# How to track your ChatGPT rankings over time, in 6 steps

> How to track ChatGPT rankings over time: which prompts to pick, how often to ask, what to record per answer, and the one number worth reporting weekly.

Canonical: https://picked.so/blog/how-to-track-chatgpt-rankings
Last updated: 2026-10-25

To track your rankings in ChatGPT over time, ask the same 10 to 20 buyer questions on a fixed schedule, from the same country, in a chat with no memory, and record three things for every answer: whether it names you, in what position, and which pages it cites. Then report one number each week: the share of answers that name you.

## Why one ChatGPT check tells you nothing

A ChatGPT screenshot is an anecdote, not a metric. The same question can get a different answer on the next run, and ChatGPT changes what it searches for depending on who asks and from where.

OpenAI says so on its [help page about searching the web](https://help.openai.com/en/articles/9237897-chatgpt-search): ChatGPT estimates your general location from your IP address and may pass it to its search providers, and with memory on it may use your saved memories when it rewrites your question into searches. A colleague in another country, or with a different chat history, can get a different list of brands for the same words.

The sources move too. In answers we recorded for the same prompts on two days in September 2026, **32 of the 69 pages ChatGPT cited on the first day were cited again on the second**. Fewer than half. Ask ChatGPT the same question three times and you'll meet three different market leaders.

So "where do I rank in ChatGPT?" has no single answer. "How often does ChatGPT name me, across many answers, this week compared with last week?" does.

## What should you record in each answer?

ChatGPT has no page two. There are three things worth recording in an answer, and they move separately:

| What to record | What it means | How to note it |
| --- | --- | --- |
| **Named** | The answer mentions your brand by name | Yes or no |
| **Position** | Where your brand sits among the brands the answer names, counting from the first | 1, 2, 3... or blank if not named |
| **Cited** | A page from your site is one of the answer's sources | The URL, or blank |

Named matters most. A buyer who reads an answer that lists four tools and not yours has no reason to go looking for a fifth. Position matters among the named: first in a list of six reads differently from sixth. Cited tells you which of your pages ChatGPT trusts, and which pages it reads instead of yours.

## How to track your ChatGPT rankings over time: 6 steps

### 1. Pick 10 to 20 buyer questions and freeze them

Write the questions a buyer asks before they know your name: "best CRM for a 10-person agency", "alternatives to [your competitor]", "how much does [your category] cost". Good sources are your sales calls, the questions prospects email you, and the queries in Search Console that are phrased as questions.

Two rules. Don't put your brand in the question, or ChatGPT will name you every time and the number means nothing. And don't edit the questions once tracking starts: change one word and you're measuring a different question. If you need a new one, add it as a new row and keep the old.

### 2. Fix the conditions you ask in

Most of the noise in a manual check comes from the person asking, not from ChatGPT. Keep these the same every time:

- **No memory.** Ask signed out (OpenAI says people who aren't signed in can use web search), or in a [temporary chat set to Unpersonalized](https://help.openai.com/en/articles/8914046-temporary-chat-faq), which doesn't use memory or custom instructions. Choose it before the first message; it can't be changed after.
- **One country.** ChatGPT reads your location from your IP, and a VPN changes it. Ask from the market your buyers are in. If you sell in several, track each as its own set.
- **Search on.** ChatGPT may search on its own when a question needs current information, but not always. Pick Search from the tools menu, so every answer is a search answer.

Write the conditions at the top of the sheet. Whoever checks next month needs to repeat them.

### 3. Ask each question 3 times per check

One answer per question is a coin toss. Ask each question 3 times in fresh chats and record all three. [Aleyda Solis's AI search checklist](https://www.aleydasolis.com/en/ai-search/ai-search-optimization-checklist/) suggests running each core prompt 3 to 5 times in a fixed 24 to 72 hour window, which is the same idea.

For a manual check, once a week is enough. Same day, roughly the same time, so a week compares with a week.

### 4. Record every answer in the same columns

One row per answer. These columns are enough to compute everything that follows:

| Column | Example |
| --- | --- |
| Date | 2026-10-25 |
| Question | best CRM for a 10-person agency |
| Run | 1, 2 or 3 |
| Named | yes |
| Position | 3 |
| Brands named, in order | HubSpot, Pipedrive, you, Zoho |
| Your page cited | yoursite.com/pricing |
| Other pages cited | g2.com/..., zapier.com/blog/... |

The "brands named" column is the one people skip and later wish they hadn't. It gives you the same numbers for your competitors for free.

### 5. Turn the rows into one number a week

The number to report is the **share of answers that name you**: answers that name you, divided by all answers that week.

10 questions asked 3 times is 30 answers. If 9 of them name you, your share is 30%. Work out the same share for your three main competitors from the "brands named" column, and you have a table a client or a head of growth can read in ten seconds.

Two more numbers are worth a line each:

- **Average position when named.** Only over the answers that name you, or a missing week drags it down for the wrong reason.
- **Answers citing your site.** How many of the week's answers link one of your pages, and which pages.

