Quick Facts
- A citation is a link or named source inside an AI answer. Tracking means recording when it appears.
- Manual checks work well for one person and up to about 20 target questions.
- Run the routine on the first Monday of each month. It takes about 90 minutes for 20 questions.
- Perplexity shows sources in a side panel. Google AI Overviews show links in a "Sources" section.
- Log the URL, the question, the date, and whether you were cited. Four fields are enough to start.
You Can Track AI Citations With a Browser and a Spreadsheet
You don't need a paid dashboard to track AI citations. A browser, a spreadsheet, and a fixed monthly slot will do it. What you need is a repeatable routine, not a tool.
A citation, in this context, is a link or a named source that an AI answer shows to support its reply. Tracking means writing down when that happens, for which question, and on which engine.
Before the first check, confirm these five items:
- One browser profile you only use for checks. This keeps your history clean.
- A spreadsheet with four columns: date, engine, question, cited (yes/no). Add a URL column if you want detail.
- A list of 10 to 20 questions your buyers actually type. Keep them short and specific.
- A calendar block on the first Monday of each month, 90 minutes.
- A note file for screenshots. Name each file with the date and the question number.
If you skip the browser profile, your history mixes personal searches with test searches. That makes it hard to review what you checked. The profile is the cheap fix.
Your Question Set Is the Spine of the Routine
If the questions are vague, the data tells you nothing. Pull them from three places: your sales team's call notes, your site search log, and the "People also ask" boxes on Google for your main topics.
Write each question the way a buyer would type it. "Strong CRM for small teams" is usable. "CRM solutions" is not, because it has no intent. Aim for 10 to 20 questions. More than that and the monthly slot gets too long.
Group the questions by theme, for example pricing, integration, and comparison. This helps you see patterns later. If three pricing questions all miss your citation, that is a signal about your pricing page, not about the engine.
| Question type | Example | What a citation tells you |
|---|---|---|
| Comparison | "Tool A vs Tool B for small teams" | Your comparison page is being read |
| Pricing | "How much does Tool A cost per seat" | Your pricing page is clear enough to quote |
| Definition | "What is a citation in AI search" | Your explainer is seen as a source |
| How-to | "How to set up X in five steps" | Your step list is being reused |
The table is a starting point, not a rule. Adjust the themes to match your own product pages. Keep the set stable so month-to-month changes mean something.
Reading One Perplexity Answer Takes Three Looks
Perplexity is the easiest engine to check by hand because it shows its sources in a panel next to the answer. Open a fresh tab, type one question from your list, and wait for the answer to finish.
Look at three things. First, the numbered citations in the answer text. Second, the source cards in the side panel. Third, the "Related" questions at the bottom, which show what the engine thinks is adjacent.
Write down whether your domain appears in the source cards. If it does, note the exact URL. Sometimes the engine cites your homepage, sometimes a blog post. That difference matters for your next content update.
Perplexity's own help centre describes how sources are displayed and how to read the citation numbers. Treat that page as the reference, not a blog post from 2024.
If your site does not appear, do not stop there. Read the top three sources. Ask what they have that you do not: a table, a dated update, a clear definition. That answer is the start of your next content task.
Google AI Overviews Don't Appear for Every Query
Google AI Overviews are harder to check because they depend on the query, the region, and whether Google decides an overview helps. Google's own documentation on AI features in Search is the place to confirm current behaviour.
When an overview does appear, look for the "Sources" section or the inline links. Record the domain and the URL. If your site is not there, note which domains are. Two or three checks will show you a pattern.
For other engines, the method is the same. Ask the question, read the answer, record the sources. Each engine has its own rules, so keep the engine name in your spreadsheet. Mixing them into one row loses the comparison.
One practical tip: use a private window for each engine check. That reduces the chance that your own browsing history shapes what you see. It is not an ideal control, but it is better than nothing.
A Log Line Tells You Whether the Crawler Could Read the Page
A log is only useful if you can compare it later. The simplest version has one row per question per engine per month. Four columns are enough: date, engine, question, cited. Add a fifth column for the URL if you want to track which page gets cited.
Use yes/no, not a score. A score invites debate about what "partly cited" means. Yes or no is clear, and it lets you count totals at the end of the year.
