Seen / Guide
How local businesses show up in ChatGPT: a checklist for agencies
Agency checklist: what to fix on a client's site, Google Business Profile and listings so AI assistants describe them correctly.
When an assistant describes a local business, it is working from whatever it can find or remember about that business. The agency's job is to make the true facts easy to find and consistent. This guide is a set of checklists you can run for any client, plus a test and a list of things not to promise. It does not require any particular tool.
What AI assistants can and cannot know about a local business
Some facts about a business stay put for years: the name, the address, the type of business, the main services. Assistants tend to get these right when the sources agree with each other.
Other facts change often and are easy to get wrong: opening hours, holiday closures, prices, whether a service is available this week, who the owner is. If an old page still shows the old version, an answer may repeat it. We cannot see inside any engine, so we cannot say which source wins when they disagree. What we can say is that disagreement between sources is a likely cause of wrong answers, and it is the part you control.
Use this as a rule when you plan the work: fix the volatile facts first, and make every source say the same thing.
Checklist 1: website basics
WEBSITE [ ] Opening hours are written as text on the page, not only in an image or PDF [ ] Hours are on the contact page and in the footer, and they match [ ] One main phone number, written the same way in header, footer and contact page [ ] Address written the same way everywhere, matching the Google Business Profile [ ] Each service has its own page or a clear section, in plain words [ ] Services the business does NOT offer are stated, if people often ask [ ] Structured data (LocalBusiness) is present and matches the visible hours and phone [ ] Holiday or seasonal hours notice is removed or updated after the date passes [ ] Old pages (previous owner, old location, old promotions) are redirected or removed [ ] Site loads on a phone and the hours can be read without zooming
Check the structured data by viewing the page source or using a testing tool of your choice. The point is that the machine-readable hours and the visible hours say the same thing.
Checklist 2: Google Business Profile
It is not established how or whether any assistant uses a client's profile, so do not tell the client that it is. It is still the most visible record of the business on Google, and Google's own AI Overviews sit on top of Google Search. Keep it correct. Google's Business Profile help covers the editing steps.
GOOGLE BUSINESS PROFILE [ ] Business name matches the real name, with no added keywords [ ] Primary category is the best fit, secondary categories are accurate [ ] Regular hours are correct for every day [ ] Special hours are set for upcoming holidays, and old ones are cleared [ ] Phone number is the correct main line [ ] Services list is filled in and matches the website [ ] Website link goes to the right page [ ] Business description is plain and factual [ ] Duplicate profiles for the same location are merged or removed [ ] Questions and answers section has no outdated answers
Checklist 3: directories and old listings
Directories are where old facts hide. Audit them in this order: the ones that appear on the first page when you search the business name, then industry directories, then the long tail.
DIRECTORIES AND LISTINGS [ ] Search the business name and note every listing on the first two pages [ ] Search the old phone number and old address in quotes [ ] For each listing: name, address, phone, hours, website match the profile [ ] Industry directories for this trade are checked [ ] Duplicates found, with a note on how to merge or report each [ ] Closed locations or former names are marked closed or redirected [ ] Listings you cannot edit are logged with a correction request and a date [ ] The profile sheet is updated so the next audit starts from the same facts
Keep a log with a link to every listing and the date you last checked it. It is dull work, and it is the work that fixes most wrong answers.
Checklist 4: reviews and mentions
Reviews and articles are text an assistant may read. If they contradict the profile, they add noise.
REVIEWS AND MENTIONS [ ] Skim recent reviews on the main platforms for quoted hours, prices or services [ ] Flag reviews that state old hours, an old address or a service no longer offered [ ] Reply politely to those reviews with the current fact [ ] Look for local news or blog mentions with outdated details [ ] Ask the editor to correct factual errors where you can reach one [ ] Do not ask the client to delete honest negative reviews
A public reply that states the correct hours is a small, honest fix and is visible to everyone who reads the review.
Checklist 5: test it
After the fixes, test what the assistants say. Use the five questions from the guide to finding and fixing wrong hours in AI answers: closing time on a given day, the phone number, whether a service is offered, holiday hours, and a "best [service] in [city]" question that does not name the client.
TEST [ ] Same five questions, same wording, in ChatGPT, Perplexity and Gemini [ ] Google search checked for an AI Overview on the same questions [ ] Each question run more than once, with the number of runs written down [ ] Every answer logged with date, engine and screenshot [ ] Each wrong answer traced to a source where one was shown [ ] A recheck date set for after the fixes have had time to be picked up
To turn the results into something the client will read, use the layout in the monthly report guide.
What not to promise the client
DO NOT PROMISE [ ] A ranking, a position or a guaranteed mention in any assistant [ ] A date by which wrong answers will disappear [ ] That an assistant reads the Google Business Profile or any single source [ ] A visibility score you cannot explain [ ] That the same question will always give the same answer [ ] Control over what an assistant says
What you can promise is the work: a confirmed profile, consistent facts across the places you can edit, a documented test and a recheck. Say that AI answers vary and are not deterministic, and put it in writing. A client who hears this early is easier to keep than one who was promised a result you cannot deliver.
If you run this across many clients, the repeated testing is the heavy part. Seen is being built to handle that step and compare the answers with each client's profile, and you can join early access if you want to see it.