Seen / Guide

How to add an AI visibility section to your monthly local SEO report

What to put in the AI section of a monthly local SEO report: questions, engines, errors found, fixes made. A simple template.

Local clients have started to ask what ChatGPT says about their business, and a monthly report with no answer leaves you improvising on the call. You do not need a new reporting system. You need one page, added to the report you already send, that is honest about what can and cannot be measured. This guide gives you the measures, the layout and the wording.

Why clients ask about this now

In agency forums and on client calls, the same three questions keep coming up. Is my business mentioned when someone asks ChatGPT for a recommendation? Does it get my hours and phone right? What are you doing about it? None of these need a statistic to justify. A client who has heard that people ask assistants for local recommendations will want to see something in the report, even if the answer is "we checked and found two errors."

A short, plain section also protects you. If the client raises it first and you have no data, it looks like you have not been paying attention. If you raise it first with a page of findings, it looks like service.

What to measure

There are four things you can put on the page. They are not equally reliable.

• Correctness of facts. Does the answer give the right hours, phone, address and services? This is the easiest to defend, because you are comparing words against a profile the client confirmed. Start here.
• Mention. Is the business named in the answer to a question like "best emergency dentist in Austin"? Useful, but it changes from run to run.
• Position in the list. If the answer lists several businesses, where does the client appear? Noisy. Report it as a range or as "listed" or "not listed", never as a rank you could be held to.
• Sentiment. How the answer describes the business. Subjective. Quote the wording instead of scoring it.

Decision rule: lead with correctness, then mention, and treat position and sentiment as notes. If a measure would embarrass you when it moves for no reason, it does not belong in the headline.

Why one run is not enough

Ask the same question twice and you can get two different answers. Assistants do not behave like a rank tracker that returns one fixed result for a query. The wording changes, the list changes, and a source that was cited once may not appear the next time.

So run each question more than once per engine and report what happened across the runs. Choose a number of repeats you can sustain every month, write it into the footer of the report, and keep it the same. The exact number matters less than using the same method each time, so that month-to-month changes mean something. Say "listed in some of the runs" when that is what happened. Do not round it up.

The one-page layout

One page is enough. Clients read the top and skim the rest, so put the verdict first. Copy this layout into your report template:

AI VISIBILITY: [Client name], [Month Year]

SUMMARY (2 lines)
Checked in [engines] using [number] questions, each run [number] times.
[Number] wrong facts found, [number] fixed since last month.

QUESTIONS BY ENGINE
Question                          ChatGPT   Perplexity   Gemini
Hours on Saturday                 correct   wrong        correct
Phone number                      correct   correct      correct
Offers [service]?                 correct   correct      wrong
Best [service] in [city]          listed    not listed   listed in some runs

ERRORS FOUND
1. Engine, question, what it said, what is true
   Source of the error: [page or listing, or "not found"]
   Status: fixed / in progress / waiting on directory

FIXES MADE THIS MONTH
- What was changed, where, on what date

NEXT MONTH
- 2 or 3 specific tasks

METHOD NOTE
AI answers vary between runs. Results show what we saw on the dates
checked and are not a ranking or a guarantee.

The table cells should be short words, not paragraphs. Keep the full answers and screenshots in your own log and offer them if the client asks. If you have not yet built the check routine, the steps in finding and fixing wrong hours in AI answers give you the questions and the log columns this page draws from.

Wording that does not overpromise

The section is only worth sending if it stays true. Use sentences like these as a starting point:

• "We checked what ChatGPT, Perplexity and Gemini say about your business on these dates."
• "We found two facts that were wrong and corrected them at the source."
• "Answers differ from one run to the next, so we repeat each question and report what we saw."
• "We cannot control what an assistant says, but we can fix the information it may be drawing on."

Avoid these: "we will get you ranked in ChatGPT," "your AI visibility score went up," "guaranteed recommendation." Nobody outside the engines can promise those, and a made-up score invites the client to ask how it is calculated. If you also send the client a list of site and listing fixes, point to the agency checklist for how local businesses show up in ChatGPT so you stay consistent about what you promise.

Tip: put the method note on every report, in the same words. It reads as routine, and it is your answer if a client later asks why one month looked different.

Do it by hand or automate it

By hand, the work is: ask each question in each engine, repeat, log the answers, compare them with the profile, trace the errors and build the page. Time one client from start to finish and multiply by your client count. That number, not ours, tells you whether to automate. For a handful of clients a spreadsheet is fine.

Seen is being built for agencies past that point. It is planned to run the questions every month, compare the answers with the profile you enter and produce a white-label PDF in this shape. If you want to try it when it is ready, join the early access list.