How to track your citations in ChatGPT: a weekly audit you can finish in an hour
A repeatable weekly routine for tracking which of your pages ChatGPT cites: prompt panel, source log, citation rate, and the point where a tool should take over from a spreadsheet.
Quick verdict
You can track ChatGPT citations by hand, and for the first month you probably should. It teaches you what a citation looks like and which pages earn them. After that, the weekly audit below stops fitting in an hour, and a platform with citation analytics and a Page Tracker does the same work daily. Promptwatch is the one we would automate it with, mainly because it links citations to crawler logs and visits.
Citation, mention, and fetch are three different things
Before tracking anything, separate the terms.
A mention is ChatGPT naming your brand in text. A citation is a link to one of your pages in the sources shown with the answer. A fetch is ChatGPT's crawler or agent reading your page, which may or may not end in a citation.
Many reports blur these. A brand can be mentioned a hundred times and cited zero, because the model learned about it from other sites. A page can be fetched constantly and never cited, because the content does not answer the question. Tracking citations means tracking the second one specifically.
The weekly audit, step by step
1. Fix a prompt panel
Choose 20 prompts and keep them stable for at least a quarter. Changing prompts weekly makes trends meaningless. Include prompts where you want to be cited: category questions, comparison questions, and how-to questions your content answers.
Tag each prompt by type. Brand prompts, competitor comparisons, and organic category prompts behave differently, and mixing them in one average hides the movement.
2. Run each prompt cleanly
Use a logged-out or temporary session so memory does not tilt the answer. Make sure web search is actually used, since a citation only appears when ChatGPT searches. Run each prompt three times. If a source shows up in one of three runs, count it as unstable rather than as a win.
3. Log every cited URL
For each run, record the full URL of every source shown, not only yours. A simple sheet with columns for prompt, run, cited URL, domain, and whether it is yours is enough.
The competitor and third-party rows are as valuable as your own. When the same Reddit thread or comparison article appears in six of twenty prompts, that is a source you need to be present in.
4. Compute a citation rate
Divide the number of runs citing any of your pages by total runs. Do it per prompt type. Then do it per page: which of your URLs earn citations, and which never do? A short list of repeat performers usually emerges, and it tells you what format ChatGPT prefers from you.
5. Write one sentence of diagnosis per change
Each week, pick the biggest gain and the biggest loss. Write one sentence on the likely cause. For a loss, check whether a new source replaced you, whether your page changed, or whether the page became unreachable to crawlers.
Where the manual audit breaks
It breaks in three predictable places.
Volume is the first. Twenty prompts at three runs is sixty checks a week, and adding a second country doubles it. The second is explanation: your sheet shows that a page lost its citation, but not whether a bot stopped fetching it. The third is attribution. A citation matters because it can send a visit, and a spreadsheet cannot show you the visit.
What to automate, and with what
| Manual step | Automated equivalent in Promptwatch |
|---|---|
| Fixed prompt panel | Prompt tracking with daily runs per model, country, and language |
| Logging cited URLs | Citation analytics at page and domain level, with URL positions |
| Watching your key pages | Page Tracker for specific URLs, yours or a competitor's |
| Spotting repeat third-party sources | Offsite mentions plus Reddit, YouTube, and LinkedIn citation reports |
| Per-page citation rate | Citation Trends, with first cited, ramp-up days, peak rate, and decay half-life |
| Why a page lost its citation | Agent Analytics crawler logs and the crawl-to-citation path |
| Did it send traffic | Visitor analytics with conversions |
Citation Trends deserves a closer look because a manual sheet cannot produce it. The decay half-life per page tells you how long a citation lasts before it fades, which turns "refresh content sometimes" into a schedule.
Plan notes before you buy
Start with Explore, which is free: 1 project, 10 prompts, 310 responses, ChatGPT only. That is enough to load your best ten prompts and compare the output with your spreadsheet. If the numbers match, you have validated the tool on your own data.
Essential at $95 a month adds 50 prompts, 6,000 responses, and visitor events, with a 7-day trial on monthly billing. Crawler logs are not on Essential. They start on Professional at $245 a month with 25M logs, so if your diagnosis step needs bot data, budget for that tier.
Who should pick which
| Situation | Approach |
|---|---|
| Under 20 prompts, one market, curious | Stay manual, or use Promptwatch Explore |
| 50 or more prompts, monthly reporting | Essential, with citation analytics |
| Losing citations and need to know why | Professional, for crawler logs |
| Several markets and a content team | Business, with state and city tracking and the Content Agent |
| Agency reporting for clients | Agency plans with unlimited prompts and projects |
If you want to compare other trackers on citation depth first, our ranking page lists them, and the Scrunch AI review covers one that tracks four LLMs on its Core plan.
The habit that matters
Whatever you use, the useful output is a short weekly note: what gained a citation, what lost one, and one action. Tools that produce charts but no action leave the note unwritten. Promptwatch's Action Board turns that note into a prioritized list, with a Discovery agent suggesting up to five evidence-backed actions per weekly run, which is a decent default when nobody on the team has an hour to spare.
Run the manual audit for four weeks. Then run Explore on the same prompts for a week. If the tool saves you the hour and adds the cause behind each loss, the choice makes itself.