How to track your visibility in ChatGPT with AIS Insights

AIS Insights does this on every model on your plan, and links every number to the answer behind it.

Nike is a demo project we run in a live AIS Insights account, so every number here is real but none of it is client data. Here's what we found asking ChatGPT its 50 prompts every day for three weeks:

• Nike was named in 93% of answers. On its weakest prompt, it was named on 3 days out of 21.
• The answers kept changing. On 41% of days, the brands in an answer were different from the day before.
• ChatGPT read more than it linked. It pulled in about 15 pages per answer and cited about a third of them.
• One review site set the shortlist. RunRepeat was cited in 45% of answers. Reddit supplied 2% of citations.

You can't track something you can't name.

So before the steps, here's one real answer, taken apart. ChatGPT gave it on 30 September 2026 to the prompt best performance running shoes for flat feet in the uk, asked as a signed-out visitor in the UK.

1. The search. Before it wrote a word, ChatGPT searched the web once: best stability running shoes flat feet UK Brooks Adrenaline GTS Asics GT-2000 Hoka Arahi. It had a shortlist before it read a single page. Nike wasn't on it.
2. The names. The answer named ASICS, Nike, On, Topo Athletic, Brooks, HOKA and New Balance. Nike got in through one product, the Nike Structure 26.
3. The product cards. Four products appeared as cards with a price and a rating, before the comparison table. The Nike Structure 26, at £94.49, was the first card.
4. The citations. Five pages were linked in the answer: two RunRepeat guides, two Podiatrist Tested guides and one from Running Bath. The line recommending the Nike shoe cited RunRepeat.
5. Pages pulled in but not cited. Seven more pages were listed as sources and never linked in the text, including Pace & Pavement, NextGait and Prevention.
6. The link tag. Every link carries utm_source=chatgpt.com. That tag is the only part of this answer your analytics will ever see.

On the other 18 days in our three-week sample, the same prompt left Nike out completely.

A mention is a name in the answer. A citation is a link. They move separately, so track both.

Open GA4 and you'll probably find a trickle of visits from chatgpt.com.

That number is the floor, not the total.

When ChatGPT names your brand, one of three things happens. The buyer clicks a citation, and you get a visit tagged chatgpt.com. The buyer searches your name later, and the visit shows up as organic or direct. Or the buyer reads the answer and buys from whoever was on the shortlist. Nothing reaches your site at all.

In the flat feet answer above, ChatGPT put four products in front of the buyer with prices before a single link. Most of what a buyer needed to choose was inside the answer.

Referral traffic counts the people who clicked. Visibility shows whether you were in the answer at all, clicked or not.

Track the clicks you do get anyway. They're the only part that connects to revenue:

• In GA4: go to Reports, then Acquisition, then Traffic acquisition. Set the dimension to Session source and filter for chatgpt.com.
• On your forms: add a free-text "How did you hear about us?" field. It catches the buyers who read your name in ChatGPT and came back another way.

Here's the problem with checking ChatGPT once.

We asked ChatGPT the same 50 prompts every day for 21 days, from 18 September to 8 October 2026. That gave us 1,000 pairs of answers to the same prompt on consecutive days.

On 41% of those days, the brands in the answer were different from the day before.

For Nike, most prompts were rock solid. On 37 of the 50 prompts, ChatGPT named Nike in every answer, every day. The trouble sat in the other 13.

Here's what a single check would have told you on the shakiest ones:

The London prompt flipped between naming Nike and leaving it out eight times in three weeks. Check it on a Monday and you'd think Nike owned it. Check it on the Tuesday and you'd think Nike had a problem.

Some guides say daily tracking is noise and you should check weekly. We'd put it differently. One day is noise. Thirty days in a row is a rate. A weekly check gives you four readings a month on prompts that change their mind several times a week.

One check tells you what ChatGPT said once. A daily series tells you what it usually says.

