How AI chatbots cite websites — and how to become the one they send traffic to

The short answer: AI assistants don't rank ten links; they compose an answer and attach a handful of citations. Those citations come from sources that are machine-readable, factually explicit, consistently described across the web, and — for systems with live search — indexed and fresh. The traffic they send is smaller than classic search but noticeably more decided: the visitor arrives after the comparison already happened. Winning citations is the new distribution game, and it's played with clarity, structure and proof rather than tricks.

This post is the practical sequel to our pillar on website presence in the age of AI — the how, after the why.

Where AI answers actually come from

Three pipelines, worth keeping apart. Training data: what the model absorbed historically — you influence it slowly, by existing consistently across the public web. Live retrieval: ChatGPT search, Perplexity, Copilot and Google's AI Overviews fetch current pages at answer time and cite them — this is where SEO fundamentals still decide everything, including the unglamorous fact that Bing's index feeds several major assistants. Knowledge graphs and structured sources: business profiles, directories, reviews and schema markup that let a machine state facts about you with confidence.

The overlap with classic SEO is large and reassuring: a crawlable, fast site with clean HTML and real authority feeds all three pipelines. The differences are in emphasis — and they're actionable.

What makes a page citable?

Watching which of our clients' pages get cited, the pattern is consistent. Answers stated plainly, early: a paragraph that survives being quoted — definitional, factual, self-contained — near the top. Structure a machine can lift: question-shaped headings, FAQ blocks with FAQPage schema, tables, stated numbers. Entity consistency: the same business name, location and service description everywhere — site, Google Business Profile, directories, LinkedIn — so retrieval systems can triangulate you without guessing. Specificity as a moat: assistants prefer sources that commit — counts, dates, prices where you can, named examples — over pages that hedge everything. Freshness where it matters: dated, updated content wins retrieval for anything time-sensitive.

And the boring prerequisite: decide your crawler policy deliberately. Blocking AI bots in robots.txt is choosing invisibility in this channel; allowing them is the default posture for a service business. Conventions like llms.txt are emerging but not standards yet — watch, don't wait.

The traffic itself: small, late, and better

AI referral visits land with UTM-visible sources — chatgpt.com, perplexity.ai and friends — and the pattern we see across client analytics is consistent: volumes are modest next to organic search, but the sessions run deeper and convert at a healthier rate. It makes sense — the assistant absorbed the browsing phase, so what reaches you is the decision phase. Treat these referrals as a segment in GA4, watch branded-search growth as the shadow metric (many AI-influenced buyers arrive by searching your name afterwards), and accept a truth of the channel: you'll be mentioned more often than you're linked-and-clicked. Mentions build the brand that gets searched; measure both.

The checklist

Clean crawlable HTML, server-rendered · answer-first paragraphs on every commercial page · FAQPage + Organization/Service schema · consistent entity data across the web · real reviews with names · dated and maintained content · crawler policy decided on purpose · AI referrals segmented in analytics. Eight lines that make you legible to machines that recommend things — this is most of what's honestly inside "AEO" and "GEO", and it ships with our SEO work.

Want to know what AI assistants currently say about your business? Ask them — then ask us how to change the answer.