Visitors arriving from ChatGPT convert at around 15.9 percent. From Perplexity, 10.5 percent. Organic search sits near 1.76 percent. One reported case has a tenth of all new signups arriving through ChatGPT referrals.
Those numbers get quoted as evidence that AI search is the future, which is not quite the useful reading. The useful reading is about what kind of visitor arrives that way, because it tells you what to do with them.
The visitor has already been screened
Someone searching Google is at the start of a process. They will open several results, compare, discard most, and possibly return later. Your page is one of ten candidates and most of your visitors were never going to buy.
Someone arriving from an assistant has already had a conversation. They described a problem, received an explanation, asked follow up questions, and were pointed at you as a specific answer. By the time they click, the evaluation has largely happened somewhere you cannot see.
So the conversion difference is not that AI traffic is magic. It is that the assistant did the filtering, and the people who make it through are the ones for whom the answer was genuinely relevant. You are receiving the end of a research process rather than the beginning.
That reframes the volume question. A hundred visitors who have already decided you might be right are worth more than a thousand who are still comparing, and comparing traffic sources by sessions makes exactly the wrong call.
The pattern we see in client analytics is a small, stubborn trickle that converts far above everything else. It is easy to dismiss on volume alone, which is exactly what most reporting encourages.
They arrive with expectations you did not set
Here is the practical problem, and it is where most of the opportunity sits.
The assistant described you. You do not know what it said. The visitor arrives holding a summary of your offering that was assembled from your site, your documentation, third party mentions, and possibly a competitor comparison.
If that summary was accurate, the visitor lands already understanding what you do, and a landing page that spends three sections explaining the basics is wasting their time. If it was inaccurate, they arrive expecting something you do not offer, and they leave confused without ever telling you why.
Both cases argue for the same thing: pages that get specific quickly. Concrete detail about what the thing does, who it suits, what it costs and what it does not do. Vague positioning copy performs badly with this audience because they already read the summary version and came for the substance.
Read More - A Search Plugin Scored 9.8: How to Read Severity Without Panicking
Find out what is actually being said
You can check this directly and almost nobody does. Ask the major assistants what your company does, what it costs, and how it compares to a named competitor. Do it from a signed out session so you are not seeing personalised results.
What comes back is frequently instructive. Outdated pricing, a product you discontinued, a feature attributed to you that belongs to a competitor, or a description drawn from a review site rather than your own words.
Where it is wrong, the fix is usually your own content rather than an appeal to the provider. Assistants draw on what is written clearly and consistently across the web. A pricing page that states prices in text rather than an image, an FAQ that answers the comparison question directly, and consistent product naming all measurably help. We covered the tracking side in how to track AI visibility.
Measure it before you dismiss it
Most analytics setups undercount this badly, which is why teams believe the traffic is negligible.
Some assistant referrals arrive without a referrer at all and land in direct traffic. Others come through domains nobody has segmented. If you have not deliberately built a segment for it, your reported AI traffic is smaller than your actual AI traffic, possibly by a lot.
Build the segment, then look at conversion rate rather than sessions. A source producing two percent of sessions and twelve percent of signups is not a rounding error, and it will never look important on a traffic report.
Then check which pages they land on. It is usually not the pages you optimised for search. Assistants tend to cite documentation, detailed comparison pages and specific answers rather than marketing pages, which means your best converting audience may be arriving on a page nobody has looked at in a year.
We ask clients to segment this before forming a view, because the instinct is almost always wrong in the same direction. A source at one percent of sessions and eight percent of signups is not a rounding error.
What this changes about the work
Search volume is falling and AI answers absorb an increasing share of informational queries. That is a genuine loss of traffic and the response cannot only be to chase the remaining clicks.
The content that earns citations is different from the content that ranked. Direct answers to specific questions, stated plainly, with numbers and specifics rather than hedged marketing language. Pages that make a claim and support it. Documentation written to be read rather than to rank.
It also matters more each year, because the channel it is replacing is shrinking, which we looked at in where demand comes from now.
That is also better for humans, which is the reassuring part. The material that assistants cite tends to be the material a customer would find genuinely useful, so this is not a new discipline so much as the old one done more honestly. Our website development work increasingly starts from what a buyer needs to know rather than from a keyword list.
Frequently Asked Questions
Why do AI referrals convert so much better?
Because the qualification happened before the click. The visitor described their problem, received an explanation, and was pointed at you as a fit. Search traffic arrives at the start of that process, so most of it was never going to convert regardless of your page.
How do I identify AI traffic in analytics?
Create a segment for the assistant domains and check your direct traffic for anomalies, since some referrals arrive without a referrer. Expect your true figure to be higher than the default reporting suggests, which is the main reason this source gets dismissed.
Should I stop investing in search?
No. Search is still much larger in absolute terms even as it declines, and the content that earns AI citations is largely the same content that ranks. The change is what you optimise for within that: specific answers rather than keyword coverage.
What if an assistant describes us incorrectly?
Fix the source material first. Make the correct information clear, consistent and in text rather than images across your own site, and check third party listings that may be outdated. Direct correction requests to providers are inconsistent; changing what they read is more reliable.
Do these visitors behave differently on the site?
Generally they arrive deeper, on specific pages rather than the homepage, and they convert faster with fewer sessions. They also tend to skip the explanatory content, so a page that buries specifics behind positioning copy loses them.
Is this worth attention at low volume?
Judge it on outcomes rather than sessions. A source producing a small share of visits and a large share of signups deserves attention proportional to the signups. It is also growing, so the work compounds rather than being a one off optimisation.
What content actually gets cited?
Specific, verifiable, plainly written material. Documentation, comparison pages that name competitors honestly, pricing stated as text, and FAQs answering the question a person would actually ask. Content written to sound impressive without saying anything is exactly what does not get cited.
Talk to our team if you want to know what assistants are telling people about you, and what to change on the pages they send them to.