The Traffic Shift Nobody Budgeted For: Buyers Are Asking Assistants First

A small channel behaving like a large one

Referrals from AI assistants are still a modest share of most sites' traffic. Their conversion rate is not modest. Similarweb's analysis put AI assistant referrals converting at around 7.1%, behind paid search and ahead of organic search and social.

That combination, low volume and high conversion, is unusual and worth understanding rather than dismissing on volume.

Why the visits are so qualified

A search engine returns ten options and the searcher begins evaluating. An assistant returns an answer, often with a recommendation and a short list, after the person has already described their situation in detail.

By the time someone clicks through from that conversation, the equivalent of several evaluation steps has happened without you. They arrive knowing what they want and having been told you fit. That is closer to a warm referral than to a search click, which is why it converts like one.

It also fits how buying already works. Given that groups spend only around 17% of their time with suppliers, a channel that participates in the other 83% is structurally valuable.

What decides whether you appear

Assistants assemble answers from what they can read and what the wider web says about you. In practice that means a handful of unglamorous things:

You must be crawlable. If your robots file blocks AI crawlers, you are absent by configuration. Many sites block them without realising.

You must state facts plainly. Assistants extract claims. Pages that make specific, attributable statements get quoted. Pages of adjectives do not.

You must be corroborated elsewhere. Being described consistently across your own site, directories, and third party mentions makes the model more confident about what you do.

You must answer real questions. The content that gets cited is the content that resolves a question completely, which is the same standard as the AI content quality bar.

The practical checklist is in getting cited by AI search.

Measuring a channel that hides

Assistant referrals often arrive without a clean referrer, or get bucketed as direct. Some do show as chatgpt.com or perplexity.ai in analytics, and those are worth watching closely because each one represents a highly qualified visit.

The rest will not be visible in analytics at all. This is the same measurement problem as dark social, and it has the same fix: ask people how they heard about you and read the free text answers. "ChatGPT recommended you" is an answer that appears more often each quarter.

What to do this month

1. Check your robots file allows AI crawlers. Five minutes, and it is the difference between participating and not.
2. Publish specific, checkable claims, with sources, on the questions your buyers ask.
3. Make your basic facts consistent everywhere: what you do, who for, where, and what it costs.
4. Watch for assistant referrers in analytics and treat each as a signal rather than a rounding error.
5. Do not chase tricks. Attempts to manipulate model output are the same category of behaviour as scaled content abuse, with the same eventual outcome.

The honest caveat

This channel is young, the measurement is poor, and anyone quoting precise forecasts for it is guessing. What is defensible: the visits convert unusually well, the cost of being eligible is low, and the work that earns citations is the same work that earns everything else content does. That makes it a cheap bet rather than a sure one.

Frequently asked

Is this replacing SEO? No. It runs on the same foundations. A site that cannot be crawled or says nothing specific fails at both.

Can we track it properly? Partially. Combine referrer data with self reported attribution and accept the gap.

Can Fixora set this up? Yes: crawler access, structured facts, and content built to be quoted rather than skimmed. Ask for the audit and get findings within 48 hours.