ChatGPT Referrals to E-Commerce Websites

The first large-scale look at AI-referred traffic

When ChatGPT began adding clickable “buy” links to its answers in August 2024, a genuinely new way for customers to reach e-commerce websites was born. This study is the first to measure it at scale, using real sales data rather than surveys or single-site case reports. The authors call it organic LLM traffic (oLLM): unpaid visits that arrive at a retailer because ChatGPT recommended it and linked to it.

What the data covers

• 973 e-commerce websites, spanning all continents and 24 product categories — from retail to finance, travel, vehicles, and business services, across both B2C and B2B — with a combined $20 billion in annual revenue.
• Over 50,000 purchases traced to ChatGPT referrals, set side-by-side with 164 million purchases from every traditional channel.
• 12 months of first-party data (August 2024–July 2025), read directly from each website's own Google Analytics — so the authors see which channel delivered each visitor and what they bought.

Crucially, oLLM is compared head-to-head against the eight established channels retailers already budget for: organic search, paid search, direct, email, affiliate, referral, paid social, and “other.” That common yardstick makes this a real benchmark for AI-referred traffic, not an isolated anecdote.

Small in volume, modest in value per visit

For all the attention AI gets in e-commerce, oLLM is still tiny. One year after launch it accounted for less than 0.2% of all visits — roughly 200 times smaller than Google organic search. Among AI platforms, ChatGPT does almost all the work, driving over 90% of this traffic (Perplexity, Gemini and others are negligible).

How much each visit is worth, fairly compared

Raw averages make oLLM look weak, but partly because its data is thin. The authors use models that control for the website, device, time of year, and how many sessions each figure rests on. Even after this fair adjustment, an oLLM visit converts less often than almost every established channel:

• Affiliate links convert about 86% more often than oLLM, and organic search about 13% more often.
• The one channel oLLM beats is paid social, which converts about 53% less often than oLLM.
• Revenue per visit follows the same ranking: above paid social, below all other channels.

One bright spot: oLLM visitors don't bounce away immediately, suggesting the links are relevant — they just don't yet buy at scale.

A young channel on an upward path

oLLM is not standing still. Over the first year, the share of ChatGPT-referred visits ending in a purchase rose steadily, while traditional channels stayed flat (apart from the usual November–December holiday bump). Revenue per visit edged up too — so the gap to established channels is narrowing, even if it hasn't closed.

A twist: more conversions, smaller baskets

The improvement comes with a catch. As conversions rose, the average order value fell — customers buy more often through oLLM, but spend less per order — so revenue per visit improves only moderately.

Why it's improving: customers are getting better at using AI

The same pattern appears in a second, independent place. Websites whose visitors are more LLM-savvy — a higher share of tech-enthusiasts, and younger audiences — show exactly the same mix of higher conversion and lower order value. Because it shows up both over time and across websites' audiences, the most consistent explanation is growing consumer skill at buying through AI — though the authors call this suggestive, not proven cause and effect.

Where AI already shines: complex purchases

The picture changes sharply by product type. On websites selling complex products — those that need real comparison and guidance — oLLM is far more present and performs much better than its overall average. The authors rate each category's complexity using three AI systems (Claude, Gemini, and ChatGPT) for an independent classification.

More traffic, and stronger conversion

• Complex-category websites receive about 4.6× more oLLM traffic share than simple-category websites — customers turn to AI more when the decision is hard.
• On these websites, oLLM's conversion rate beats five traditional channels: paid social, referral, email, organic search, and direct.
• Revenue per visit also moves up toward organic and paid search levels.

This fits the intuition that a conversational assistant helps most when a purchase involves weighing many options, and little for quick, routine buys. High-complexity categories include vehicles, finance, business services, and heavy industry; low-complexity ones include news, sports, and entertainment.

Match your response to your product

The study's practical message isn't “act now” or “ignore it” — it depends on what you sell.

If you sell simple, routine products

• The study measures the lower funnel — whether a visit converts and what it earns — and there oLLM is still weak for routine purchases. So for now, don't shift lower-funnel budget out of proven performers like affiliate and search — but treat oLLM's discovery value as real, not zero.
• LLMs may still be very useful upper-funnel, in discovery — but because the data credits only the last click before a purchase, it can't capture that role. The study's findings are consistent with consumers starting their search on an LLM, then switching to a more familiar or convenient channel to actually buy.

If you sell complex products

• Here oLLM already deserves attention: on complex-product websites it pairs comparatively strong conversion and revenue per visit with low bounce rates.
• Treat it as a complement to organic and paid search — worth optimizing now, even though the volumes are still small.

Three things to watch

• It's free traffic, for now. Because oLLM is unpaid, it's attractive when return on ad spend matters — but ChatGPT has started to implement paid ad formats and built-in checkout, which could change the economics.
• Getting found by AI is the new frontier. The authors point to generative engine optimization — shaping how AI assistants surface and link to your site — as a fast-emerging discipline worth understanding.
• It's early, and the field is moving fast. This is a first snapshot of a channel that barely existed a year ago — the platforms, the technology, and how people buy through AI are all evolving quickly, so expect the picture to keep changing. Grips Intelligence, which supplied the data behind the study, tracks this traffic across retailers as it develops.