How Do I Fix Wrong Business Information in ChatGPT?

From Oct. 9 to 31, 2026, 92.5% of US ChatGPT home-service answers in the Cheers Market Baseline, our own research panel, carried at least one phone number, up from 12.6% the week before. A customer can repeat that number, or the hours beside it, to whoever answers your phone. When it's wrong, the fix starts outside ChatGPT, on the page the answer read.

Last verified: November 1, 2026.

That jump landed with a new layout: Sistrix found ChatGPT switched to a building-block layout overnight on Oct. 8 to 9, 2026, though the phone figures are ours. Star ratings were already there, mentioned in 76.0% of US and Canadian answers the week before the change and 75.7% after, general advice included.

OpenAI hasn't published a listing editor for local businesses, as far as we could find on Nov. 1, and its Get Quote beta is a quote flow, not a correction tool (our Get Quote guide explains it). The first two steps cost nothing: a line in your intake script and a note of the number you want shown at each branch.

What ChatGPT prints about you now

Warranty or guarantee mentions rose after the change too, from 48.4% of US answers to 79.4%, so add your warranty terms to the facts you check.

When we checked 900 phone numbers ChatGPT printed in October 2026 against each business's own website, 17.6% of the US numbers we could check weren't on the site in the pages we checked, and none of the confirmed nonmatches matched the single-digit or transposition errors we checked. In the US the share was 41.0% when the site ran a call-tracking script, against 12.1% when it didn't. We didn't dial them, but they were properly formed local lines, so any one could be a tracking line, an old number still listed somewhere or another branch's line. Inspect the cited and other public pages, confirm the branch's intended number and correct any stale record you find. The four cases below show how to handle the source and the answer separately.

A rating lower than your Google one is often the Yelp card's: for the same US businesses, ChatGPT's map cards linked to Yelp averaged 4.34 stars where Google AI Mode showed 4.89, on far fewer reviews. Why ChatGPT shows a lower star rating than Google explains which cards carry it.

Write down exactly what the caller heard

The first person at your company to hear an AI claim is usually whoever answers the phone. One owner told us his callers open by repeating an AI's price comparison (accurate, in his case), and another said people call saying ChatGPT recommended them. That's reason enough to add "AI assistant (ChatGPT, Gemini, other)" to your how-did-you-hear options this week.

Then add one line to the intake script: when a caller says "I saw that you..." or "ChatGPT said...", whoever's on the phone logs it in a shared sheet. Give it columns for call time, branch and number dialed, the caller's exact words, which assistant ("unknown" is fine), fact type, what's true per the approved record, who took the call, fix owner and incident number. Keep fact types to a short dropdown: phone number, hours, open or closed, service offered, service area, price or offer, rating or review count, and other.

Decide which number you want shown

Before anyone touches a listing, write the approved record for each location, starting with which number you want customers to see. Otherwise a team can end up "fixing" a tracking or branch line it wants live.

Google's guidelines for representing your business ask for a local number for the location instead of a central call-center line whenever possible, one the business controls directly, and nothing that redirects callers to other businesses. They don't mention call tracking. Our read is that a tracking number fits if you own it and it forwards straight to the branch, and our call-tracking guide covers keeping one without splitting the listing.

Pick one number per location, list every other live number and where it's published, and have operations sign off hours, services, service area and open status, with a date. Keep any old number ChatGPT still prints ringing at the right branch until it stops showing up.

The record matters past ChatGPT, too: on Oct. 29, 2026, Search Engine Roundtable (SER) reported wider early access to a Google pilot where an AI agent answers customers' texts. Google's example is a restaurant, so don't assume a home-service brand can switch it on today.

Find the page the answer read

In one of our ads I say, "When we started tracking what AI and Google were saying about these badge customers, we realized it was telling us exactly what was holding each one of them back." A wrong fact points the same way, at a page somebody can fix.

