How to Measure Your Visibility in AI Answers

A fixed question list, stable test conditions, a tracking sheet and a one-hour monthly routine: a simple method for tracking what AI assistants say about your brand.

Published on October 1, 2026 · Strategy

To measure your visibility in AI answers, ask the same fixed list of questions to the same assistants every month, under the same conditions, and log in a spreadsheet whether your brand is mentioned, how it's described and which sources get cited. Then add the traffic you actually receive, which you can spot through the utm_source=chatgpt.com parameter.

This method builds on our pillar guide GEO: how to get recommended by ChatGPT without paying: you can't improve what you don't look at. It needs no paid tool, about an hour a month and a spreadsheet. As a running example we'll follow Harbor & Pine Realty, a fictional real estate agency in San Diego.

Why do answers change from one try to the next?

Before you measure, you need to know what makes an answer vary. Otherwise you'll read random noise as progress or decline. According to OpenAI's page Searching the web with ChatGPT, several factors come into play:

• Location: for local questions, ChatGPT can use an approximate location based on your IP address, or your device location if you've shared it.
• Memory: when it's on, ChatGPT may use your saved memories to rewrite its search.
• Query rewriting: ChatGPT often turns your question into one or more queries sent to partner search providers, and those results shift over time.
• Model and timing: the same question asked two days in a row can produce two different answers.

So measurement isn't about finding the one true answer. It's about spotting a trend by keeping everything you control identical from month to month.

Step 1: write your fixed question list

Write the questions your customers would really ask, in their own words. Split them into four families, a few questions each:

Two rules: always include the city or neighborhood in local questions so results don't hinge on your own location, and once the list is set, don't change it. If you want to try new questions, keep them in a separate section.

Keep your brand name out of the discovery, comparison and problem questions. Those three families exist to tell you whether assistants bring you up on their own, the way they would for a prospect who has never heard of you. Only the brand family should name you directly. And word the questions the way a real customer would, including their doubts and constraints ("I'm selling for the first time," "I need to move within three months"), because that's how people actually talk to ChatGPT.

Step 2: lock in your test conditions

1. The same assistants every month. For example ChatGPT with web search and Google's AI Mode. Add another assistant if your customers use it.
2. A fresh conversation for each question. The context of one conversation shapes the next.
3. No memory. OpenAI says Temporary Chat lets you chat without using or updating memory. It's the most neutral mode for testing.
4. Search turned on. In ChatGPT you can trigger it from the tools menu to make sure the answer draws on the current web.
5. The same week of the month, for instance the first one, to smooth out calendar effects.

Step 3: log each answer in your tracking sheet

Fill in one row per question and assistant. Here's the template to copy into a spreadsheet:

Add a "Notes" column for anything that doesn't fit: wrong information (hours, address), a missing service, a review being quoted. That's often where your most valuable fixes are hiding.

Step 4: the monthly routine, step by step

1. Open the spreadsheet and duplicate last month's tab.
2. Ask each question in a fresh conversation, no memory, search on.
3. Fill in the row, copying the cited sources.
4. Add up three simple totals: answers where you're mentioned, answers where the description is accurate, answers where your site is a source.
5. Compare with last month and note the changes.
6. Pick one or two actions for next month, no more.

Step 5: cross-check with real traffic

The sheet tells you whether you're mentioned; it doesn't tell you whether that brings visitors. Two sources fill the gap:

• Traffic from ChatGPT. OpenAI automatically adds utm_source=chatgpt.com to links clicked from ChatGPT search, according to its publisher FAQ. Filter on that source in your analytics tool and track the visits, leads and sales it drives.
• Google Search Console. Clicks from AI Overviews and AI Mode are counted in the "Web" search type, per Google's page AI features and your website. They aren't broken out from other traffic, so watch the overall trend on your key pages.

To turn those visits into business metrics, our lexicon of CPM, CPC, CTR, CPA and ROAS covers the vocabulary, and our guide ChatGPT Ads vs. Google Ads explains how paid and organic visibility differ.

Step 6: turn results into actions

What are the limits of this method?

It relies on a sample of questions, not on every conversation your customers have. One month proves nothing; wait for several readings before drawing conclusions. A changed answer might come from a model update rather than anything you did, which is why you keep conditions fixed and watch the trend. Finally, traffic data only covers clicks: someone who reads your name in an answer and then calls you without clicking won't show up anywhere. Ask new clients how they heard about you. Adding "an AI assistant like ChatGPT" as an option on your intake form or in your first call takes a minute and fills in the part of the picture no dashboard can show.

Where Oposto fits in this routine

The fixes that come up most often in this kind of tracking are a consistent description and a regular publishing cadence. Oposto learns your brand from your website and keeps the same description across all your posts, carousels and videos, published to 8 networks once you approve. It measures real performance so you can reuse what works. Try Oposto for free, no credit card required.

Frequently asked questions

How many questions should I track?

Enough to cover your four families (discovery, comparison, brand, problem) with a few questions each, and no more than you can log in an hour a month. Consistency matters more than volume.

Why use Temporary Chat to test ChatGPT?

According to OpenAI, Temporary Chat lets you chat without using or updating memory. Your answers depend less on your personal history, which makes month-to-month comparisons more reliable.

How do I see visits coming from ChatGPT?

Filter on utm_source=chatgpt.com in your analytics tool. OpenAI adds that parameter automatically to links clicked from ChatGPT search.

Does Search Console separate AI Overviews traffic?

No. Google says clicks from AI Overviews and AI Mode are included in the "Web" search type along with the rest of your search traffic.