Skip to content
← All posts
TroubleshootingUpdated 10 min

AI Says My Business Is Permanently Closed: How to Fix

By the MentionedOn team

Fixing what AI already gets wrong about your business.

A customer tells you they almost did not come in, because they asked ChatGPT and it said you had closed. Or the phone volume drops for a week before anyone mentions why. This is a revenue emergency with a boring cause: somewhere in the chain of records that describe your business, a status field says closed, and the assistants are repeating it. The fix is mechanical. The part nobody tells you is that fixing the source does not un-say the answer, because the assistant may not be reading your profile live at the moment it answers.

Why an assistant thinks you are closed

In practice there are six common causes. Identify yours before you start editing, because the remedy differs.

  • A stale "permanently closed" flag on a map profile. Major map platforms accept public edits (Google's "Suggest an edit" is the best-known), and a competitor, a confused customer, or an over-eager local guide can propose a closure. Owners and local-search practitioners also report closures appearing after signals like a disconnected phone number or long inactivity, though the platforms do not document what actually triggers it.
  • An old listing for the prior tenant at your address. Address records are sticky. If a restaurant closed in that unit before you leased it, its dead listing sits on the same coordinates, and a retrieval system asked "is this address open" can surface the wrong one.
  • A move that left a duplicate behind. You relocated, created a fresh profile at the new address, and the old one stayed live and eventually got flagged closed. Now there are two records with your name and one says closed.
  • An aggregator propagated a wrong status. Wholesale business-data compilers feed a long tail of apps and directories. One bad status upstream re-emerges in places you have never heard of, weeks later, and can even come back at you as a suggested edit on your primary profile.
  • You were temporarily closed and never reopened the record. A remodel, a seasonal break, a storm. Someone toggled the status, nobody toggled it back, and after a long enough dormancy a temporary flag can harden into a permanent-looking one.
  • The model is reciting a snapshot, not a live lookup. This is the one that surprises owners most, and it is covered below.

The expectation to set before you touch anything

Assistants answer local questions along two different paths, and they behave differently after a correction.

Some answers use a live search or map lookup. Open the actual citations and check that they identify your business and support the closure claim. A response without a verifiable source leaves the cause unresolved.

Other answers come from the model's weights, which encode a snapshot of the web from before the model was trained. If "permanently closed" was true of your business in the sources at snapshot time, no amount of profile editing changes that answer until a later model or a retrieval step overrides it. Caching between a search index and a generated answer can add further lag, though the assistants do not document how much.

An edit can appear on its original source before it appears in an assistant’s answer. Check the source directly, then repeat the buyer question on the same surface. There is no fixed retest interval that proves a correction has propagated.

Triage, in order

1. Find out exactly what each assistant says, and screenshot it

Do not rely on the customer's paraphrase. Open a fresh, logged-out session in each assistant your customers actually use and ask the way a customer would: "Is [business name] in [city] still open?", "What are [business name]'s hours?", "Best [your category] near [neighborhood]". That last one gives you a read on whether you are being silently dropped from recommendation lists as well as being declared closed. Run it a few times, since the answers vary.

Screenshot every wrong answer with the date visible, and note the exact prompt. Two reasons: assistants are non-deterministic, so an answer you cannot reproduce later is still evidence you had the problem; and if the wrong information originated from a third party, dated evidence is what escalation paths ask for. Ask each assistant where it got the claim: retrieval-backed answers will often name a source, which hands you the record to fix.

2. Fix the authoritative profile first

For most local businesses the highest-leverage record is the Google Business Profile, because it feeds Google's own surfaces and is widely mirrored. Sign in as the owner, set the open/closed status back to open, and confirm hours, address, phone, and category in the same pass. Apple's own business portal controls how your place appears in Apple Maps, and Microsoft's business-listing tool is worth claiming for the same reason on its surfaces. Both have been renamed and consolidated in recent years, so start from each vendor's current business-listing entry point rather than an older product name you find in an old post. If you cannot log in, recovering ownership is step zero. An unclaimed profile is one where anyone can suggest changes and you have no privileged way to override them.

Expect the possibility of re-verification. Platforms increasingly ask for a video walkthrough showing signage, the interior, and evidence of your control of the premises. Verification flows change often, so follow the prompts in the dashboard rather than a checklist from a blog post.

Confirm the listing’s name, street address and map link before editing. For a mistaken closure on the same Google listing, follow Google’s reopening instructions. That guidance treats a permanent closure recorded as part of a move or name change differently. Follow the instructions for the current location, and keep the old record distinct from the active one.

3. Hunt duplicates and merge them

Search the map platform for your business name, your phone number, your street address, and any former name or former address. Look for records with slight spelling variations, suite numbers present or absent, and old phone numbers. Duplicates split your reviews, contradict each other on status, and are a common reason a correction "does not stick": you fixed one record and an assistant read the other. Use the platform's own duplicate-reporting or merge flow rather than editing both to say the same thing.

