A growing share of buying journeys now starts in a conversation with an AI. ChatGPT, Perplexity, Claude, Google's AI Overviews: these tools recommend products, compare solutions — and send traffic. Most sites don't even see it. Here is how to measure it, evaluate it and grow it.
Quick Answer: how do you measure AI traffic?
Through the referrer — provided your analytics classifies it. Visits from assistants carry identifiable referrers: chatgpt.com, perplexity.ai, gemini.google.com, claude.ai, copilot.microsoft.com. Most tools drown them in generic "referral" — check that yours makes them a dedicated channel.
What to know about this traffic:
- It is still small in volume (a few percent), but growing fast.
- It arrives late in the decision journey: the AI already provided the context and the comparison; the click is verification or action.
- Judge it on engagement and conversion, never volume.
To attract more: question-and-answer structure, sourced figures, FAQPage JSON-LD, topical authority — the field of GEO (Generative Engine Optimization).
A new channel, invisible in most stats
When a user clicks a source cited by ChatGPT, the visit lands on your site with a referrer (chatgpt.com). The problem: most analytics tools file these visits under catch-all "referral", between some directory and a forum. As a result the AI channel doesn't exist in your reports, its trends are undetectable, and you steer 2026 with 2015's categories.
Referrers to watch: chatgpt.com and chat.openai.com (ChatGPT), perplexity.ai, gemini.google.com, claude.ai, copilot.microsoft.com — plus the AI Overviews share, harder to isolate as it blends into organic Google traffic.
Low volume, high value: what these visits are worth
Across the sites we observe, AI traffic shares three traits:
- It is still modest — a few percent of traffic at best, growing fast.
- It arrives late in the journey. The user has already asked their questions, received the comparisons, often a recommendation. The click to your site is verification or action — not discovery.
- It converts above average. The logical consequence: less volume, more intent. The exact inverse of viral social traffic.
Practical conclusion: judge this channel on engagement and conversion, never on volume. A 200-visits/month channel converting at 8% deserves more attention than a 5,000-visit channel at 0.3%. Exactly the kind of trade-off a channel-segmented conversion rate reveals.
In practice with Mirage — The Acquisition module natively classifies AI and LLM traffic into a dedicated channel, alongside search, social and referral. You see the AI channel's volume, engagement and conversions at a glance — no configuration, no referrer list to maintain. Free 30-day trial.
Growing the channel: GEO fundamentals
GEO (Generative Engine Optimization) aims to make your site a cited source for answer engines. The levers overlap with good SEO, with specific accents:
1. Answer questions, literally
AIs assemble answers. Content structured as crisp questions and answers (interrogative headings, the direct answer at the top of the section, FAQ marked up as FAQPage JSON-LD) is easier to cite than elegant but diffuse prose.
2. Provide sourced facts
Figures, dates, comparisons, prices: concrete verifiable data is the fuel of AI answers. A "Mirage vs Matomo" page with a factual table stands a better chance of being cited than a "we're the best" page.
3. Build topical authority
AIs, like Google, favour sources that are consistent on their subject. A cluster of articles that reference each other — consent-free measurement, the consent exemption, GA alternatives — outweighs isolated pages.
4. Stay technically accessible
Clean server-rendered HTML, content readable without JavaScript, a robots.txt that doesn't exclude GPTBot/PerplexityBot (unless that is an editorial choice), decent response times. AI crawlers are less patient than Googlebot.
5. Measure the loop
GEO without measurement is a prayer. Track the AI channel in your stats, spot which pages receive those clicks, reinforce what gets cited. The SEO module crossed with acquisition closes the loop: queries → citations → visits → conversions.
Setting up the tracking: three maturity levels
Level 1 — see the channel. A tool that natively classifies AI referrers (or, failing that, a custom channel grouping the domains above). You finally know whether the topic concerns you, and how fast it grows for you.
Level 2 — qualify the channel. Engagement (duration, pages/session), conversion rate and landing pages of AI traffic, compared with classic search. This is where the investment decision is made: on most B2B sites we observe, the AI channel's conversion clearly beats social, sometimes search.
Level 3 — close the loop with content. Cross the cited pages (those receiving AI clicks) with your editorial output: which formats, topics and structures do answer engines pick up? This loop turns GEO from an act of faith into a measured programme.
What AIs cite: observable regularities
By regularly querying the assistants about your own sector (a monthly manual audit is enough to start), patterns emerge:
- "Answer" pages — crisp definitions, how-to guides, FAQs — are massively over-represented compared with marketing pages.
- Comparison tables are regularly reused almost verbatim: a factual "X vs Y" page is a citation machine.
- Dated, sourced figures ("in 2026, the CNIL…") reassure models seeking to anchor their answers.
- Cluster coherence counts: cited sites are usually cited across several neighbouring queries — topical authority works as in classic SEO.
Conversely, mass-generated content with no expertise signal, keyword-stuffed pages and unstructured walls of text almost never appear.
SEO and GEO: one programme, two outlets
Do you need a separate "GEO budget"? No — 80% of the work is shared with SEO: intent research, quality structured content, structured data, internal linking, authority. The specific 20%: writing directly citable answers (an AI cites a paragraph, not a page), maintaining dated verifiable facts, allowing AI crawlers in robots.txt, and measuring the channel to arbitrate. Treat GEO as an additional outlet of your existing editorial programme — not a competing discipline.
Block or welcome AI robots?
The publisher's dilemma: allowing GPTBot and friends feeds models with your content; blocking them removes you from answers. For a media business living on ad audience, it's debatable. For a business site — SaaS, services, e-commerce — the trade-off clearly favours openness: every citation is a free recommendation late in the buying journey. You don't turn away a referrer because they read your brochure.
FAQ
How do I know whether ChatGPT sends me traffic?
Through the referrer: visits from ChatGPT arrive with a chatgpt.com referrer, Perplexity's with perplexity.ai, and so on. Your analytics tool must still classify them into a dedicated channel rather than drowning them in generic "referral" — that is the thing to check.
Does AI traffic convert better than Google traffic?
Early studies and our observations converge: volume is low but intent is high. A visitor clicking from an AI answer has already received the context, the comparison and often the recommendation — they arrive late in the decision journey. Measure engagement and conversion per channel to verify on your own site.
What is GEO (Generative Engine Optimization)?
SEO's equivalent for AI answer engines: optimising content to be cited by ChatGPT, Perplexity or Google's AI Overviews. The levers: structured factual content, direct answers to questions, sourced figures, topical authority, and technical accessibility for AI crawlers.
Should I block or allow AI robots on my site?
A strategic question. Blocking (via robots.txt: GPTBot, PerplexityBot…) protects content but excludes you from answers — and from the channel. For most business sites, being cited by AIs is an acquisition opportunity that outweighs the content-reuse risk.