Traffic from AI assistants: small, but it converts. How to earn it and measure it
There is little traffic from ChatGPT and other assistants, but studies agree it converts markedly better than organic search. How to get into the answers, how to measure it, and why most companies have no idea how much arrives.
- Volume is small — one analysis puts it at 47 to 190 times less than organic search.
- Conversion rate is markedly higher; estimates range from +31 % to more than tenfold depending on the dataset.
- Most companies have not measured it: visits from assistants often land under direct traffic in GA4.
When someone asks today “what should we do about AI”, the answer has two parts. One is about what AI does to search, covered in AI Overviews and AI Mode. The other is about something new: people arriving straight out of a conversation with an assistant. There are few of them, but they behave differently.
How many there are and how they convert
Volume is small so far. One analysis puts traffic from assistants at 47 to 190 times less than organic traffic from Google. For most sites that means a fraction of a per cent of visits.
What is interesting is what happens to those visits. The available studies agree on direction and disagree on magnitude:
| Source | Finding | What limits it |
|---|---|---|
| Adobe Digital Insights, Q1 2026 | Visitors from AI assistants converted 42 % better than others in March 2026; a year earlier 38 % worse | The most credible source here, but we only saw it second-hand |
| Analysis of 94 online shops | ChatGPT 1.81 % vs. 1.39 % for non-brand organic, that is 31 % better | The most sober figure; e-commerce, smaller sample |
| Conductor, AEO Benchmarks 2026 | 1.66 % sign-up rate from AI vs. 0.15 % from organic, roughly 11× | Largest sample (13,770 domains), but we could not verify it at source |
| Seer Interactive | 15.9 % conversion rate for a B2B client from ChatGPT | A single client; the figure is repeated by many sites, which makes it look like more data than it is |
First: older data said the opposite. A 2024 study found ChatGPT converting 13 % worse than organic search. The reversal between 2024 and 2026 may be a real change in behaviour, or a change in what can be measured at all.
Second: some visits from assistants are misfiled as direct traffic in analytics. That distorts the comparison in both directions — missing visits and missing conversions.
Why it converts better
The likely explanation, and one that matches what we see in accounts: the assistant does part of the research for the user. Someone clicking through from an answer already has a narrowed shortlist and often a specific reason to come to you. It is like the difference between “what types of flooring are there” and “where to buy vinyl flooring with installation in Leeds”.
It follows that the volume will never be large — and does not need to be. For a channel with a tenfold conversion rate, a tenth of the visits is enough.
How to get into the answers
Answer engine optimisation (AEO, sometimes GEO) has no settled methodology yet, and anyone claiming otherwise is selling a course. What does appear to work from observation, and does no harm to classic SEO either:
- Answer the question immediately. A short, quotable summary at the top, detail below. We do that on this site with the “In short” section on every article — for exactly this reason.
- Write concretely and with numbers. Models cite claims, not atmosphere. A sentence with a figure, a date and a source has a far better chance of being picked up than a general piece of advice.
- Name your sources. Not only for honesty: text that grounds its figures in traceable sources is easier to cite.
- Provide structured data. Article, FAQPage, BreadcrumbList, Organization. Machine-readable structure helps search engines and models alike.
- Consider an
llms.txtfile. It is not a standard and nobody guarantees anyone reads it; it is a cheap attempt to hand models a map of your content. Ours is at /llms-en.txt. - Do not gate the content. What sits behind a registration will never reach an answer.
- Be mentioned elsewhere. Models draw on discussions, reviews and trade sites too, not only your own.
One honest note: a citation is not a click. A brand can be mentioned in an answer a hundred times and nobody arrives. That is an argument for tracking mentions separately from visits.
How to measure it
Most companies do not have this, and it is an hour’s work. By one survey, only 14 % of marketers track AI search performance at all.
- Create a segment in GA4 by referral source. Domains such as chatgpt.com, perplexity.ai, claude.ai, copilot.microsoft.com and gemini.google.com.
- Expect some to leak into direct traffic. Some answers do not pass a referrer. Watch for rises in direct traffic to deep pages — that is usually hidden AI traffic.
- Track conversions, not visits. At these volumes visit counts are noise; conversion rate and value are what mean anything.
- Ask in the order form. “How did you hear about us?” is imprecise, but at low volumes it is often the only usable method.
- Do not evaluate it weekly. At ten visits a month a weekly view carries no information.
How much attention it deserves
Soberly: today it is a supplement, not a channel. The right balance of effort, as we see it:
- Do now, because it costs an hour and helps elsewhere too: the analytics segment, structured data, a summary at the top of articles.
- Do continuously, because it pays off regardless: concrete content with numbers and sources that answers real customer questions.
- Do not do yet: buy expensive “AEO audits” or rebuild the site for this. The methodology does not exist yet and the data underpinning it changes every quarter.
What to take away
Traffic from AI assistants is small today and probably high quality. The most sensible response is not a strategy but measurement: find out how much you have and how it converts, because without that you will be making decisions next year on someone else’s numbers from a US dataset. And the best thing you can do for being cited — concrete content with numbers and sources — happens to be exactly what works without AI too.
Common questions
Does traffic from ChatGPT convert better than from Google?
2026 studies agree that it does, but they differ widely on how much. Adobe Digital Insights reports that in March 2026 visitors from AI assistants converted 42 per cent better than others, having converted 38 per cent worse a year earlier. A more sober analysis of 94 online shops found a 31 per cent difference, while other datasets claim up to elevenfold. Older data from 2024 showed the opposite. The reliable conclusion is therefore the direction, not a specific number — your own analytics segment will tell you more than any of these sources.
How do I measure traffic from AI assistants in GA4?
Create a segment by referral source with domains such as chatgpt.com, perplexity.ai, claude.ai, copilot.microsoft.com and gemini.google.com. Expect some visits to miss it, because some answers do not pass a referrer and the visit lands under direct traffic — a hint is a rise in direct traffic to deep pages. At these volumes track conversions and their value rather than visit counts, and do not evaluate in a weekly window.
What is AEO and is it worth investing in?
AEO, or answer engine optimisation, is the effort to get into the citations that AI assistants give in their answers. There is no settled methodology yet, so expensive audits and courses can wait. What is worth doing now, because it is cheap and helps elsewhere: put a short quotable summary at the top of your content, write concretely with figures and traceable sources, add structured data, and do not gate content behind registration. Remember too that a citation is not a click — a brand can be mentioned in answers without any traffic resulting.
- AI Shoppers Now Convert 42% Better Than Google Traffic, Digital Applied (summarising Adobe Digital Insights, Q1 2026)Source of the Adobe Digital Insights figure of +42 % in March 2026 against −38 % a year earlier. We saw Adobe’s primary analysis only second-hand through this agency text.
- AI Referral Traffic vs Organic Search: Conversion Rates and Performance Compared, AirOpsSource of the 94-shop comparison (1.81 % vs. 1.39 %), the Conductor AEO Benchmarks 2026 figures (13,770 domains) and the Seer Interactive number (15.9 % for one B2B client). A vendor blog; we did not verify the individual studies at source.
- ChatGPT vs Google Traffic: Conversion Rate Truth (2026), Relevant AudienceSource of the older 2024 data showing ChatGPT converting 13 % worse than organic search. Shows how quickly conclusions in this area reverse.
- AI Referral Traffic Stats 2026, RCTRLSource of the estimate that assistant traffic is 47 to 190 times smaller than organic, and that only 14 % of marketers track AI search performance. A compilation, not primary research.