AI Visibility Monitoring: Know What ChatGPT and Google's AI Tell Your Customers

Your analytics cannot see the conversation

Most AI answers never send a visitor. The customer reads three names, picks one, and calls. If yours was not one of them, nothing shows up anywhere: no bounce, no lost session, no line on a chart. Owners find out the old way, when a regular mentions that ChatGPT suggested someone else, or when they ask it themselves at midnight and wish they had not.

The free reports help, and we set all of them up, but each one sees a sliver. None of them records the words of the answer, which is the part that decides the call.

Google also says plainly that no third-party tool has access to its internal ranking or AI systems. So monitoring does the honest thing: it asks the engines the same questions a customer would and writes down what comes back. If you want to see the GA4 side first, our guide to tracking AI traffic in GA4 builds the channel step by step.

Seven things we record for every question

A single yes or no is not enough. Being named fourth with the wrong phone number is a different problem from being left out, and it gets a different fix. For each question, on each engine, on each run, the record holds:

• Named or not. The headline number, counted across repeated runs rather than one lucky answer.
• Position in the answer. First of three reads differently from a footnote under five rivals.
• What it said about you. Hours, phone, prices, services, service area and claims, checked against your real facts so errors surface before customers repeat them.
• Who was named instead. The competitors that appear, how often, and on which questions.
• The sources it cited. The directories, articles, review sites and pages the engine leaned on, which become the to-do list.
• The tone. Whether the description is neutral, glowing, or built around a complaint.
• Engine, date and wording. So a change can be tied to a fix, an engine update or plain variation.

Rolled up, the named-or-not column becomes your AI share of voice: the share of answers that name you, set against the same figure for each competitor. It is the closest thing this field has to a ranking, and we report it with the sample size attached so nobody mistakes noise for news.

Choosing questions your customers really ask

The prompt set is where most monitoring goes wrong. Track vanity questions and you get vanity results. Say a roofer in Boerne tracks "best roofing company" with no town attached: the answers will be national brands and roofing blogs, and the chart will be flat forever. The questions have to sound like the people who pay you.

• Service and place: "Who installs metal roofs near Boerne?" and its neighbors in the suburbs you serve.
• Problem first: "My roof is leaking after hail, who should I call?" Customers describe symptoms, not services.
• Comparisons: "Is a local roofer better than a national chain?" and head-to-head questions naming you and a rival.
• Your name: "What are the hours for your company?" and "Is your company any good?" These catch wrong facts and bad summaries.
• Spanish: where your customers ask in Spanish, the same questions in Spanish, because the answers can differ.

Each question is asked more than once per cycle, because AI answers vary from one asking to the next. We revisit the set whenever services, seasons or competitors change. The method is laid out in full in our prompt tracking guide, and you can try a single question today with the free AI visibility check.

AICiteBeam, Delphi and the free consoles

We did not want to rent our core instrument, so we built it. Monitoring at DataCram runs on three kinds of tools that check one another.

A thermometer does not lower a fever

Plenty of dashboards will tell you that your AI visibility is 23, down from 26, with a small red arrow. Very few tell you why, and fewer still do anything about it. Monitoring on its own is a thermometer. It is worth having, and it is worth nothing if nobody treats the patient.

A chart that only moves when the engines update is a weather report, not a marketing program.

That is why our monitoring is part of the monthly SEO program, not a separate subscription. The answers feed the work: the cited sources become a list of directories and articles to earn a place in, the wrong facts become listing corrections, the competitor gaps become pages. When an answer goes wrong, the matching guide in AI search fixes becomes the playbook, whether the problem is a competitor named in your place or wrong business details.

How the engagement runs, month by month

1. 1

   Free GEO audit

   We run a first set of your customers' questions through the engines, check the site and listings, and show you where you stand. The free GEO audit ends with a fix list and a fixed monthly quote.

2. 2

   Agree the questions and the rivals

   Together we settle the question set, the towns or regions, the languages and the competitors worth tracking. You know who takes your jobs better than any tool does.

3. 3

   Record a baseline

   Before anything changes, every question is run on every engine more than once. This is the before photo, and it keeps everyone honest later.

4. 4

   Fix, then watch

   The program works through the fix list: crawlability, listings, reviews, content and mentions. Alerts flag sharp moves between reports so a wrong fact gets caught between reports, not at the next meeting.

5. 5

   Report and reset

   Each month you get the answers, the trend and the next fixes. When the business changes, we prune questions that no longer matter and add new ones.

Inside the monthly report

If you are comparing agencies, our page on what a GEO report should show lists the red flags, starting with any single "AI score" nobody can define.

Price, and when a tracking tool is enough

Monitoring comes inside our monthly programs, which start at $625 a month; most local businesses land between $625 and $2,250, depending on how competitive the questions are and how much needs fixing. There is no long contract, the scope is written down, and the tier is set after the free audit. The SEO pricing page and GEO pricing explain what moves the number, and SEO financing is available.

Now the part most agencies leave out. If you have someone in-house who will act on the numbers, a tool alone may be enough. As of October 2026, Otterly.AI's Lite plan is $29 a month for 15 prompts, Semrush's AI Visibility Toolkit is $99 a month per domain billed annually, and Ahrefs' Brand Radar AI starts at $199 a month. Our GEO tools comparison covers the rest. If your team wants a plan and a second opinion rather than a managed program, GEO consulting is the better fit. And if your customers never ask an AI about businesses like yours, monitoring is money better spent on local marketing.