At the end of each check, count the yes answers per engine and per theme. Write those counts in a small summary block at the top of the sheet. That block is what you show your team, not the raw rows.
Keep the sheet in one place and do not rename columns mid-year. If you change the question set, start a new sheet and note the change date. Otherwise the trend line is meaningless.
Here is where the anatomy walk-through helps. Imagine a log line from your server that shows a visit from an AI crawler. It has a timestamp, an IP, a user agent, a requested URL, and a status code.
The user agent tells you which crawler it is. The requested URL tells you which page it read. The status code tells you whether it got the page or an error. A 200 means it got the page. A 403 means you blocked it, which is a common reason a page never gets cited. The HTTP status code meanings are defined in RFC 9110, the IETF standard for HTTP semantics.
You do not need a tool to read this. Your hosting panel usually has a raw log view. Search for the crawler name, then look at the URLs it requested. If it never requests your pricing page, that page cannot be cited from that crawl.
This is the link between manual checks and server logs. The check tells you whether you were cited. The log tells you whether the crawler could read the page at all. Put the two side by side once a quarter and the picture gets clearer.
Compare Engines Across Three Months, Not One Check
The same question can produce different answers on different days, even on the same engine. That is normal. Treat a single check as a snapshot, not a verdict.
Look for patterns across three months before you draw a conclusion. If Perplexity cites you in month one and not in months two and three, that is a pattern. One miss is noise.
Keep the engine list short. Two or three engines is enough for a manual routine. Adding more engines adds rows, not insight, unless you have a specific reason to watch one of them.
Record the date and time of each check. AI answers can change within a day, so a timestamp helps when you compare notes with a colleague. It also protects you from arguing about a result that has already changed.
The monthly check is the measurement. The work happens between checks. Three changes tend to help, based on how AI answers pull from pages.
First, make one page per question theme. A page that answers "how much does X cost" in a clear table is easier to quote than a page that buries the price in a paragraph. Second, date your pages. A visible "last updated" line helps both readers and crawlers judge freshness.
Third, keep definitions short and on the page. If your page defines a term in one sentence, an AI answer can lift that sentence. If the definition is spread over three paragraphs, it is harder to use.
These are content changes, not tricks. They also help human readers, which is the point. A page that is clear to a buyer is usually clear to an AI answer engine.
Manual Tracking Stops Working When the Routine Takes Over Two Hours
Manual tracking has a ceiling. If you have more than about 20 questions, or more than three engines, the monthly slot gets too long and you start skipping it. That is the signal to change the method.
There is no public figure for the exact ceiling, because it depends on how fast you read and how many engines you check. A practical test: if the routine takes more than two hours, cut the question set or drop an engine.
When you do move to a tool, keep the spreadsheet. Tools change their interfaces and their coverage. Your own log is the one record you control, and it lets you check whether a tool's numbers match what you saw by hand.
For many small teams, manual tracking is enough for a year or more. The value is not in the volume of data. It is in the habit of looking, and in the notes you build up month by month.
Key Facts at a Glance
- Run the check on the first Monday of each month. Block 90 minutes for 20 questions.
- A citation is a link or named source inside an AI answer. Record yes or no, not a score.
- Four columns are enough: date, engine, question, cited. Add a URL column when you need detail.
- Perplexity shows sources in a side panel. Google AI Overviews show a "Sources" section when they appear.
- Server logs tell you whether a crawler could read the page. A 403 status is a common blocker; RFC 9110 defines what each status code means.
- Manual tracking works up to about 20 questions and three engines. Past that, cut the set or change the method.
How do I check if Perplexity cites my site?
Open a fresh tab, type one question from your list, and read the source cards in the side panel. If your domain appears, note the exact URL. Record yes or no in your spreadsheet.
Can I track AI citations for free?
Yes. A browser, a spreadsheet, and a calendar block are enough. The cost is your time, roughly 90 minutes a month for 20 questions.
How often should I manually check AI citations?
Once a month is a practical rhythm. A weekly check adds rows without adding insight, unless you are testing a specific content change.
What metrics should I track for AI citations?
Start with four: date, engine, question, and whether you were cited. Add the cited URL when you want to know which page is doing the work.
Why doesn't my site appear in AI citations?
Common reasons include a blocked crawler, a page that is hard to quote, or a question set that does not match your content. Check the server log for 403 errors first.