AI Citation Tracking: What It Measures and How to Do It

Jim Wrubel
1/4/2026
AI citation tracking is the practice of running a fixed set of prompts through AI assistants like ChatGPT, Gemini and Google AI Overviews on a schedule, and recording which brands each answer mentions and which web pages it cites as sources. Done over weeks, it shows your share of AI answers against competitors and the exact pages and publishers AI relies on to talk about your market.
Key Takeaways
- AI citation tracking records two separate things for every answer, the brands it mentions and the pages it cites, and the two often point in different directions.
- A single check of an AI assistant is a sample of one. Answers vary from run to run, so you need the same prompts tracked over time to see a real trend.
- A useful setup has four parts; a prompt set built around buyer questions, the assistants your buyers use, a daily cadence, and a named competitor set.
- The most valuable output isn't the score. It's the list of cited pages, which shows you which publishers to pitch and which of your own pages to fix.
What AI citation tracking measures
People use "citation" loosely, so it helps to separate the measurements a tracker records.
| Measurement | What it is | What it tells you |
|---|---|---|
| Mention | Your brand's name appears in the answer text | Whether AI brings you up at all |
| Citation | The answer links to a source page | Which pages the assistant leaned on |
| Cited URL | The specific page behind a citation | Whether it's your page or someone else's |
| Share of voice | Your mentions as a share of all tracked brands' mentions | Where you stand against competitors |
| Cited publishers | The domains that show up most often as sources | Who shapes AI answers in your category |
Mentions and citations come apart more often than you'd think. An assistant can recommend your brand while citing a review site that lists you. It can also cite your blog post on a topic and then recommend a competitor. If you only count one, you miss half the picture.
The cited URLs are where the work starts. When an answer cites a page, that page was part of what the assistant read before it wrote. Some of those pages are yours. Many belong to publishers, review sites, forums and comparison pages. Those third-party pages that mention your brand are earned media, and they shape AI answers even when the answer never names the page in its text.
Why one-off spot checks mislead
The most common way teams start is typing a question into ChatGPT, reading the answer, and taking a screenshot. That tells you almost nothing.
AI answers are generated fresh each time. Before an assistant answers a question that needs current information, it writes its own web searches (often called grounding searches or fan-out queries), reads the results, and builds an answer from what it found. The search results shift day to day, and the model's wording shifts on every run. The same prompt can name four brands at 9 a.m. and three different ones at noon.
So a single answer is a sample of one. It can make a brand look dominant or absent purely by chance. In our own tracking data, a brand's share of voice moves 5 or more points on about half of all nights with no change to the prompts. A 7-day average moves that much far less often, which is why the weekly trend is the number to report. There's more on reading those numbers in how to measure AI visibility and track it over time.
The fix isn't to check more often by hand. It's to track a fixed set of prompts on a schedule, so every day's answers are comparable with the day before, and to judge the trend instead of any single answer.
How to set up AI citation tracking
![]()
You need four decisions before any numbers mean something.
1. Build a prompt set around real buyer questions
Start with the questions your buyers ask when they don't know your brand yet. Category questions ("what's the best project management tool for construction firms?"), segment questions, use case questions, comparison questions and decision questions ("is X a good fit for a 50-person team?"). Add a few brand questions so you can see how AI describes you when asked directly.
Keep the set focused. A prompt set nobody reads is worse than a small one somebody reviews every week. Our guide to choosing the right prompts to track walks through building one from your brand profile.
2. Pick the assistants your buyers use
ChatGPT, Gemini and Google AI Overviews cover most of the AI answers your buyers see. Each one runs its own searches and picks its own sources, so the same prompt can cite different pages on each. Track them separately, because a brand can lead on one assistant and barely appear on another. Add other assistants when your buyers use them.
3. Run the prompts daily
A daily cadence gives you enough answers to see a trend within a week or two, and it means a change in your citations shows up close to the event that caused it. Mark those events (a new page, a placement, a site change) on your timeline so you can connect cause and effect later.
4. Name your competitors
Share of voice only means something against a defined set of rivals. Start with the brands AI already names alongside you, then add the ones your sales team actually loses deals to.
Daily Prompt Tracking in Spyglasses handles all four. It runs your prompt set every day across ChatGPT, Gemini and Google AI Overviews, records every mention and citation, and builds the trend lines and share of voice for you. You can group prompts into projects tied to a specific SEO or PR effort, so you can show whether that work moved the numbers.
What to do with citation data
Tracking is the proof layer. The value comes from acting on what the citations show you.
Find the sources AI relies on
Sort every cited page by how often it appears for your prompts. A short list of publishers and pages usually accounts for a large share of citations in a category. Those are the pages that shape what AI says about your market. Citation Intelligence groups those sources by type and publisher, separates pages AI cited in its answer from pages it only read, and shows the third-party pages that mention your brand.
Work the earned media
When a publisher is cited often for your category and doesn't mention you, it's a PR target. When it mentions a competitor and not you, it's a priority. Check that AI can actually reach the outlet before you invest in a pitch, because some publishers block AI crawlers. How to connect earned media to AI visibility covers that process in detail.
Fix your own pages
Look at the grounding searches behind the prompts where you're absent. If your page should answer one of those searches and doesn't get cited, the problem is usually that it doesn't rank for the search, or that it ranks but doesn't answer the question in a way the assistant can lift. Rewrite it for the search, then watch the citation data to see whether it starts getting picked up. How to win a target AI grounding search walks through that step.
Watch what AI says about you
Citations show where answers come from. The answer text shows what they say. If AI repeats an outdated claim about your pricing or a product you discontinued, the citation list usually points to the page feeding it. How to monitor what AI says about your brand covers that side of the work.
Getting started
Start with a baseline. Run a set of buyer prompts across ChatGPT, Gemini and Google AI Overviews, note which brands get mentioned and which pages get cited, and compare your share of voice with your top competitors. The AI Visibility Report gives you that snapshot in one run. From there, put the prompts on a daily schedule so you can see the trend, then work the cited sources one page and one publisher at a time. See pricing for the plans that include daily tracking.