AI says my company is closed

What is actually happening

You type your own company name into ChatGPT. It tells a buyer you are permanently closed. Or Gemini says the location shut in 2023. Or Perplexity hedges: "reports suggest this business may no longer be operating." Meanwhile your doors are open, your phone rings, and your invoices go out on time.

Three things are usually going on at once.

1. One upstream record is wrong

Assistants do not visit your shop. They read. The usual culprits are a map profile that someone marked closed, a directory page that was never updated after a move, an aggregator feed carrying an old status, or an archived news page about a different branch closing.

2. Silence looks like closure

If your website has no dated content since 2023, no visible hours, no structured data, and your social accounts stopped posting, a model has nothing recent to weigh against the one page that says closed. Absence of evidence gets read as evidence.

3. Each app reads a different pile

Five assistants asked the same question on the same day can give five different answers, because they pull from different indexes, different cached crawls and different live search partners. That is why a single screenshot is not a diagnosis.

There is also a naming trap. If you ask an assistant "is AI Knows Us still open?" you have already told it the business exists and matters. Ask the question the way a buyer asks it, without the name, and you see what buyers actually see.

How to confirm it is happening to you

Last verified: 17 October 2026

Do this before you change anything. You need a baseline, or you will never know whether your fix worked.

1. Write five blind questions. No company name. Use the words a buyer uses. For example: "Is there a late night pharmacy open in Indiranagar, Bangalore?" and "Which commercial laundry services in Pune are still operating in 2026?"
2. Write two named questions. "Is [company name] still open?" and "What are [company name]'s hours?" These are not your measurement. They are your diagnostic, because the answer often names the source it trusted.
3. Ask all five apps. ChatGPT, Gemini, Perplexity, Claude and Google AI Overviews. Same day, same wording, fresh chat each time, history and memory off where the app allows it.
4. Record four fields per answer: app, date and time, verbatim status wording, and any source link the app shows.
5. Repeat each question three times per app. Answers vary run to run. One run is an anecdote.
6. Count, do not guess. If 9 of 15 runs say closed, write "9 of 15 runs on 17 October 2026", not "60 percent of AI says we are closed". A handful of questions cannot carry a percentage.
7. Open the cited sources. Click every link the assistants show. One or two URLs usually explain the whole problem.

For a longer version of the prompt protocol, including how to keep wording stable across months, see #.

What to change, in order

Last verified: 17 October 2026

Honest order: fix the record that the most sources copy from, then your own site, then the long tail. Do not start with the long tail because it feels easy.

The steps, in order

1. Reclaim and correct the map profile first. If it reads permanently closed, reopen it and set exact hours, including holiday hours.
2. Put hours and a dated proof of life on your homepage. A visible "Open today until 7pm" line, plus a dated post or notice from this month.
3. Add structured data. Mark up your business with LocalBusiness and openingHours. Keep the same name, address and phone format everywhere.
4. Push the page for recrawl. Use the URL Inspection tool in Google Search Console and submit to Bing Webmaster Tools or IndexNow.
5. Check your robots rules. If you block AI crawlers, assistants may have only old third party pages to read about you. OpenAI documents its crawlers at platform.openai.com/docs/bots.
6. Correct the secondary listings in the table above, one by one, same wording, same hours.
7. Ask a few third party sites to say you are operating. A supplier page, a trade association member list, a local chamber directory, a dated press mention. Assistants weigh sites that are not yours.
8. Re-run the same blind questions after 14 days, then 30, then 60. Same wording. Compare counts, not feelings.

For a source-by-source correction checklist with the status fields to change on each platform, see #.

A worked example you can repeat

This is a hypothetical, written so you can copy the method. It is not a customer case study.

A two branch dental clinic in Pune closed its older branch in 2022 and kept the new one. Buyers started saying "the internet says you shut down".

1. Baseline, 1 September. Five blind questions, five apps, three runs each: 75 runs. Clinic appears in 12 runs. In 7 of those 12, the answer says the clinic closed or may have closed. Two apps cite the same old local news page about the 2022 branch closure.
2. Diagnosis. One map profile for the old branch still exists and is not marked closed properly, so the two listings merge in people's reading. The news page has no correction note. The clinic site has no hours in HTML, only inside an image.
3. Fixes. Old branch profile marked permanently closed with the correct date. New branch profile hours corrected. Homepage gains a text hours block and LocalBusiness markup. A dated October notice is published: "Our Kothrud branch closed in 2022. Our Baner clinic is open six days a week." The news outlet is emailed and adds a one line update.
4. Re-measure, 1 October. Same 75 runs. Clinic appears in 14 runs. Closed wording drops to 2 of 14.
5. Reporting. Write it as "7 of 12 on 1 September, 2 of 14 on 1 October". With counts this small, say the direction moved and keep measuring. Do not convert it into a percentage improvement.

How long before it moves

Nobody can promise a date. The honest version has three clocks.

• Platform clock. How fast the map or directory accepts and publishes your edit. Some edits appear in minutes, some need verification steps first.
• Crawl clock. How soon each AI crawler and search index fetches your corrected pages again. You can nudge this with recrawl submissions. You cannot control it.
• Model clock. Answers that come from training data rather than live retrieval change only when the model is retrained or when live search overrides the memory. This is the slowest clock and the one you have least influence over.

Practical guidance: measure at 14, 30 and 60 days after the fix. If the closed wording is unchanged at 60 days on apps that cite live sources, a wrong record is still out there and your diagnosis missed it. Go back to the cited links.

Where AI Knows Us fits

Most of the work above is something a careful owner can do alone with a spreadsheet and a free afternoon. If you want it measured and repeated rather than done once, that is where a tool helps.

AI Knows Us measures across ChatGPT, Gemini, Perplexity, Claude and Google AI Overviews. Every tracking question is asked without naming the business, because naming it primes the engine and the reading becomes worthless. Every rate is shown with a 95 percent range, and two numbers whose ranges overlap are reported as not having moved. It also does the fixing as well as the measuring: it writes the pages, publishes them, and asks other sites to name the business. The product runs 13 agents: intake, discovery, analyst, diagnosis, planner, writer, editor, publisher, scout, outreach, experiment, lifecycle and the copilot. It is made by Clyra Labs in India. Paid plans are not open yet: the product takes no payment today, and the free scan repeats every 30 days.

Who should not choose AI Knows Us

• You have one wrong map listing and nothing else. Fix the listing yourself. You do not need a measurement product.
• You need a signed commercial contract and an invoice today. Paid plans are not open yet.
• You want . No vendor can sell that, including this one.
• You want listing management across dozens of local directories as the main deliverable. Other categories of tool are built around that job.

In the adjacent category of AI visibility monitoring, several tools track assistant answers, including Profound, Peec AI, Otterly.AI and Scrunch AI. Compare their current pricing and method on their own sites before you decide.

FAQs