A Weekly Routine for Tracking Your Citations in ChatGPT

Checking ChatGPT once tells you almost nothing. Citations move, answers vary between runs, and a page that is cited on Monday can vanish by Friday. What helps is a routine you repeat, short enough that you actually do it. This is ours: four steps, about forty minutes, once a week.

Step 1: Fix the prompt panel (10 minutes, first week only)

Tracking citations means tracking a stable set of questions. Pick the prompts that map to money: the ones where a buyer would be comparing options, asking for a recommendation, or checking whether you are legitimate. Tag them by type, which in Promptwatch means organic, brand, or competitor comparison, and by topic.

Resist the urge to add two hundred prompts. A panel you review every week beats a panel you export once. Start with the questions you would pay a salesperson to answer well.

One practical tool here is the query fan-out view. ChatGPT often turns one question into several searches behind the scenes, and the pages it cites tend to answer those sub-questions. Promptwatch shows query fan-outs per model, so you can see what ChatGPT actually searched for, not just what you typed.

Step 2: Read the citation table (15 minutes)

Open Citation analytics and filter to ChatGPT. Look at three things in this order.

First, which of your own URLs were cited, and how many distinct prompts produced those citations. A page cited by many prompts is part of how the model describes your category. A page cited many times by a single prompt is fragile.

Second, which competitor and third-party pages are cited for the same prompts. If a review site or a comparison post keeps appearing where you do not, that is your target list.

Third, the domain split. Citation analytics in Promptwatch separate page-level and domain-level citations, and there are dedicated reports for Reddit, YouTube and LinkedIn. If a forum thread is doing the work your site should be doing, you want to know.

Write down two things: one page that gained citations, and one that lost them. That pairing is the whole weekly story.

Step 3: Check the crawler side (10 minutes)

When a page loses citations, the first question is whether ChatGPT can still reach it. OpenAI uses several bots with different jobs. GPTBot, OAI-SearchBot and ChatGPT-User are the three to know, and they are not interchangeable: the search bot and the user-triggered fetcher are the ones tied to showing a page in an answer.

Promptwatch's Agent Analytics shows real-time, IP-verified AI crawler logs, which means spoofed user agents are filtered out. Open the page that lost citations and look at three columns: were there crawls, did they succeed, and what did errors say? Agent Analytics includes a crawl-to-citation path with a citation rate per page, plus error tracking with failure explanations.

Common findings in this step:

A 403 or timeout the bot hits but human visitors never see.
A page that gets crawled often and never cited, which points to content rather than access.
A page that is cited without recent crawls, which is normal for cached material but worth a re-check next week.

Crawler logs need a connection to your CDN or host. Promptwatch lists Cloudflare, AWS CloudFront, Fastly, Vercel, Netlify, Akamai, Google Cloud CDN, a custom HTTP option, and a WordPress plugin. Note that on brand plans, crawler logs start on Professional; Essential does not include them.

Step 4: Pick one action and write it down (5 minutes)

Not five. One. The Action Board in Promptwatch helps here: a Discovery agent investigates your data on a weekly schedule and suggests up to five evidence-backed actions per run, and you can assign, comment and track them. Choose the one with the clearest evidence and put a name and a date on it.

Typical outcomes from a week's review:

Unblock a crawler path that produced errors.
Refresh a page whose citations are decaying.
Write a missing comparison page where a third party owns the citations.
Add a page to the Page Tracker so you can watch it next week.

What to record in a running log

Keep a plain document with a line per week: date, cited pages gained, cited pages lost, crawler findings, action chosen. After six weeks you will see patterns that no single snapshot shows, such as one prompt family that flips every time you publish, or a competitor page that holds a citation for a month and then drops.

Mistakes that ruin the routine

Changing the prompt panel every week makes trend lines meaningless. Treat the panel like a fixed instrument and change it deliberately, once a quarter.

Reading ChatGPT in isolation misleads too. Promptwatch tracks other engines in the same panel, so check whether a loss is ChatGPT-specific or across the board before you rewrite anything.

Finally, do not skip step 3. Teams love debating content when the real problem is a blocked bot.

Where Promptwatch fits

You can run steps 1 and 2 with any tracker that exposes citations. Steps 3 and 4 are where most tools stop, because crawler logs and an action queue are separate products for most vendors. Promptwatch has both in the same workspace as the citation data, which is why this routine fits in forty minutes instead of three tabs and a spreadsheet. The free Explore plan covers 10 ChatGPT prompts, enough to run the routine for a single week and see whether it earns a slot in your calendar. If it does, Essential at $95/mo adds 50 prompts and a 7-day trial on monthly billing, and Professional is where crawler logs begin.

Do the routine for a month. By week four, you will know which pages ChatGPT trusts, and which ones are only hoped for.