Does ChatGPT traffic convert better? What the studies say

Your analytics shows a few hundred visits from chatgpt.com last month, and a handful of them bought something. That's a small slice of your traffic. The question is quality: do the people an assistant sends convert better than people from Google, or worse?

Search for an answer and you'll find both. One widely shared post says AI search visitors convert 23 times better. A peer-reviewed study of 973 online stores says ChatGPT referrals convert worse than almost every other channel. Both are real measurements. They measured different things.

This post covers what the main studies found, why they disagree, and how to measure the number that matters for you: your own.

What the studies found on AI traffic conversion rates

The case for "better"

Ahrefs. In June 2025, Ahrefs reported that AI search sent 0.5% of its visitors over the previous 30 days and 12.1% of its signups, and that "AI search visitors convert at a 23x higher rate than traditional organic search visitors for Ahrefs" (Ahrefs). The post is open about the limits: it's one site, engagement was "a close call" (lower bounce rate and more pages per visit, but shorter visits), and a larger multi-site study was still in progress.

Seer Interactive. Seer's case study looked at one client's GA4 data from October 2024 to April 2025. ChatGPT converted at 15.9%, Perplexity 10.5%, Claude 5% and Gemini 3%, against 1.76% for Google organic (Seer Interactive). AI sources together came to 0.07% of organic traffic: just under 11,000 sessions against almost 14 million. The client's industry and the conversion events aren't named, and the authors offer their explanation, people finishing their research inside the assistant before clicking, as "likely", not tested.

Microsoft Clarity. In November 2025, Clarity reported on more than 1,200 publisher and news sites. Over one month, traffic from LLMs converted to sign-ups at 1.66%, against 0.15% for search, 0.13% for direct and 0.46% for social, and to subscriptions at 1.34%, against 0.55% for search (Microsoft Clarity). Conversions were detected with Clarity's smart events, and AI referrals were still under 1% of traffic. Microsoft also makes Copilot, which led the subscription figures.

The case for "no better" or "worse"

Amsive. In September 2025, Amsive compared LLM and organic traffic on 54 sites over six months, counting only audited macro conversions: purchases for ecommerce, demo requests or form fills for B2B (Amsive). The mean conversion rate was 4.87% for LLM traffic and 4.60% for organic. A paired t-test found the difference was not significant (p = 0.794), and it stayed not significant when only the 33 higher-volume sites were kept. Against each site's average conversion rate, LLM traffic converted better on 56% of sites and worse on 41%. In the exploratory industry split, organic converted higher for ecommerce and consumer services, and LLM traffic for financial services and travel, on samples the author warns are small.

Kaiser and Schulze. The largest ecommerce study, and the only peer-reviewed one, comes from Maximilian Kaiser (University of Hamburg) and Christian Schulze (Frankfurt School of Finance & Management). They analyzed first-party data from 973 ecommerce sites with about $20 billion in combined revenue: more than 50,000 transactions referred by ChatGPT against 164 million from other channels (University of Hamburg). Between August 2024 and July 2025, ChatGPT's conversion rate and revenue per session were higher than paid social and lower than every other established channel they compared (Frankfurt School). Two findings soften that: ChatGPT conversion rates rose over the period (though falling order values offset part of the gain), and for "complex products with high information needs", its referrals can create value. The study is published in Marketing Science. When the working paper came out in October 2025, the authors noted it counts last-click referrals only, so it misses ChatGPT's influence earlier in a purchase, and that ChatGPT was under 0.2% of the traffic in their data (Digiday).

Why the studies disagree

Read side by side, the studies aren't contradicting each other so much as answering different questions.

