AI-driven traffic isn’t (yet) as good as it’s hyped to be

🤖 Introducing: AI special month, 2026 edition

This month, Science Says insights are going to be a bit special.

To say there is a lot of noise around AI and marketing right now is an understatement. There are the overhypers, claiming you can make millions in a few hours (but only if you buy their course/system/hacks!), and there are the detractors, stuck in thinking AI will produce images of people with 7 fingers, and little more.

As is usually the case, the reality is somewhere in between. With these specials, I’ll bring you up to speed with the latest actual evidence on how AI should impact your key marketing decisions, now and in the future.

These are the kind of findings that change how you think about AI and what you do with it on a Monday morning.

We’re going to start with 2 fundamental insights you need to know about (the 1st one being today’s insight, the 2nd next week). Then the rest of the month will be a bit different. 

On April 21st we’ll launch and I’ll share with you our newest collaboration with Wharton, the Blueprint for AI Agent Adoption (if this sounds interesting, you might also like our previous Blueprint for Effective AI Chatbots).

Then, in the last week of the month, we’ll zoom out and look at things from a high-level strategic view. Based on everything that’s happening, what are the key things you should be focusing on, and preparing for?

For this, I will leave the floor to our Chief Academic Advisor, Stefano Puntoni. He is Professor of Marketing and Co-Director of Human-AI Research at Wharton, and one of the world’s leading figures on AI’s role in marketing.

💜 I hope you’ll love this special month as much as my team and I enjoyed bringing it together.

(Are you new around here? Subscribe for $0 so you don’t miss the rest of these insights).

📝 Context

Topic: AI, Ecommerce, Website
For: B2C. Can be tested for B2B 
Research date: March 2026
Universities: University of Hamburg, Frankfurt School of Finance & Management
Note: This is a working paper that has not yet been peer-reviewed and published. This means it’s more likely to be revised in the future.

We’re all feeling massive pressure to invest in GEO (Generative Engine Optimization) to increase the likelihood of AI chatbots mentioning us or our products.

As people switch to asking AI chatbots (e.g. ChatGPT, Claude) about products and recommendations (vs asking traditional search engines like Google), it’s only logical that you want to be the product that shows up in relevant conversations.

But how actually good is AI traffic for your sales, compared to other channels?

Scientific research gives us the answer.

📈 Recommendation

AI-driven traffic is one of the channels you can use in your mix. Strategically, consider it part of your middle funnel (people actively considering options, including yours).

Prioritize it if your product is complex or research-heavy (e.g. electronics, health supplements, vehicles, many B2B purchases). It performs strongly for these products, with 4.6x higher traffic share than average.

If you sell simple products (e.g. apparel, personal care), optimizing for AI traffic should be lower on your priority list, for now.

🎓 Findings

• On average, people who land on ecommerce sites via AI chatbots (e.g. ChatGPT) generate less revenue per session than visitors from other digital channels. The only channel AI-driven traffic outperforms is paid social (e.g. TikTok ads).

• In an analysis of $20 billion in sales across 973 ecommerce websites, researchers compared AI-driven traffic to other digital channels and found that AI referrals:

  • Had the lowest revenue per session (spend per visit), except for paid social

  • Performed slightly better, but still not strongly, when it came to conversion rates (rate of visitor buying):

    • 11.5% less than organic search

    • 46.2% less than affiliate traffic

    • Less than email, but are 16.7% less likely to leave a page without interaction (e.g. clicks, navigation) 

  • Are improving their conversion rate over time, but their average order value is decreasing

• Scientists also analyzed the share of traffic coming from AI referrals and found that:

  • Between August 2024 and July 2025, only ~0.2% of visits came from AI referrals, mainly ChatGPT

  • Websites for complex products (e.g. healthcare, vehicles), had 4.6X higher ChatGPT traffic share than those selling simple products (e.g. stationery)

  • Websites with younger visitors have 5.5x higher AI referrals

  • Websites with tech-savvy visitors also had 3.8 x greater AI-driven traffic.

🧠 Why it works

• Shopping through AI adds extra steps to our usual shopping journeys (e.g. prompting, reading the answers, clicking on links, then browsing the website).

• We’re not willing to go through this extra friction unless it brings real value, which is not the case when we shop for simple products (e.g. clothes).

• However, when shopping for complex products (e.g. construction machinery), we require more information and detailed comparisons.

• This makes AI shopping more worthwhile for research-heavy complex products, but not so much for others.

✋ Limitations

• As people get more familiar with AI shopping, and technologies improve (e.g. better interfaces and recommendations), this may change. The fact that tech-savvy people are more likely to use AI for shopping supports this direction.

• The research focused on organically-delivered AI traffic. Some AI chatbots have introduced ads. It’s unclear how good paid traffic from AI chatbots is.

• The study used last-click attribution. This means people could have started their search with AI, then returned to buy later as paid search and be counted as coming only from the latter. Or vice versa.

• Showing up in AI results is likely to improve brand awareness, which can eventually lead to higher sales. This was not measured or compared with other channels.

👀 Real-life example

Emma is a mattress brand. This product type is usually heavily researched, which may lead to a higher share than average of AI traffic.

❌ Issue: Emma does a good job with their website, but could optimize it further for AI.

✅ Solution: Emma should prioritize optimizing for AI-driven traffic. For example, by:

• Providing text-based comparisons with other mattresses and competitors (this is most likely what people are asking ChatGPT for)

• Include further information in a Q&A section. This would also increase average product ratings.

• Showcase products by benefits, instead of features (e.g. explaining the benefits of each layer, rather than the technicalities). Making people even more likely to convert.

🔍 Study type

Market observation from (164,875,690 online purchases on 973 websites in 49 countries from August 2024 to July 2025 across 24 categories, including fashion, engineering, gardening, beauty, and cosmetics)

📖 Research

ChatGPT Referrals to E-Commerce Websites: How Do LLMs Compare Against Traditional Channels?. Working paper (Version: March 2026)

Important: We usually only cover peer-reviewed, published research papers. This study is neither; it’s a work-in-progress paper. I decided to make an exception because of its importance and timeliness. However, this means these findings are at a higher risk of changing in the future compared to how our typical insights would be.

🏫 Researchers

• Maximilian Kaiser, University of Hamburg

• Christian Schulze, Frankfurt School of Finance & Management

🎁 Bonus mini-insight

This trivia question is from our Blueprint for Effective AI Chatbots, a collaboration with Wharton, where we collected all the latest science-based tactics to make chatbots more effective. Download it for free here.

Here is how else Science Says can help your marketing:

• 📈 Join the Science Says Platform to unlock all 300+ insights, real-world case studies, and exclusive playbooks

• 📘 Boost your sales and profits with topic-specific Science-based Playbooks (e.g. Pricing, Ecommerce, SaaS, AI Best Practices)

• 🔬 Get science-based thought leadership content for your brand. We help market leaders in AI, tech, and across B2B enterprises, build authoritative thought leadership using a scientific perspective. Find out more.

🎓 It took me, a PhD analyst, and an experienced marketer 16.5 hours to accurately turn this 68-page research paper into this 3min insight.

If you enjoyed it please share it with a friend, or share it on LinkedIn and tag me (Thomas McKinlay), I’d love to engage and amplify! 

If this was forwarded by a friend you can subscribe below for $0 👇