4. Correct the aggregators and the directories that feed them

Below the map platforms sits a wholesale layer of business-data compilers (Data Axle, Foursquare, and Localeze are the names that come up most in the US), plus consumer platforms like Yelp and Facebook that many apps read directly. Each has its own claim-and-correct route, and the routes change; go to each company's current business-listing portal instead of following an old link.

Then do your industry's directories, which matter more than generic ones for exactly the queries you care about: the state license lookup, the trade association roster, the chamber of commerce, the insurance or franchise locator, the booking platform, the delivery marketplace. These are the pages retrieval-backed assistants cite for professional categories, and a closed status on a licensing lookup is unusually damaging.

If you are considering a paid syndication service to push corrections at once, ask one specific question first: what happens to the pushed data if you cancel? Some of what these services provide is a managed overlay rather than a permanent edit at the source.

5. Make the correct status unambiguous on your own site

Your website is the source you fully control and the one an assistant can cite verbatim. Put a plain, dated sentence above the fold: "Open and serving customers as of [date]. Hours below." Assistants extract sentences; give them one worth extracting.

Add or repair LocalBusiness structured data with a complete openingHoursSpecification, plus specialOpeningHoursSpecification for holidays, and make sure name, address, and phone in the markup match the profile character for character. Note the limitation honestly: schema.org has no standard "we are open" boolean, so full, current hours plus visible dated copy is how you signal it. Validate the markup with a schema testing tool before you call it done.

Then add a page that answers the question directly: an FAQ entry titled "Is [business name] still open?" with a one-paragraph answer. It reads as strange to a human and it is the kind of page a retrieval step can find and quote.

6. Generate a few fresh signals dated after the correction

A correction with no corroboration is one record's word against a dozen. In the days after you fix the profile, create things the web can see and date: post an update to the profile, publish a short post on your own site, ask a handful of recent customers to leave dated reviews (never scripted, never incentivized; review platforms police that), update social profiles, and get one local mention if you can. The goal is a cluster of recent evidence that contradicts the stale claim.

Worked example: the answer points to your old address

In this example, a bike repair shop moved from 12 Market Street to 80 Pine Avenue. An assistant answers “permanently closed” and links to the old address. The active profile and the shop’s location page already show current hours at Pine Avenue. The owner keeps the old closure intact and checks which source still connects customers to Market Street.

  1. Save the evidence. Save the question, date, assistant and any visible search mode. Copy the exact cited URL, name and address, then open the link. A citation that points to Market Street supports the status of that location only.
  2. Correct the source. A local directory still sends customers to Market Street. The owner uses the directory’s correction form to update the address and links to the shop’s current location page. If the search citation had instead named the active Google listing and shown a wrong closure flag, the owner would correct that listing directly.
  3. Recheck the source and the answer. Once the directory shows Pine Avenue, repeat the original question in a fresh thread with the same city and search settings. Also ask whether the Pine Avenue location is open. Record each answer and its citations. Mark the correction as visible at the source even if the assistant still returns the old claim.
  4. Escalate where the error persists. If the directory still displays the old address, reply through its support channel with both public links and the date of the request. If public sources are accurate and the assistant still returns the claim, use the product’s feedback control if available and include the dated answer and current public location URL.

When a closure flag will not come off

If the status reverts or a correction is rejected, use the profile’s support or appeal path. Share the affected listing URL, the dates you saw the error, and a link to the current business location. Use the private support channel for any verification documents the platform requests. A public community post can use the public URLs and a screenshot with private information removed.

Protecting revenue while it propagates

  • Put the dated "we are open" line on the homepage, the booking page, and the voicemail greeting.
  • Reply publicly to any review or comment that mentions the closure, with the current date in the reply. Those replies are publicly visible text that search and retrieval systems can pick up.
  • If you run paid search, you control the ad copy directly and can put "Open today" in a headline within the hour. One caveat: location assets and map ads are typically linked to the business profile itself, so those may misbehave until the profile is fixed. Check what is actually serving.
  • Email and text your existing list. Your regulars are the customers most likely to have quietly stopped coming.
  • Brief whoever answers the phone: "You may have seen we were closed online. That was an error: we are open, and here are today's hours."

Preventing the recurrence

Keep a single source-of-truth document with your exact name, address, phone, hours, and categories, and make every listing match it. Keep the profile owner account on an address more than one person can access, so notifications about suggested edits are actually seen. Never use a "temporarily closed" toggle without a calendar reminder to undo it. If you move, use the platform's move flow on the existing record instead of creating a new one. And re-run the search queries yourself on a schedule, because nobody downstream will tell you.

That last point is the real gap. Owners find out about this from a customer, weeks in. Checking what the assistants say about you on a recurring cadence, rather than after the damage, is the difference between a bad week and a bad quarter. That is the job monitoring does, and it is also something you can do manually with a calendar reminder and ten minutes. Either way, the check has to exist. See also the local business AI visibility checklist for the underlying signals that keep your record correct in the first place.

See where you actually stand

The trial covers ChatGPT, Claude, Gemini, Perplexity, Grok and Copilot: all six when it starts and in the Monday sweep, with daily checks on ChatGPT and Gemini.

$0 today, then $99/month. Card required. Cancel anytime.