How to measure your brand's visibility in ChatGPT

When someone asks ChatGPT "what's the best Webflow agency for a SaaS startup?", the answer is a paragraph, not a list of ten blue links. Your brand is either in that paragraph or it isn't, and it may or may not come with a link to your site. Measuring that is different from measuring rankings, and most of the mistakes we see come from treating it the same way.

This guide explains a method you can run yourself, by hand or with a tool.

1. Start from buyer questions, not keywords

Write down 10 to 15 questions a buyer would ask an AI assistant before choosing a provider like you. Good questions are specific and have commercial intent:

• "best Webflow agency for SaaS startups"
• "Framer or Webflow for a startup website"
• "how much does a Webflow website cost"

Avoid questions that only your brand can answer ("what does Acme do?"). They tell you nothing about whether you're recommended when the buyer doesn't know you yet.

Keep each question in one language. "Best CRM for agencies" and "mejor CRM para agencias" are different questions with different answers, and should be tracked separately.

2. Separate "mentioned" from "cited"

For each answer, record two things:

• Mentioned: your brand name (or an alias) appears in the text.
• Cited: your website appears as a source link.

They behave differently. You can be mentioned without being cited (the assistant knows you but links elsewhere), and cited without being mentioned by name (your page is used as a source for a general statement). Both matter, for different reasons.

Also record which competitors are mentioned and which URLs are cited, including pages that belong to neither you nor your competitors. That last group is often the most useful part of the data.

3. Ask each question more than once

This is the step most people skip, and it's the one that matters most. AI answers are not stable. In public research, SparkToro and Gumshoe ran 12 prompts about 3,000 times and found that the chance of getting the same list of brands twice was less than 1 in 100 (sparktoro.com/blog).

So a single answer is an anecdote. If you check once, see your brand, and celebrate, you may simply have been lucky. If you check once and don't see it, you may have been unlucky.

A reasonable routine is to ask each question three times per engine, every week, in fresh sessions without personal history, and pool the results over a few weeks.

4. Report a share with a range, not a position

With repeated samples, the honest metric is a share: "you were mentioned in 7 of 12 answers". Because the sample is small, add a confidence range. A Wilson interval works well for small samples; at 90% confidence, 7 of 12 is roughly 36% to 78%.

That range looks wide, and it should. It tells you that a jump from 58% to 67% next week is probably noise, while going from 0 of 12 to 8 of 12 probably isn't.

Avoid reporting a "position in ChatGPT". The order in which brands appear changes too much between runs to be a meaningful number.

5. Look at who gets cited instead of you

Once you know where you're missing, look at the sources the engines cite for those questions. Two patterns show up again and again:

• Listicles carry a lot of weight. Ahrefs found that 43.8% of the pages ChatGPT cited for "best X" queries were blog listicles (ahrefs.com/blog).
• Most of those listicles are third-party pages. In Peec AI's research, third-party listicles received about 81% of listicle citations (peec.ai/blog).

This changes what "fixing" a gap means. For some questions, the best move is a page on your own site that answers the question clearly, with sources. For others, it's getting onto the third-party pages the engines already trust.

6. Act on a few gaps, then measure again

Pick the questions where competitors are mentioned or cited and you aren't, and where you have something genuine to say. For each one, decide between:

• On-site: a focused article that answers the question, cites its sources and includes a short FAQ.
• Off-site: the listicles, reviews or directories the engines cite, where you could reasonably be included.

Then keep measuring the same questions. The cleanest signal that an action worked is the URL you published showing up as a cited source. Changes in the overall mention rate are useful too, but read them with their range: only changes larger than the noise count.

A simple spreadsheet template

If you want to start by hand, one row per answer is enough:

Week	Question	Engine	Sample	Mentioned	Cited	Competitors mentioned	Cited URLs
38	best Webflow agency for SaaS	ChatGPT	1	yes	no	Studio A, Studio B	studio-a.com/…, listicle.example/…

After a few weeks you can compute shares per question and engine and see where the gaps are.

FAQ

How many questions should I track?

Ten to fifteen well-chosen buyer questions are enough for a small business. More questions with fewer samples each will give you noisier numbers, not better ones.

Does ChatGPT give everyone the same answer?

No. Answers vary between runs and can be personalised for logged-in users. Measuring in fresh sessions and repeating each question is how you get a number that reflects the general tendency.

Can I guarantee that ChatGPT will cite my site?

No. You can make it more likely by publishing clear, sourced answers to the questions buyers ask and by being present on the third-party pages engines already cite, and you can measure whether that changes anything.