Ask ChatGPT the question your customer asked, naming the customer's city. Open the block for your business, find the source linked for the fact that's off, and save the URL with the date, because that page is what you fix, not ChatGPT.

If the link doesn't settle it, start with the sites ChatGPT cites most. In October 2026, Angi or HomeAdvisor appeared among the sources of 69.4% of US ChatGPT answers in the Cheers Market Baseline that listed any source, Expertise.com in 63.8% and Yelp in 30.3%. Check best-in-city list sites too. Those shares tell you where to look, not which site supplied a claim; run the source audit by industry to see which sites the engines lean on for your trade.

What you find will fit one of four cases:

• The source is wrong. Fix it; the next section goes through each kind.
• The source is right and the answer misread it. Save both, send feedback on that response, and don't rewrite an accurate page to match a bad summary.
• No source shows. Record what the interface showed and don't guess which directory said it.
• The answer picked the wrong branch. Record the city you asked about, the branch it named and the phone route it gave.

Sometimes the page it read is yours. Suppose an answer files your battery-backup company under solar because your homepage talks about solar charging; then the fix is your own copy.

Fix the fact where it lives

Assign each correction to the team that owns the record, then verify the live public page and repeat the answer check.

Google Business Profile (listings or marketing). Edit the primary number, plus up to 2 more, hours and services to match the approved record, never to match a wrong answer. Google reviews edits before they go live and gives no time, so screenshot the live profile once the change shows.

Google's automated calls (branch manager). Decide who at each branch answers Google's automated calls, because what staff tell one can become a profile update that sits as "Pending" for up to 5 business days, and Google may post updates to a verified profile on your behalf. Check the three switches under Advanced settings and keep a dated note of how they're set.

Your location page (web). Put the branch's number in visible text, a tel: link and the LocalBusiness telephone property with the country and area code, matching the profile, and keep the live URL and the date you changed it.

Your crawler access (web). Allow OAI-SearchBot in robots.txt and at the CDN or firewall. OpenAI's crawler overview names it as the crawler behind ChatGPT search, set independently of GPTBot, and says search picks up a robots.txt change in about 24 hours; keep the diff.

Yelp (listings). For a page marked closed, check for a "Moved to" banner and rule out an after-hours "Closed," then suggest an edit. On a page you've claimed, locked fields go through the contact option behind the "i" icon. Yelp publishes no timeline, so save a public page that confirms the right value.

BBB (listings or the regional operator). If the profile says out of business or shows returned mail, send the local BBB office proof the location operates, such as a live page or a recent invoice address. BBB reports a real closure for a year, so ask for a wrong one to be corrected rather than waiting it out.

Directory, dealer-locator and best-in-city list pages (marketing). Send the publisher the URL, the exact wrong line, the approved value and a public page that confirms it, and keep that note with its date.

Photos and ads (listings; paid media). In September SER covered a Maps contributor whose photos put one scammer's phone number on unrelated businesses, so check your listing the way a customer sees it, report any photo like that and keep a screenshot. If your Google Ads use location and call assets, Google's statement in an SER report says location ads "may direct calls to the phone numbers associated with those locations," so make sure the profile number is one you want those calls to reach.

ChatGPT itself (whoever logged the incident). When the cited page is right and the answer misread it, use the feedback control on that response and save it with its source. Feedback covers only that response, so send it alongside the source fix, never instead of it.

If the branch really did close, move or merge, follow the closed-location cleanup instead of deleting every old reference.

Ask again in 30 days before you close it

Treat 30 days as your team's checkpoint, not a refresh time ChatGPT promised. In days 1 to 3, reproduce the answer in the same city, save the source and the approved value, and assign an owner; in days 4 to 10, correct the source, then check the public page itself. Between days 11 and 30, ask the original question again in the same city, plus two variations, and record whether the answer, the cited source, the named branch and the phone route are right.

If the source is fixed and the answer hasn't moved by day 30, look for a second copy of the old fact before you decide the fix failed. Date every check and save where each link went, because ChatGPT's link format kept changing in late October; an undated screenshot can't tell you which version you saw.

