How to measure traffic and leads from AI assistants
Three layers, because no single one tells the truth: a custom analytics channel that groups AI referrers (chatgpt.com, perplexity.ai, gemini.google.com and the rest); server logs showing which pages AI bots fetch; and a "how did you hear about us" question on your intake forms. The last layer matters most — several AI platforms strip referrer data, so analytics undercounts, and self-reported attribution is where AI-driven revenue actually becomes visible.
How do I measure traffic and leads coming from AI assistants?
Layer one: the AI channel in your analytics
Create a custom channel group matching referrers from chatgpt.com, chat.openai.com, perplexity.ai, gemini.google.com, copilot.microsoft.com and claude.ai. This gives you a trend line and a conversion rate for the channel — and the conversion rate is the number to watch. Ahrefs measured AI-referred visitors converting at roughly 23 times the rate of organic search visitors in their own funnel: about half a percent of traffic driving twelve percent of signups. Small traffic numbers from this channel are routinely worth more than they look.
Know the caveat: several platforms strip referrers entirely, so this layer is a floor, never the total.
Layer two: what the bots themselves tell you
Your server logs record every fetch by ChatGPT-User, OAI-SearchBot, PerplexityBot and their peers. Pages fetched repeatedly are your live citation sources — the pages assistants actually pull when composing answers. That list is operational gold: keep those exact pages fresh, make sure each carries a clear next step for a human reader, and watch for AI-referred 404s, since assistants send roughly 2.9 times Google's rate of traffic to URLs that do not exist. Redirect every one to the closest real page.
Layer three: ask people
Add "How did you hear about us?" to every intake form, with AI options listed plainly — ChatGPT, Perplexity, an AI summary in Google. This is the ground truth that survives stripped referrers, cross-device journeys, and the common pattern where an assistant names you unlinked and the buyer searches your brand afterwards — a journey every analytics package misattributes to branded search.
For paid placements the chain is tighter: campaign and ad identifiers on every landing URL, captured on the first touch, stored with the lead, and carried through to the booked call. That is how we run our own ChatGPT campaigns, and it is the difference between knowing a channel works and believing it does.
The monthly review that makes it a system
Once the three layers exist, the review is short: share of voice from prompt testing, AI channel traffic and conversion, top cited pages from logs, and self-reported AI leads from intake. Four numbers, monthly, against last month. Quarterly, redo the competitive gap analysis — the citation landscape churns too fast for an annual view.
Why does my analytics show almost no AI traffic?
Three usual reasons: no channel grouping (visits hide inside referral noise), platforms stripping referrers (invisible by design), and unlinked mentions converting through branded search. Set up all three layers before concluding the channel is small.
What conversion rate should I expect from AI referrals?
Higher than organic, in most reported data — Ahrefs' 23x figure is the extreme, but directionally the pattern repeats: a visitor arriving from an assistant's specific recommendation lands pre-qualified. If your AI referrals convert below organic, the mismatch is usually landing-page relevance.
Can I see which prompts led people to my site?
No — assistants do not pass the conversation. The workable proxy is your own monthly prompt testing plus which pages the referrals land on, since each cited page maps to a topic.
How soon after starting AEO should the numbers move?
Log-file activity moves first — bots fetching your refreshed pages within days or weeks. Prompt-test share of voice moves next. Referral traffic and self-reported leads lag by weeks behind that. If nothing has moved in the logs after a month, revisit the technical checklist before touching content strategy.
Last reviewed 23 August 2026