Key takeaways
- A useful platform shows the exact page that was cited, the answer it appeared in, the run and the assistant, not only a list of websites.
- ChatGPT and Perplexity cite in different ways, so an answer without sources means something different in each and a good platform records the difference.
- Sorting the cited sources by type shows which evidence an assistant relies on and which is missing for your category, and that is the reading that leads to work.
Several platforms show which sources ChatGPT and Perplexity cite. The ones worth a trial show the exact page that was cited and not just the website, the answer it appeared in, the run and the assistant, and they let you set your own buyer questions. Below we set out how the two assistants cite, what a platform must show, how we measure citations ourselves, and four platforms that state source tracking on their own pages. It is not a ranking.
How ChatGPT and Perplexity cite, in their own words
The two assistants build an answer from the web in different ways, and that changes what a missing citation means.
OpenAI's help page says ChatGPT may search the web automatically when a question would benefit from current information, and that responses which use web search may include citations. It describes a Sources view, when available, that lists the cited sources and other relevant links, and it adds that search results and citations can be incomplete, outdated or incorrect. See OpenAI's help page on searching the web with ChatGPT Last checked .
Perplexity's help centre describes a different default. It says Perplexity searches the internet for a question and that each answer includes numbered citations linking to the original sources. See Perplexity's help centre page on how it works Last checked .
Three things follow for anyone counting citations.
- In Perplexity, an answer is expected to carry sources, so an answer without them is unusual and worth reading.
- In ChatGPT, whether an answer carries sources depends on whether it searched, so some answers will have none. Report those separately and do not read them as a gap in your pages.
- Neither page promises the sources are complete or correct, so the cited page should be opened and read, not only counted.
Our guide to how an AI answer is assembled walks through where sources fit in the larger picture.
What a platform has to show to be useful
A list of websites is a start. To act on it you need more.
- The exact address of the cited page. A website tells you who. The page tells you what was used, and that is what you can learn from or compete with.
- The answer it appeared in. Keep the question, the full answer and the sources together. A source with no answer attached cannot be checked.
- Which run. Answers vary from run to run, so a platform should show the date and repeat, and count over several runs.
- Which assistant. ChatGPT and Perplexity should be separate columns, never one blended figure.
- Who owns the source. Your own pages, a competitor's, a review site, a publisher, a study. Sorting by owner turns a list into a picture.
- Whether you were named. A cited source and a named brand are different outcomes. Mention, citation and qualified recommendation explains how to keep them apart.
- Change over time. Which sources rise and which drop out after you publish something new.
How we measure citations
We keep our method short and plain, so you can compare it with any platform's.
We start from a fixed set of buyer questions written in the buyers' words. Each question is asked of each assistant more than once, by default twice, because answers vary. We store every answer as it was given, with the sources it showed, and we never edit it afterwards. We then count answers, not impressions: a source counts as cited in an answer when its address appears among that answer's citations, and we report the count over the number of answers asked, such as cited in 9 of 40 answers.
There is one limit we state rather than hide. Citations from the GPT models and from Perplexity arrive with the full page address. Gemini sends its sources through a Google redirect, so in our collection today we can see which website was cited but not which page. Any platform that shows page-level sources for every assistant is doing something different from us, and you should ask how.
Platforms that do this
The lines below say only what each company's own page states on the date shown. They are not a ranking and nothing here says one is better than another. Several pages describe citation tracking in general terms, so use the test above: does it show the exact address, the answer behind it, the run and the assistant?
OtterlyMeasurement tool
Its analytics page lists website citation tracking: every cited address, whether it names you, and which rival it names instead, along with how often each page is cited. It says it runs prompts across ChatGPT, Perplexity and five other engines and tracks them daily. Its site has separate pages for agencies and for marketing teams.
Their pageLast checked
ProfoundMeasurement tool
Its citation page describes monitoring when ChatGPT, Perplexity, Google AI Overviews and other answer engines cite your content or a competitor's across every prompt you track. It sorts each source into a category, such as owned, competitor, earned media, press wire, social or institution, compares citation share by platform, topic and prompt, and exports the data as CSV or JSON.
Their pageLast checked
ScrunchMeasurement tool
Its Monitoring and Citations page describes showing which domains and addresses are cited most often across the answers you track, breaking citations down by content type, and following how citations move over time. Its example views show tracked ChatGPT and Perplexity answers. Its site lists enterprises and agencies as use cases.
Their pageLast checked
Peec AIMeasurement tool
Its pricing page lists source analytics with a domain and address detail view, citation share, a classification of each source as competitor, editorial, reference or user-generated, and a gap analysis of sources that name competitors but not you. The models it lists include ChatGPT and Perplexity. The page states prices; none are repeated here.
Their pageLast checked
Read together, the four pages agree that cited sources are worth tracking and describe source types and citation share. They differ in the assistants listed, the way sources are classified and what you can export. Those are the differences to test in a trial.
What to do with a list of cited sources
A list of sources only helps if you read it. Sort the sources for one buyer question by type: your own pages, competitors, review sites, publishers, studies and press. Then ask which type an assistant leans on for that question and which of them says nothing about you. The gap between those two lists is the work. Our guide to evidence gaps shows how to read the gap, and whether assistants recommend your competitors shows how to read the names beside the sources.
If you want it run for you
Answer Position runs the measurement as a managed service, so you do not log into software. We agree the buyer questions with you, ask them across AI assistants, keep every answer as it was given, including the sources it showed, and read the sources against the question so they are interpreted and not only counted. See our services for the full ladder.
- Snapshot is £995. It is a clear first read: 30 to 50 buyer questions across three assistants, set against three competitors, with a standard report and a 30-minute walkthrough. The fee is credited against an Audit booked within 60 days. It suits a team that wants a first look at which sources the assistants lean on.
- Monitor is £595 a month, minimum 3 months. It tracks 50 buyer questions every month, with a monthly report setting out the five actions that matter most and a call each quarter. It suits a team that wants to watch the sources move after it publishes.
Nobody can promise what an assistant will cite. We can keep the evidence, show how often a source appeared and say how uncertain the reading is.
