, 2026.

Not-on-site share by call-tracking script, October 2026
Numbers fromSite runs a call-tracking scriptNo call-tracking script
US41.0% (32 of 78 numbers)12.1% (34 of 280 numbers)
US and Canada34.0% (36 of 106 numbers)8.5% (51 of 603 numbers)

We read each page's HTML as the server sends it, before any script swaps the number, so on a tracking site we compared ChatGPT's number with the site's default number, the one a visitor sees if the script never runs. A tracking line can still ring the right branch without ever showing up in that HTML. And a tracking number on your profile or in a directory can land in an answer while your site shows something else. If you run tracking numbers, our call-tracking playbook covers how to keep one stable public number per location and still measure every call.

The page ChatGPT linked showed a second pattern. When it linked a page deeper than the homepage, such as a location page, the number was missing from the site 21.9% of the time (21 of 96, both countries), against 10.8% when it linked the homepage, and we checked the linked page itself in both cases.

Trace the unmatched numbers back to their source

We read all 87 confirmed misses, and not one was a one-digit slip or a swapped pair of digits. Of the 87, 83.9% shared an area code with a number on the business's site, and 60.9% sat on sites that show exactly one number, a different one. More than half came from sites with no call-tracking script at all. None was printed in three or more metros, so they don't look like one national line repeated across markets.

All 900 passed our number-format checks. Tracking numbers on profiles, older directory listings and other branch lines are the first places to investigate. Dial the number, identify its owner and correct the source if it sends customers to the wrong place. Our check covered website matches and number formats; the call-routing check belongs in your branch audit. Yelp told shareholders on Aug. 6, 2026 that its ratings and reviews "recently began powering ChatGPT's local experience in relevant categories," but in our sample, numbers on Yelp-labeled map cards were missing from the business's site about as often as numbers on other cards.

Check your five busiest markets this week

Pick your five busiest markets and ask ChatGPT what customers there ask: an emergency plumber, same-day AC repair, no-heat furnace repair, the best electrician in that city. If you work one metro, ask five different job questions there instead, and ask each one twice, since answers vary from run to run. For every answer, write down the number printed under your business name and the page ChatGPT links beside it. Then call that number from a phone outside the office and note who picks up, because what you're testing is whether it reaches someone who can book the job.

If the right branch answers, the number is doing its job even when your site doesn't show it, so add it to the list of lines you keep routed. If anyone else picks up, or nobody does, find where that number is published before you change anything: start with the page ChatGPT linked, then search the 10 digits in quotes on Google to find the directories and old profiles that still list it. Our guide to fixing wrong business information in ChatGPT has the steps for correcting each source.

Which number should ChatGPT show?

Four steps per location

Settle one number per location, then make every source agree

  1. 01

    Pick one number per location

    The line you want customers and AI tools to call, with a named owner.

  2. 02

    List every tracking number

    Write down where each one is published: Business Profile primary and up to two more, ads, directories, the website.

  3. 03

    Match the page's default number

    The number in your HTML before any script swaps it, plus the tel: link and LocalBusiness telephone, should match the profile and listings.

  4. 04

    Keep printed numbers ringing

    Every number ChatGPT prints for you should ring at the right branch until it stops appearing.

A number ChatGPT prints can be a tracking line you own. Keep it routed until it stops appearing in answers.

Google Business Profile guidelines and Google Search Central LocalBusiness docs, read Nov 1, 2026.

  1. Four steps to settle which phone number ChatGPT should show for a location
  2. 1. Pick one number per location: The line you want customers and AI tools to call, with a named owner.
  3. 2. List every tracking number: Write down where each one is published: Business Profile primary and up to two more, ads, directories, the website.
  4. 3. Match the page's default number: The number in your HTML before any script swaps it, plus the tel: link and LocalBusiness telephone, should match the profile and listings.
  5. 4. Keep printed numbers ringing: Every number ChatGPT prints for you should ring at the right branch until it stops appearing.

Start with one number per location, the line you want any AI tool to show, and put a person's name next to it as the owner. Then list every tracking number in use and where each one is published. A Business Profile can hold a primary number plus up to two more, so all three go on the list, along with ads, directories and the website.

