Prompt tracking vs citation tracking: which question each one answers
Prompt tracking asks whether engines name you. Citation tracking asks which pages they rely on. Why you need both, how they disagree, and how to read them together.
Teams new to AI visibility often use "tracking" as one word. It's two different activities, and mixing them up leads to bad reports. One asks whether an engine names your brand. The other asks which pages the engine used to say it. They can move in opposite directions, and when they do, you've learned something.
Prompt tracking: the mention question
Prompt tracking runs a fixed set of questions through AI engines on a schedule and records the answers. From those you get Visibility Score, Share of Voice, sentiment and position in answer. In Promptwatch, monitors run prompt panels daily per model, country and language.
It answers: "Out of the questions we care about, how often are we named, how prominently and in what tone, compared with competitors?"
That's a brand-level view. It's the first thing a CMO wants, and it's useful.
Citation tracking: the source question
Citation tracking records which URLs engines cite in their answers, at page and domain level, with positions. Promptwatch's citation analytics adds a Page Tracker for specific URLs and Citation Trends for timing.
It answers: "Which of our pages do engines rely on, for which prompts, and which pages do they use instead of ours?"
That's a content-level view, and it's what the content team needs.
Where they agree
When both rise, the story is clean: more prompts name you, and your pages are cited more. You've probably improved something real.
When both fall, look for a shared cause. A crawler block, a site migration or a competitor launch might do it.
Where they disagree
This is where it gets interesting.
Mentions up, citations flat. Engines name you more, but not because of your pages. Third-party sources are probably carrying you. That's good news with a risk: you depend on pages you don't control. Check offsite mentions.
Mentions flat, citations up. Your pages are cited more, yet you aren't being named more. The engine uses your content but credits the answer differently or doesn't surface your brand. Look at how your pages present the brand name and what the answers say.
Mentions down, citations flat. You're cited but not named. Examine what the answers say. Possibly a competitor is recommended while your page is used as background.
Mentions flat, citations down. A warning sign. Your pages are being replaced as sources, and mentions may follow. Check crawler logs and competitor pages.
What each can't do
Prompt tracking can't tell you why. Citation tracking can't tell you whether the answers helped your brand. A third thing, crawler logs, can tell you whether bots can reach you at all. And visitor analytics tells you whether any of this produced visits.
No single measure is enough, which is the reason a platform that combines them has an advantage over a pure tracker. Promptwatch combines prompt tracking, citation analytics, Agent Analytics crawler logs and visitor analytics in one place, and 360 Insights joins crawler logs, citation clicks and AI-referred visits per page.
How we report them together
Our monthly report puts them side by side:
| Question | Measure | Source |
|---|---|---|
| Are we named? | Visibility, Share of Voice | Prompt tracking |
| Which pages are used? | Page and domain citations | Citation analytics |
| Can bots reach us? | Fetches and errors | Crawler logs |
| Did it bring visits? | AI-referred visits | Visitor analytics |
For the layout of the full report, see what we send clients each month.
Choosing the right one for the decision
- Deciding whether to invest in AI visibility: prompt tracking.
- Deciding which pages to rewrite: citation tracking.
- Deciding whether to fix infrastructure: crawler logs.
- Deciding whether it's worth the budget: visitor analytics.
Using the wrong measure for a decision is a frequent source of wasted effort. A team that watches only mentions might rewrite pages that were never the problem.
Common mistakes
- Reporting mention rates without the prompt panel definition.
- Changing the panel mid-quarter.
- Treating citations from one engine as proof for all.
- Counting a brand name in text as a citation.
- Comparing your numbers with another vendor's, which use different panels.
Keep definitions written down, and put them in the report's footer.
Reading together, a quick method
Once a month, make a two-by-two for your top ten prompts. One axis is "named / not named". The other is "our page cited / not cited". Each quadrant tells you what to do:
- Named and cited: protect.
- Named, not cited: strengthen your own page.
- Cited, not named: examine how the answer frames you.
- Neither: it's a gap, so check access, then content.
That fits on a slide and drives real work.
What we'd recommend
Track both. Promptwatch lets you see mentions and citations in the same workspace, with the crawler and traffic data alongside. Log what you find on the Action Board so each quadrant becomes tasks with owners.
A team that knows the difference between "named" and "cited" makes noticeably better decisions than one that doesn't.