You don't need a paid dashboard to track AI citations. An AI citation is a link or brand mention that an answer engine shows as a source for its answer. You can check it by hand once a month, on the first Monday, and learn more than many tools will tell you.
The routine takes about 90 minutes for a small site. You ask a fixed set of questions in each engine, write down what comes back, and compare it to last month. The pattern matters more than any single hit.
A fixed monthly date beats a daily check
Answer engines change their answers often. A daily check burns time and shows noise. A monthly check is slow enough to show real movement and fast enough to catch a drop before a quarter ends. Pick a fixed date so nobody has to ask when you last did it.
The first Monday also gives you a clean month of content to test. If you published something in the first week of the month, the next check shows whether it entered any answer set. That's a simple before-and-after you can actually read.
Write the date at the top of your sheet before you start. If you change your question list later, the old rows still make sense because each one carries its own date.
A short, frozen question list is what makes months comparable
Your question list is the set of prompts you ask every engine, every month. Keep it short: 10 to 20 questions is enough for a small B2B site. Each question should be one a real buyer would type, not a keyword you want to rank for.
Group the questions into three buckets. First, category questions ("what is a [your category]?"); second, comparison questions ("[your product] vs [competitor]"); third, problem questions ("how do I fix [the problem your product solves]?"). That mix shows where you appear and where you don't.
Don't edit the list mid-month. If a question stops being useful, retire it at the next monthly check and note the change in a comment column. A stable list is the only way to compare months.
- Keep 10 to 20 questions, no more.
- Use buyer language, not internal jargon.
- Write the exact prompt in a cell, word for word.
- Freeze the list for at least three months.
Perplexity shows its sources, so start there
Perplexity is the easiest engine to check because it shows numbered sources under each answer. Open a fresh thread for each question so the engine doesn't carry context from your last prompt. Paste the question exactly as written and read the answer, then the source list.
Perplexity's help centre describes how it cites sources inline and in a numbered list below the answer. That list is your record. Note whether your domain appears, which position it holds (1 is the top source), and which page was cited. The page matters more than the domain: a homepage citation and a deep guide citation tell different stories.
Perplexity sometimes cites a page you didn't expect, like a PDF or an old blog post. Write down the exact URL. That tells you which content is actually citation-worthy, and it often isn't the page you spent the most time on.
Answers can shift between runs as the index refreshes. If you get a surprising result, rerun the same prompt in a new thread once. If it changes, note both results.
ChatGPT and Gemini need a follow-up question to show sources
ChatGPT with browsing enabled can show sources, but the display varies by plan and by whether browsing was used. The reliable method is to ask the question, then ask a follow-up: "Which sources did you use for that answer?" Record whatever it names. If it names nothing, record "no sources shown" rather than guessing.
Gemini shows source links under answers when it uses web results. The pattern is similar: paste the question, read the answer, then read the linked sources. Note the position and the exact URL, same as Perplexity.
Don't expect every engine to cite you the same way. An engine that answers from its training data may mention your brand without a link. That is still a citation in the loose sense, so log it as "brand mention, no link" and keep it separate from linked citations in your totals.
One row per question per engine per month, with fixed columns
A citation log is a spreadsheet with one row per question per engine per month. Keep the columns fixed. The table below is the minimum you need; add columns only if you will actually fill them every month.
| Column | What goes in it | Example |
|---|---|---|
| Date | Day you ran the check | 2026-10-05 |
| Engine | Perplexity, ChatGPT or Gemini | Perplexity |
| Question ID | Short code from your fixed list | Q07 |
| Cited? | Yes, no, or brand mention only | Yes |
| Position | Source number in the list | 3 |
| URL cited | Exact page the engine linked | /guides/setup |
| Context | One line on how the source was used | Cited for setup steps |
| Notes | Anything odd, like a changed answer | Rerun gave same result |
The context column is the one people skip, and it's the one that pays. "Cited for setup steps" tells you the page is doing a specific job. "Cited, unclear why" tells you to read the answer again next month.
Keep one tab per month and one summary tab that pulls counts. The summary needs three numbers: total linked citations, total brand mentions, and the number of questions where you appear in the top three sources. Those three move at different speeds, and watching them together stops you overreacting to one bad month.
Read each log line in three parts
Take one row: Perplexity, Q07, cited, position 3, URL /guides/setup, context "cited for setup steps". Read it in three parts. The engine column tells you where the win happened. The position tells you how strong it is: position 1 is the first source a reader sees. The URL tells you which page earned it.
Now compare the same row last month. If it moved from position 5 to position 3, the page is getting stronger for that question. If it dropped off entirely, check whether the page still exists and still answers the question. A 404 or a rewritten intro can remove a citation without any warning.
Two rows side by side often explain each other. If Q07 cites your setup guide and Q08 cites a competitor's setup guide, the gap is the question, not the page. That points you to content work, not link work.
Cited pages share a shape: a clear answer near the top
After two or three months, sort your log by URL. The pages that keep appearing tend to share traits: a clear definition near the top, a short numbered procedure, and a named source for any number.
These traits match how answer engines summarise. A page that states "X is Y" in one sentence is easy to quote. A page that buries the definition in paragraph six is hard to quote, even if it's better researched.
Use the log to find pages that get cited for one question but should answer three. Add a short section that answers the related question directly, then watch whether the next monthly check picks it up. That is a cheap test with a clear result.
Manual checks win on learning; paid tools win on scale
Paid AI visibility tools automate the asking and the logging. They are useful at scale, say 200 questions across five engines. For a small site with 15 questions, the manual routine is cheaper and you learn more, because you read every answer.
The trade-off is coverage and consistency. A tool runs on a schedule you don't have to remember. A manual check depends on you showing up on the first Monday. If you miss two months, the trend line breaks.
No public figure exists for how often AI answers change, so treat any tool's "share of voice" number as a snapshot, not a law. The same caution applies to your own sheet.
Key Facts at a Glance
- Cadence: once a month, first Monday, about 90 minutes for 15 questions across three engines.
- Engines: Perplexity, ChatGPT with browsing, Gemini with web results.
- Core columns: date, engine, question ID, cited, position, URL, context, notes.
- Key metric: linked citations plus brand-only mentions, tracked separately.
- Review trigger: a page that drops out of the top three sources for two months in a row.
- Source rule: write down the exact URL every time; a domain alone hides which page did the work.
How do I check if Perplexity cites my website?
Open a new thread, paste one question from your fixed list, and read the numbered source list under the answer. Search that list for your domain. If it appears, note the position and the exact URL. If it doesn't, record "no" and move to the next question.
Can I track AI citations for free?
Yes. The manual routine uses only the engines' own interfaces and a spreadsheet. You pay in time, not money. Budget about 90 minutes a month for a small question list.
How often should I manually check AI citations?
Monthly is a good default. Weekly checks mostly show noise, and quarterly checks are too slow to catch a drop. Pick a fixed date so the comparison stays honest.
What should I record in a citation tracking sheet?
At minimum: date, engine, question ID, whether you were cited, source position, exact URL, a one-line context note, and any oddity. Keep the columns fixed so month-to-month comparison works.
Why does Perplexity cite some sources but not others?
Perplexity builds its answer from pages it retrieves for that prompt. Pages that state a clear answer in plain language are easier to quote than pages that bury the point. The exact retrieval rules are not public, so treat any explanation as a working theory and test it against your log.
Primary documents: Google Search Console: Performance report and Google Search Central: AI features and your website.