Google's rule for profiles fits in one line: "The phone number must be under the direct control of the business." Its guidelines also ask for a number that connects to the individual location, prefer a local number over a central call-center line where possible, and rule out numbers that redirect or refer callers to other businesses. They don't mention tracking numbers, so what matters is whether you control the line and where it rings.

Next, make the default number in your page's HTML, the one visitors see before any call-tracking script swaps it, match the number your profile and listings show. Put it in visible text, in a tel: link and in the LocalBusiness telephone property with the country and area code, as Google's structured data documentation (last updated Sept. 8, 2026) recommends. Do that on the homepage, the contact page and every location page, so each branch's page carries that branch's number in plain HTML. For brands with many branches, the LocalBusiness schema guide shows how to give each location page its own telephone value that matches the number on the page.

Plenty of contractor sites do half of this. In a Cheers census of 976 readable US home-service contractor homepages, read on Nov. 1, 2026, 86.0% had a clickable phone link, but only 47.0% put the phone number inside LocalBusiness schema. Schema gives crawlers an explicit phone-number field alongside the number customers see. Keep both consistent with the branch line you want published.

While you're in there, check that your robots.txt file doesn't block OAI-SearchBot, which OpenAI's crawler overview describes as the crawler that surfaces websites in ChatGPT's search features.

Keep every number ChatGPT prints for you ringing at the right branch. Don't cancel or recycle a tracking number that still shows up in answers, even if it's long gone from your website. After an acquisition, list every number the acquired brand ever published and keep each one routed until it stops appearing; our guide to transferring Business Profile ownership covers the profile side while the answers catch up.

Re-run the check monthly, and again after any call-tracking change, merger, move or website vendor switch. ChatGPT changed its whole answer layout overnight in October, so date every check you run.

AI agents have started placing calls to businesses

A public phone number is now one that software can dial. Meta's Muse beta for calls to US businesses expanded on Sept. 16, 2026, according to an X post by Ryan Fox that TechCrunch reported the next day. TechCrunch reported Google's test of Gemini Call for Me on Sept. 24, 2026; Google says Gemini starts every call by disclosing that it's an AI assistant from Google calling on a recorded line, and the call comes from the customer's own number.

Google's help page says Call for Me can ask for hours, availability and service quotes and can book appointments. For now it's limited to adults in the US with a Pixel 11 on a US SIM, a Google AI subscription, the Phone by Google beta and English. Neither company says where its agent finds a business's number, so assume it can dial any number listed for you and make sure each one reaches someone who can quote the job and book it.

Google's own answers lean on the phone when the job is urgent. On urgent questions in October 2026, Gemini put a phone number in 44.8% of US answers and Google AI Mode in 32.6%, against about 5% on planned-job questions.

Your Business Profile settings have switches for the automated calls Google itself places to businesses, which the Call for Me help page doesn't mention; our guide to Google AI phone calls covers those switches.

The phone-versus-booking-link comparison shows how ChatGPT and Google AI Mode offer different routes from an answer to a business.

Method

The Cheers Market Baseline asks ChatGPT, Google AI Mode and Gemini the same 32 home-service questions every week in 100 metro areas, 60 in the US and 40 in Canada. Each day runs a different subset, so one day isn't like-for-like with a full week. This study uses answers from Oct. 2 to 31, 2026, with the same 32 questions in the same 60 US cities in both windows; the smallest has about 300,000 people.

Phone share. The US comparison uses 1,910 ChatGPT answers from Oct. 2 to 8 and 6,307 from Oct. 9 to 31; Google AI Mode had 1,910 and 6,299 answers with text, and Gemini 1,909 and 6,306. Volume figures count repeats. The website check covers ChatGPT only.

Sample. We drew 900 business numbers at random by printed number from 772 answers, 400 from Oct. 2 to 8 and 500 from Oct. 9 to 31, and set aside 9 utility, city and 311 or 911 lines. Each number goes to the business named nearest above it in the answer and is checked against the website the same answer links for it. Within one answer, a repeated number counts once.

Weighting. We oversampled the week before the change (400 of about 2,257 printed numbers, against 500 of about 47,625 after), so pooled shares are unweighted. Weighted by window, about 12.8% of checkable numbers across both countries weren't on the site and about 65.4% of all numbers were. For current figures, use Oct. 9 to 31: 17.8% of 219 checkable US numbers weren't on the site, and 59.2% of 289 sampled US numbers were.