Don't react to one week. With 30 answers, a few answers either way moves the share by 10 points without anything changing. Look at four-week trends, and per question: a share that falls on one question after a competitor publishes a comparison page is something you can act on.

### 6. Watch the visits as well as the answers

The answers show whether you're named. Two other signals show whether it brings anyone:

- **Referrals in your analytics.** ChatGPT adds `utm_source=chatgpt.com` to its referral links, OpenAI says in its [publisher FAQ](https://help.openai.com/en/articles/12627856-publishers-and-developers-faq). Filter your analytics for that source and you see which pages people open from ChatGPT answers.
- **`ChatGPT-User` in your server logs.** It's the bot that opens a page because someone asked ChatGPT a question right then. A rise in its hits on a page means that page is being read for live answers, whether or not anyone clicks.

## Why not use the API instead?

Some guides suggest scripting the check with OpenAI's API, because it's cheap and repeatable. It is, but it measures a different product. A plain API call answers from what the model learned in training, with no web search, unless you add OpenAI's web search tool. With the tool, you still aren't looking at the chatgpt.com interface your buyers use.

If you automate through the API, label the numbers "API with web search" and don't compare them with what people see in the app. The point of tracking is to know what your buyer reads, so record from the interface they read it in wherever you can.

## When the spreadsheet stops working

The manual method works up to a point. 10 questions, 3 runs, once a week is 30 answers and about an hour of copying. It stops working when:

- you want daily numbers, because a weekly check misses a drop for six days;
- you track more than one brand or client;
- you need other models too: Gemini, Perplexity, Claude and Google's AI answers each name different brands for the same question;
- someone has to read 30 answers for brand names every week without missing one.

That's what a [ChatGPT rank tracker](https://picked.so/chatgpt-seo) automates. Picked, an [AI visibility tool](https://picked.so/), asks your prompts every day on ChatGPT and the other models (Claude less often), records the answer from ChatGPT's own interface, not the API, and shows the share of answers that name you and each competitor, the position, and the pages each answer cited. To see where you stand before tracking anything, run the [free AI visibility checker](https://picked.so/free-ai-visibility-checker).

## When you don't need to track it

If your buyers don't ask AI tools what to buy, there's nothing to track: a walk-in trade or a market that buys by phone is better served by local SEO. And if you have fewer than five questions that matter and nobody to report to, a careful manual check once a month is enough. Tracking pays off when someone asks "are we in ChatGPT?" every week and needs a number, not a screenshot.

Tracking also doesn't change the answer by itself. It tells you which questions you lose and which pages ChatGPT reads instead. Our guide on [how to rank in ChatGPT](https://picked.so/blog/how-to-rank-in-chatgpt) covers what to do about them.

## How we checked

- **How ChatGPT searches, location, memory and temporary chats:** OpenAI's help center ([searching the web](https://help.openai.com/en/articles/9237897-chatgpt-search), [temporary chat](https://help.openai.com/en/articles/8914046-temporary-chat-faq)), read October 25, 2026.
- **Day-to-day citation changes (32 of 69):** ChatGPT answers for 9 prompts on September 27 and 28, 2026, the chatgpt.com interface, US. Small sample, two niches; the full breakdown is in our [ChatGPT citations post](https://picked.so/blog/chatgpt-citations).
- **`utm_source=chatgpt.com` and `ChatGPT-User`:** OpenAI's publisher FAQ and crawler docs.
- **Google's top results** for "how to track chatgpt rankings over time": the US first page, October 25, 2026.

## Frequently asked questions

### Can I track my ChatGPT rankings for free?

Yes, by hand. Pick 10 buyer questions, ask each 3 times a week in a temporary chat set to Unpersonalized, from the same country, and log whether you're named, your position and the pages cited in a spreadsheet. It costs about an hour a week. Paid trackers do the same daily and across more AI models.

### How often should I check my ChatGPT rankings?

Weekly for a manual check, daily if you use a tool. ChatGPT's answers and sources change from day to day, so a single check is a sample, not a ranking. Whatever the cadence, compare four-week trends rather than reacting to one week's number.

### Does ChatGPT have rankings like Google?

Not in the same way. There's no list of ten positions. Each answer names some brands, in an order, and cites some pages. "Ranking" in ChatGPT means how often you're named across many answers, where you sit among the brands named, and whether your pages are among the sources.

### Why does ChatGPT give my colleague a different answer?

Because the conditions differ. ChatGPT estimates location from the IP address, may use saved memories when it builds its searches, and gives a different answer on each run anyway. To compare like with like, ask from the same country, signed out or in an unpersonalized temporary chat, and ask several times.

### Can Google Analytics show traffic from ChatGPT?

Yes. ChatGPT adds `utm_source=chatgpt.com` to its referral links, so visits from them show up under that source in Google Analytics and most other tools. It counts people who clicked, not answers that named you without a click, so track both.

### What's a good share of answers in ChatGPT?

There's no public benchmark that holds across categories. Compare yourself with the competitors named in the same answers instead: if the leader is named in 60% of answers and you're in 15%, that gap is your number to close, question by question.