Best software to monitor AI citations and referral traffic: how we do it
How we monitor AI citations and referral traffic for clients: what to measure, which software covers both, and the mistakes that skew the numbers.
Clients rarely ask for "citation monitoring". They ask whether AI is sending them customers. To answer that honestly you need two datasets that most software keeps apart: what AI engines cite, and who arrives on your site afterwards. Here is how we set up both, and the software we use to do it.
Citations and traffic are different problems
A citation is an engine naming your page as a source in an answer. A referral is a visitor arriving from an AI product. They are linked, but not one-to-one. Many answers cite you and nobody clicks. Some visitors arrive after an AI mention without any referrer at all. If you only track one side, you will misread the other.
So the minimum viable setup has three layers:
- Prompt tracking to see which questions produce answers and who gets cited.
- Citation analytics to see which URLs and domains are cited.
- Visitor analytics to see AI-referred sessions and what they do.
We add a fourth for most programs: crawler logs, because an engine cannot cite a page its bot never fetched or failed to read.
What we track in each layer
Prompts. We build a panel of real buyer questions, tagged by topic and intent, and run it daily per model and country. The panel should include category questions, comparison questions and a few that name the client.
Citations. We look at page-level citations first. Which of the client's pages are cited, and for which prompts? Then domain-level: which third-party sites does the engine trust in this market? Offsite mentions matter too, meaning pages that name the brand without linking. Reddit and YouTube get their own check because they show up often as sources.
Traffic. We want AI-referred sessions split by engine, then landing pages, then conversions. The conversion number is the one a finance lead will accept.
Crawlers. We read the bot logs for GPTBot, OAI-SearchBot, ChatGPT-User, ClaudeBot, PerplexityBot and the rest, looking for blocked paths and error responses on the pages we want cited.
The software we run this on
We run client programs on Promptwatch because it covers all four layers in one place. Citation analytics gives page-level and domain-level views, and Citation Trends classifies each citation by content type and source type, showing when a page was first cited, how quickly it ramped, and whether the rate decays. Visitor analytics reports AI-referred traffic and conversions through a lightweight script, a GTM template or the WordPress plugin. Agent Analytics provides real-time, IP-verified crawler logs, so spoofed user agents are filtered out.
The view we open most is 360 Insights, which joins crawler logs, citation clicks and AI-referred visits per page. When a page is crawled, cited and visited, the content is working. When it is crawled but never cited, we rewrite it. When it is cited but gets no visits, we look at the snippet and the title.
A practical note on plans: crawler logs start on Professional ($245 a month), so Essential ($95) gives you citations and visitor analytics without bot logs. Agency plans from $199 a month include crawler logs and visitor analytics.
Where GA4 fits
GA4 can show sessions by source, so you can filter for chat and answer-engine domains. We use it as a cross-check, not a source of truth. Referrer data is incomplete for AI products, and GA4 cannot tell you which prompt or which cited page led to the visit. Use it to sanity-check totals, and use the AI-specific tooling for attribution.
Common mistakes
- Reading single runs. Citations change run to run. Look at weekly trends across a panel.
- Counting mentions as citations. A brand mention in text is not a linked source. Track both, separately.
- Installing the traffic script late. You cannot backfill referrals. Put it in before the first campaign.
- Ignoring blocked bots. If robots.txt or a CDN rule blocks the crawler, no content change will help.
- Tracking only your own domain. The domains your competitors earn citations from are your outreach list.
A monthly review that works
Every month we pull four views: the pages gaining citations, the pages losing them, the prompts where a competitor owns the citation, and AI-referred conversions by landing page. Then we pick two actions. The Action Board helps here, since it produces a prioritized list from the data, and the content gap analysis shows which questions the site does not answer well enough to be cited.
What software cannot tell you
It cannot promise a citation, and it cannot see private conversations. It can show how often you are cited across many runs, and what those citations lead to. If a vendor shows you a perfect, stable citation count, be suspicious.
Where to start
If you have no tracking today, set up the visitor script first, then a prompt panel of 30 to 50 questions, then citation views. Add crawler logs when the first two layers show gaps you cannot explain. Within a month you will know whether AI is a channel worth investing in.