Site check. We crawled the sites publicly on Nov. 1, 2026, honoring robots.txt: 706 sites, 663 of them readable. On each we read the homepage, the contact page, the page ChatGPT linked and up to eight more contact, location, about and service-area pages, checking tel: links, schema telephone values and visible text, plus a second pass over the raw HTML. A number is checkable when its site loaded and showed a phone number. Of the 900, 94 had no business site and 45 pointed at one that didn't load.

Exclusions. We set aside 16 toll-free numbers and 3 whose map card linked a Facebook or Google page, so the on-site shares (62.3% US, 69.1% overall) leave out 12 toll-free numbers we found on the business's site.

Hand review. We read all 108 numbers that failed the first match by hand. Of those, 87 are confirmed misses and 21 stay unresolved (16 of them US); unresolved ones never count as misses or enter the call-tracking and linked-page splits. Intervals are Wilson 95%, close to what resampling whole answers gives: the US share of 17.6% runs 14.1% to 21.8%, and with no single-digit or transposition matches among 87 nonmatches, the upper bound for those specific patterns is 4.2%. We did not test every possible error or dial the numbers.

Map cards. Numbers on Yelp-labeled cards were missing in 17.8% of 101 cases, against 14.4% of 202 on other cards (Oct. 9 to 31, before hand review).

Schema census. On Nov. 1, 2026, Cheers read the homepages of 1,096 US home-service contractor websites from Google's local map results, with Cheers customers removed; 976 were readable. We read raw HTML without running JavaScript, so the phone-link and schema shares are floors.

Limits. This is one engine over one month. We read static HTML, so a number a script inserts after the page loads is invisible to the check; the second pass found 11, and finding more would lower the share, not raise it. A number only on a deeper page than we fetched counts as a miss. We didn't crawl directories or dial any number, so a miss means "not on the business's own website in the pages we checked," never that the number is wrong.

To cite this study: Dylan Allen-Arnegard, "Is ChatGPT Showing the Right Phone Number for Your Business?," Cheers GEO Academy, November 2026, https://cheers.tech/geo-academy/chatgpt-phone-numbers-local-answers.

Sources

Sources read November 1, 2026.

Dylan Allen-Arnegård is the CEO & Co-Founder of Cheers, the done-for-you local visibility partner for established home-service companies and PE-backed portfolios.

Frequently Asked Questions

Call the number ChatGPT shows from an outside phone and check which branch answers. In our October sample, 17.6% of the checkable US numbers were absent from the business's website. That finding identifies numbers to investigate; the call confirms whether they reach the right business.

Check the page ChatGPT links beside the number, then search the 10 digits in quotes on Google. Tracking lines, old directory listings and other branch numbers are the first places to investigate. In our check, 83.9% of the unmatched numbers shared an area code with a number on the business's site.

Keep your Business Profile, website and directories in agreement so each gives an assistant the right branch number. OpenAI hasn't identified a single source for business phone numbers, and our website comparison did not trace each number to its source. Treat consistency across those pages as the practical job.

A tracking number can still ring the right branch. It becomes a problem when it stops ringing there, when your reports can't tell where the call came from, or when it's a pool number your vendor can take back. In our check, US numbers on sites running a call-tracking script were missing 41.0% of the time, against 12.1% without one.

Treat every public number as one a customer or their AI assistant can call. Muse has a US business-calling beta, and Google's Call for Me can ask for service quotes and book appointments. Keep each published line routed to someone who can help, including tracking numbers that still appear in answers.

Written by

Dylan Allen-Arnegard, CEO and Co-Founder of Cheers

Dylan Allen-Arnegård

CEO & Co-Founder, Cheers

Dylan leads Cheers, the done-for-you local visibility partner for established home-service companies and PE-backed portfolios. He started out helping frontline teams earn reviews.

Share this article

ChatGPT prints a phone number in 9 of 10 US home-service answers. We tested 900: 17.6% of the US numbers we could check weren't on the business's site.

Next step

Is AI recommending your business?

Find out how visible you are across ChatGPT, Gemini, Perplexity, and AI Overviews.

Keep reading