How to track AI citations over time

Decide what you’re tracking

Pick a fixed set of questions, between five and twenty for most small sites or clients. Choose questions that lead to revenue, not vanity queries. Write them exactly as a customer would ask them, and keep the wording fixed from week to week. Changing the wording changes the answer, which breaks the comparison.

For each question, record which domain counts as yours. If you track a client with several domains or subdomains, decide upfront which ones count.

Record three outcomes, not one

A yes/no metric hides useful detail. Use three outcomes per engine per run:

• Cited: your domain is one of the linked sources.
• Partial: you are named in the answer, but the link goes to another site, such as a review site or a directory.
• Absent: you are neither linked nor named.

Also record who was cited instead

The list of cited domains is the most useful data you’ll collect. Week over week, it shows which sites gained and lost citations for each question. When a competitor appears suddenly, look at what they published. When a review site keeps showing up, that’s a placement worth pursuing. When you drop out, compare the current sources with the last run where you were cited.

Pick a cadence and stick to it

Weekly is a good default. It’s frequent enough to connect a change to something you shipped and infrequent enough to keep costs and effort low. Run checks at a similar time each week. Daily checks mostly measure random variation unless you’re watching a specific launch.

Log every page change, outreach win, or new mention with a date. Without that log, you’ll see movement in the data and have no idea what caused it.

Read trends, not single points

Because answers vary, look at runs in groups. A simple rule: count how many of the last four runs had you cited, per engine. Moving from one in four to three in four is a real change. Bouncing between two and three is probably noise.

Look at engines separately. Perplexity and ChatGPT retrieve differently, and a change that helps on one may not show on the other right away.

A simple client report

If you report this to clients, keep it short. One table and three notes is usually enough:

• Rows: each tracked question.
• Columns: each engine, showing the outcome for the last four weeks (cited, partial, absent).
• Note 1: domains gained or lost this period for the most important questions.
• Note 2: what you shipped this period and when.
• Note 3: the three fixes you plan next, each tied to a specific question and page.

Doing it by hand versus automating

You can run this in a spreadsheet: one tab per week, paste each question into each engine, and record the outcome and the cited domains. It works for a handful of questions, and it’s a good way to learn how the engines behave. It gets tedious quickly, and manual checks can be colored by your own search history and location.

AI Cite Check’s Pro plan automates this. You track a query from any report, and it’s re-checked weekly against Perplexity and ChatGPT. The tracking page shows a cited / partial / absent timeline per engine and the domains gained or lost since the last run. Pro tracks up to 10 queries and Agency tracks 25.

What to do when you lose a citation

First, confirm it over two or three runs. If you’re consistently absent, compare the sources being cited now with the ones from your last cited run. Look for a new competitor page, a refreshed roundup that dropped you, or a change on your own page, such as a redesign that pushed the answer lower or a template change that moved the text into JavaScript. Fix the most likely cause, log the change, and keep watching. Most drops trace back to one of those three things.

Common tracking mistakes

• Changing the question wording between runs.
• Tracking too many low-value questions and ignoring the ones that drive revenue.
• Treating one bad week as a trend.
• Not logging what you changed and when.
• Reporting citation counts without saying who was cited instead.