You don't need a paid rank tracker to see whether AI answer engines cite your site. You need a fixed date, a short list of prompts, and a spreadsheet. An AI citation is when an answer engine lists your page as a source under its reply. This routine runs on the first Monday of each month and takes about 90 minutes.
Run the check on the first Monday of the month, in the same order every time
Block 90 minutes. Open your spreadsheet, your browser, and the Perplexity answer page side by side. Keep the order fixed so the conditions match every month.
Step 1: pull out the same ten prompts you used last month. Step 2: run each prompt in Perplexity and copy the source list. Step 3: paste the citation URL and the answer text into the sheet. Step 4: score each hit on the rubric below. Step 5: write down what changed since last month.
If you skip a month, the next run won't compare cleanly. Four prompts on time beats ten prompts late, because you're tracking a trend, not taking a snapshot.
A citation, a mention and a link are three different things, so log them separately
A citation is a source the engine lists under the answer, usually as a numbered card or footnote. A mention is the engine naming your brand in the answer text without listing your page. A link is any URL in the answer body, whether or not it appears in the source list.
They behave differently. A citation means the engine's retrieval step pulled your page and the generator picked it. A mention may come from training data or from another page that names you. A body link may be a citation or just an inline reference.
Give each one its own column. Mixing them hides the thing you actually want to know: is the engine fetching your page, or just remembering your name?
Build the sheet before you start, and keep it narrow
Make one tab per engine and one summary tab. The engine tabs hold raw rows. The summary tab counts citations, mentions and links per prompt per month. Keep the prompt text in column A so you can copy it straight back into the engine.
Here's the column layout to start with. It's deliberately small, because a wide sheet gets abandoned by month three.
| Column | What to enter | Example |
|---|---|---|
| A: Prompt | The exact question text, unchanged | strong b2b seo agency for saas |
| B: Date run | YYYY-MM-DD, first Monday | 2026-10-05 |
| C: Engine | Perplexity, ChatGPT, Copilot, Gemini | Perplexity |
| D: Our URL cited? | Yes / No | Yes |
| E: Citation position | 1, 2, 3 ... or blank | 2 |
| F: Answer snippet | First 200 characters of the answer | "For SaaS teams the usual picks are..." |
| G: Competitor URLs | Other cited domains, comma separated | example.com, other.com |
| H: Change since last month | New / Lost / Same / Changed position | New |
| I: Score | 1 to 5, see rubric below | 4 |
Copy that table into a blank sheet and you have a working tracker. There's no template file to download; the layout above is the template.
Check Perplexity by reading the source cards, not the answer text
Open Perplexity in a fresh tab. Type the prompt exactly as written. Wait for the answer to finish streaming, because sources can appear late. Perplexity's own help centre describes how it lists sources under each answer; the cards sit next to or below the reply depending on the layout.
Click each source card. Note the domain and whether it's your page. Copy the URL into column D or G. Screenshot the answer and save it with the date in the filename, so you have a record if the answer changes next month.
This is the manual citation check in its simplest form. It's slow, but it's honest: you see the answer the same way a buyer would.
Use the same browser profile each month. Logged-in and logged-out sessions can return different answers, and you want the comparison to be apples to apples.
Run the same prompts on other engines the same day, and log each engine separately
Perplexity is the easiest to check because sources show by default. ChatGPT with browsing shows sources in a panel; Copilot shows footnotes; Gemini shows links under the answer. Each engine has its own display, so keep them in separate tabs.
Run the same ten prompts in each engine on the same day. Don't compare across days, because answers move. One sheet, one date, four engines is the cleanest setup.
Not every engine shows sources on every answer. If a reply has no source list, log it as "no sources shown" rather than "not cited". Those are different facts and you'll want them separated later.
Score each row 1 to 5, then work the top five first
Not every citation is worth chasing. A score from 1 to 5 helps you decide where to spend time. The rubric below is a starting point, not a standard.
- 5 — cited on a prompt that maps to a service you sell, in position 1 or 2.
- 4 — cited on a relevant prompt, any position.
- 3 — mentioned by name but not cited.
- 2 — competitor cited, you absent, on a relevant prompt.
- 1 — no sources shown, or the prompt is off-topic for you.
Sort the month's rows by score, highest first. The top five rows are your work list for the next four weeks. Everything below is background noise until the top five are handled.
A score of 2 is the most useful row in the sheet. It tells you a competitor is being pulled for a question you should own. That's a concrete gap, not a vague worry.
Compare month to month, because single-prompt flips are noise
AI answers move. The same prompt can return a different source list a week later, even with no change on your side. That's normal and doesn't mean your tracking is broken.
To measure change, compare month to month, not day to day. Count how many of the ten prompts cited you this month versus last month. A move from two to four is a signal; a single prompt flipping is noise.
Keep a short note column for anything that changed on your site between runs, like a new page or a rewrite. That way you can connect a citation gain to a cause, or rule it out. There's no public baseline for how often citations flip, so treat your own history as the baseline.
Shrink the routine before it gets skipped
Most manual trackers die in month three. The fix is to cut the routine down before you skip it. Drop to five prompts if ten feels heavy. A five-prompt routine that runs every month beats a ten-prompt routine that runs twice a year.
Put the first Monday in the calendar as a recurring event with the sheet link in the description. Open the sheet the day before and check the prompts are still the ones you care about. Swap one prompt only if it has gone stale, and note the swap so the trend line stays honest.
At the end of each quarter, read the summary tab, not the raw rows. The summary shows whether citations are growing, flat, or slipping. That's the number you report.
Key Facts at a Glance
- A citation is a source the engine lists under the answer; a mention is the brand named in text; a link is any URL in the body.
- Run the check on the first Monday of each month, same prompts, same browser profile, same day across engines.
- Ten prompts across four engines is about 90 minutes of manual work.
- Score each row 1 to 5 and work the top five rows first.
- Compare month to month, not day to day, because AI answers move.
How do I check if Perplexity cited my site?
Open Perplexity in a fresh tab, type your prompt exactly as written, wait for the answer to finish, then look at the source cards next to or below the answer. Click each card to confirm the domain. Copy the URL into your sheet and screenshot the answer with the date in the filename.
How often should I manually track AI citations?
Once a month, on a fixed date, is enough for most B2B teams. Weekly checks add noise because answers move on their own. If you need faster feedback, add a second run mid-month but keep the monthly run as the one you compare.
What is the difference between an AI citation and a mention?
A citation is a source the engine lists under the answer, usually as a numbered card or footnote. A mention is the engine naming your brand in the answer text without listing your page as a source. Log them in separate columns because they mean different things.
Can I track AI citations for free without tools?
Yes. You need a browser, a spreadsheet, and a fixed date. The engines show their sources on the page, so the work is reading and copying. The cost is time, not money: about 90 minutes a month for ten prompts across four engines.
Why do AI citations change between queries?
Answer engines retrieve sources at the moment of the query, so the pool of pages they pull from can shift as the web changes and as the engine updates its index. The same prompt on a different day can return a different source list. That's why you compare month to month and keep the prompt text unchanged.