1. What counts as a conversion. A free signup on a software tool, a newsletter subscription on a news site, a demo request and a purchase in an online store are different events with different base rates. Ahrefs counted signups, Clarity counted sign-ups and subscriptions on publishers, Kaiser and Schulze counted store purchases.
2. One site against many. A single site's 23x tells you a lot about that site. Across 54 sites, Amsive found LLM traffic converting better on some and worse on others, which is exactly how one site can show a large lift while the average shows almost none.
3. What it's compared with. "Organic search" usually mixes people searching your brand name, who are often ready to buy, with people browsing a category. Which slice is the baseline moves the comparison.
4. Small numbers. AI referrals were under 1% of traffic in every study above. A few hundred visits and a few dozen conversions can produce a dramatic rate that moves a lot from month to month.
5. Missing referrals. Cloudflare found that traffic from Claude's native app carries no Referer header and believes the same holds for other native apps (Cloudflare). OpenAI says ChatGPT adds utm_source=chatgpt.com to referral URLs from its search results (OpenAI), but untagged visits with no referrer land in direct. So part of AI-referred traffic is counted under another channel, and whether it converts better or worse than what's left, nobody has measured.
6. Last click. Amsive and Kaiser and Schulze say they credit the last visit before the conversion; the other three don't say how they attribute. Someone who reads about your product in ChatGPT and buys a week later after searching your name counts as organic or direct.
7. Timing. The data runs from August 2024 to late 2025. Kaiser and Schulze saw ChatGPT's conversion rate rise over their year, so any of these figures may already be out of date.

The fair summary: there is no general answer yet. There is evidence that AI-referred visitors can convert very well on some sites, including software and information-heavy purchases, and evidence that for typical online stores ChatGPT referrals have so far converted below most channels. Your site is the only sample that settles it for you.

How to measure it on your own site

1. Define what finishing means. A purchase confirmation, a signup, a booked demo. Use the same definition for every channel, and prefer the outcome you'd pay for over a soft one such as a newsletter signup.
2. Count AI referrals by referrer and by tag. Match the referrer host (chatgpt.com, perplexity.ai, claude.ai, gemini.google.com, copilot.microsoft.com and others) or a utm_source naming an assistant. Server logs also catch visitors whose browsers block analytics scripts. We covered the counting in more detail in do AI crawlers send visitors back?
3. Compare like with like. Set AI referrals against non-branded search or a single landing page group, not your whole site. Compare by assistant too: the studies above found large differences between them.
4. Look at landing pages. People sent by an assistant usually land deep on a product page, a docs page or a comparison, not your home page. The pages that convert tell you which content assistants are quoting well.
5. Wait for enough data. With small counts, a month's rate is noise. Track it monthly and compare the trend, not the single figure.
6. Ask people. Amsive suggests a self-reported field such as "How did you hear about us?" on lead forms, which catches people who read an AI answer and arrived some other way later.
7. Remember the agents that don't refer anyone. A browser agent that buys or books for a person finishes the task itself, so no person is ever referred. Those tasks only show up when you measure agents directly, as we describe in where agents get stuck.

How Ethogram helps

Ethogram runs on your server, so it counts people's visits whether or not their browser runs an analytics script. A person counts as sent by an AI assistant when they arrive from its site (chatgpt.com, perplexity.ai, claude.ai and others) or with its utm_source tag; the tag is read on arrival and not kept, and only the referrer's origin is recorded (AI referrals).

AI visits that converted, on the overview, counts the people an assistant sent who went on to finish in the same visit, from the page they landed on: totals and each assistant against the period before, and the landing pages those visits started on. Finishing means reaching one of your goals, a page path or a named action reported with track(), or, with none set, a confirmation page such as /thank-you or /order-received. A goal page that answered 4xx or 5xx doesn't count. On the People tab, Visits that converted shows the rate for people AI assistants sent beside everyone else, and the goal each converting visit reached first (People).

The limits are the ones above. Conversions count only within the same visit, so someone who reads an answer today and buys tomorrow isn't credited, and a person's visitor key changes every UTC day, so a return visit on another day isn't linked to the first one. Only a visit still open at midnight carries over. Visits from apps that send neither a referrer nor a tag count as direct. And for the agents that do the task themselves, the Scorecard ranks browser agents and answer agents such as ChatGPT agent and ChatGPT-User by the share of their visits that reached a goal (Task scorecard). Pricing includes a free trial.

If you want to know whether the people AI assistants send you go on to buy, see how Ethogram works.