Seen

Seen / Guide

Why AI assistants give wrong hours for a local business, and how to fix it

ChatGPT, Gemini or Perplexity gave the wrong hours or phone for a client? A step-by-step way to find the source of the error and fix it.

A client calls you. A customer drove to the shop on Sunday because ChatGPT said it was open. Or Gemini gave out a phone number the business dropped two years ago. This guide is a four-step routine to find the wrong fact, trace it to a source and fix it. It works by hand with a spreadsheet and no special software.

What the wrong answer looks like

The mistakes are usually small and specific. That is what makes them costly, because a confident answer sounds like it was checked. These are the kinds you will see most often:

Take a screenshot of each wrong answer with the date, the engine and the exact question. You will need it in step 2 and again when you show the client what changed.

Where AI engines get local facts

Here is what is documented and what is not. Some assistants can search the web when they answer. ChatGPT search, Perplexity and Google AI Overviews show sources for at least part of an answer, so you can often click through and see which page was used. An assistant that is not searching answers from its training data, which can be out of date.

What is not public is how much weight each source gets. We cannot say that an engine reads a client's Google Business Profile, or that it prefers one directory over another. It is reasonable to suspect that the business website, the Google Business Profile, directories, review sites and old pages all feed answers somewhere. Treat that as a working guess and let the evidence in step 3 tell you which source mattered for each error.

Step 1: write down the correct profile

You cannot call an answer wrong until you have written down what is right. Ask the client to confirm each field, in writing, once. Keep one profile per client in a shared sheet. Copy this template:

CLIENT PROFILE
Business name (exact):
Also known as (other names people use):
Address:
Service area (cities or radius):
Main phone number:
Other numbers that are valid:
Numbers that are NOT valid (old lines):
Website:
Hours, Monday to Sunday:
Holiday hours this quarter:
Services offered:
Services NOT offered (things people ask about):
Prices that are public (if any):
Booking method (walk-in, phone, online):
Confirmed by (name) on (date):

The "not valid" and "not offered" lines matter most. They tell you what to search for when you trace an error, and they stop you from arguing with the client later about what was true.

Step 2: ask the same five questions in each engine

Use the same wording every time so results can be compared. Run them in ChatGPT, Perplexity and Gemini, and check a Google search for the Google AI Overview. Here is the set, using Maple Street Dental, a made-up practice in Austin, as the example:

  1. What time does Maple Street Dental close on Saturday?
  2. What is the phone number for Maple Street Dental in Austin?
  3. Does Maple Street Dental offer emergency appointments?
  4. Is Maple Street Dental open on public holidays?
  5. What is the best emergency dentist in Austin?

The last one is different. It does not name the client, so it shows whether the business is recommended at all. For a real client, swap in the service, the city and the correct hours.

Log every answer in a sheet with these columns: date, engine, question, answer as written, correct or wrong, wrong fact, source link if one was shown. Run each question more than once. AI answers vary between runs and are not deterministic, so a single wrong answer is a lead and a repeated wrong answer is a problem. Pick your own number of repeats and write it down, so next month's check uses the same method.

Step 3: trace each wrong answer to a source

Start with the source links, if the engine showed any. Open them and look for the wrong fact. When there are no links, search for the wrong fact yourself:

Write the finding next to the error in your sheet. Some errors will have no findable source. Note that too, and move on.

Step 4: fix at the source, then recheck

Fix in this order, because it puts the most control in your hands first:

  1. The client website: hours as plain text, one phone number in header and footer, structured data matching the profile.
  2. The Google Business Profile: hours, special hours, phone, services. Google explains how to edit these in its Business Profile help.
  3. The directories that showed the wrong fact in step 3. Claim or update the listing, or ask for a correction where you cannot edit.
  4. Duplicate or closed listings. Merge or mark them where the directory allows it.

Then wait before you recheck. Search engines and directories update on their own schedules and we do not know how long an assistant takes to reflect a change, so do not promise the client a date. Recheck with the same five questions and compare against your first log. If the wrong answer remains, go back to step 3 and look for a source you missed.

Tip: keep the screenshot of the wrong answer and a screenshot of the corrected one in the same row. Clients understand a before and after at a glance.

How to make this a monthly routine

One check finds today's errors. A monthly check shows whether fixes held and catches new ones, such as holiday hours that were never changed back. Put the five questions in a recurring task for each client and keep the log in the same sheet. When it is time to report, the layout in how to add an AI visibility section to your monthly report turns the log into one page. If you want a list of what to fix on a client's site and listings before you test, use the checklist for how local businesses show up in ChatGPT.

Doing this by hand across many clients takes real time, and you should measure your own before deciding whether it is worth automating. Seen is being built to run these questions every month and compare the answers with your profile. If that would help you, you can join the early access list.