And don't close a ticket because a dashboard says "saved." Mark the source correction complete when it is public, record fresh answer checks separately, and keep monitoring: one correct answer does not establish a lasting fix.

Also compare the ratings an answer quotes with their source pages; our AI review-rating study shows why the source platform matters.

Method

Cheers Market Baseline. The Cheers Market Baseline is our own research panel, not customer data: every week it asks ChatGPT, Google AI Mode and Gemini the same 32 home-service questions in 100 metro areas, 60 in the US and 40 in Canada. US figures come from 8,217 ChatGPT answers in the 60 US cities in October 2026: 1,910 from Oct. 2 to 8 and 6,307 from Oct. 9 to 31, of which 5,835 carried a phone number. Phone and warranty shares compare those two windows, counted Nov. 1.

A star rating or phone number counts when the answer text mentions it, general advice such as "look for 4.5 stars" included. For the before-and-after star figures we used all 13,692 October answers from both countries, 3,185 of them from Oct. 2 to 8 and 10,507 from Oct. 9 to 31. Source shares use the 8,100 US answers that listed at least one source, and they show where answers linked, not which source supplied each fact. The Yelp-card comparison matches 1,252 US businesses by name and city across ChatGPT and Google AI Mode answers from Oct. 20 to 31.

Phone check. We sampled 900 phone numbers from 772 Market Baseline ChatGPT answers (Oct. 2 to 31, 2026, US and Canada) and checked each one on Nov. 1 against the business's homepage, its contact page, the page ChatGPT linked and up to 8 more contact, location, about and service-area pages, reviewing every mismatch in context. The US share uses the 374 numbers we could check, 66 of them not found; the call-tracking comparison uses 358 resolved US numbers, 78 on sites with a tracking script and 280 without. Across both countries, 73 of the 87 confirmed mismatches shared an area code with a number on the site.

Nobody dialed a number and we didn't crawl directory pages, so a mismatch means only that the number wasn't on the business's own website in the pages we checked. A number a script adds after the page loads can slip past a check like this, which would push the not-on-the-site share up, not down.

Sources

• What changed in ChatGPT in October: GPT-6, UTM tracking and sources, Sistrix, Oct. 30, 2026. The building-block layout and its dates, not the phone figures.
• Guidelines for representing your business on Google, Google Business Profile Help, undated. The local-number, direct-control and no-redirect rules.
• Google Business Profiles AI-Powered Messaging Expands, Search Engine Roundtable, Oct. 29, 2026. Wider early access to Google's AI messaging pilot.
• Edit your Business Profile, Google Business Profile Help, undated. A primary number plus up to 2 more, and Google's review of edits.
• About automated calls and texts from Google to your business, Google Business Profile Help, undated. The Advanced settings switches and the 5-business-day "Pending" window.
• Local Business (LocalBusiness) Structured Data, Google Search Central, Sept. 8, 2026. The telephone property with country and area code.
• Overview of OpenAI crawlers, OpenAI, undated. OAI-SearchBot, its setting apart from GPTBot and the 24-hour robots.txt window.
• How do I report that a 'closed' business page is actually open?, Yelp Support Center, undated. The "Moved to" banner, after-hours check and suggested edit.
• How do I report incorrect business information if the fields are locked?, Yelp Support Center, undated. The contact option behind the "i" icon on a claimed page.
• What BBB reports on, BBB, July 23, 2026. The one-year closure note and returned-mail notes.
• Google Maps Phone Number in Photo Scam, Search Engine Roundtable, Sept. 10, 2026. The contributor whose photos carried a scammer's number.
• Google LSAs Showing Google Business Profile Number, Not Google Ads Number, Search Engine Roundtable, Sept. 25, 2026. Google's statement on location and call assets; the headline says LSAs, the body covers Google Ads.

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.

First published August 7, 2026 by Amadeus Peterson; rewritten November 2026.