Here's the process we run, using our Nike demo project as the example. In seven steps.

Step 5: See what ChatGPT cited, and what it read and skipped

ChatGPT pulls in more pages than it links. In Nike's answers, it pulled in about 15 pages per answer and cited about a third of them. The rest were read and dropped.

Nike's own site shows why this matters. ChatGPT pulled in nike.com pages 1,222 times in three weeks and cited them 407 times. That's about 1 in 3. The other two times, ChatGPT had Nike's page in hand and chose someone else's.

AIS Insights lists both kinds of page for every answer. Cited pages are marked in mint, pages it read but didn't cite in grey, and your own pages carry an Owned badge.

ChatGPT · “best performance running shoes for flat feet in the uk”

01podiatristtested.comCited

02runrepeat.comCited

03runrepeat.comCited

04runningbath.co.ukCited

05podiatristtested.comCited

06runrepeat.comRetrieved

07paceandpavement.comRetrieved

08nextgait.comRetrieved

09prevention.comRetrieved

10flawlessshoereviews.comRetrieved

11towerpodiatry.comRetrieved

12runlabgear.co.ukRetrieved

Step 7: Make one change, and track it as a campaign

Tracking is worth nothing if it doesn't change what you do next week.

Pick one prompt from Step 4. Make one change aimed at it: a page, a review, a correction. Add the URLs to a campaign in AIS Insights and watch the citations from the day it goes live.

Then give it time. Nobody can tell you how long ChatGPT takes to notice a change, us included. Track every change for at least 30 days before you judge it.

Here's where Nike's ChatGPT answers came from, over the same three weeks:

Three things stand out.

One review site does the heavy lifting. RunRepeat was cited in 45% of all answers and supplied more than twice as many citations as Nike's own site. If you sell running shoes in the UK, RunRepeat's guides are part of your marketing whether you like it or not.

Your own site is in the mix, but not in charge. nike.com was the second most cited site, at 8.5%. That's 1 citation in 12, for the most famous brand in the category.

Reddit isn't the source everyone says it is. You'll read that Reddit is ChatGPT's favourite source. In Nike's answers it was, until August. In July, Reddit supplied about 20% of ChatGPT's citations. Since mid-September it has supplied about 2%.

This is one category in one market. Yours will look different. That's the point of tracking it.

ChatGPT builds its shortlist on other people's sites. Find the sites that build yours.

Every result points to a different fix. Here's the map:

For Nike's flat feet prompt, that map points at the fourth and fifth rows. The fix isn't a new page on nike.com. It's getting the Nike Structure into the stability-shoe roundups that RunRepeat and Podiatrist Tested publish.

Every tracker samples ChatGPT from the outside. Here's what that means in practice:

• It sees the signed-out answer. We ask the way a new visitor in your market would. A logged-in buyer with months of chat history can get a different answer, and no tracker can see that.
• It sees one market at a time. Nike's numbers are for the UK. Ask from the US and the shortlist changes.
• ChatGPT changes underneath you. OpenAI changes the model behind ChatGPT, and the answers change with it. On 8 October 2026, GPT-6 became the default for free and signed-out users. Any tracker that doesn't record when that happens will show you a trend that isn't yours.
• It asks a sample, not every conversation. Fifty prompts a day is a strong signal. It's not a census.

Trust the tracker that shows you its answers. In AIS Insights, every number opens to the stored answer behind it, so you can check our work.

Whichever tool you pick, ask these before you pay. Here are our answers.

ChatGPT is one model your buyers use. AIS Insights asks the same prompts, on the same day, across every model on your plan, so a gap between two models is a real difference in how they answer.

You can track up to seven models, depending on your plan: ChatGPT, Claude, Google AI Overviews, Google AI Mode, Microsoft Copilot, Perplexity and Gemini. ChatGPT is available on every plan. Starter, Growth and Scale each track three models of your choice, and Enterprise tracks all seven, including Claude.