The Invisible Brands of ChatGPT

In this chapter, I want to dig into something I’ve been thinking about a lot lately: are all eCommerce categories created equal in the eyes of AI? Are products like $500 gadgets and niche gear getting most of the LLM attention, while basics like t-shirts and shampoo stay invisible?

The short answer is no. Let’s unpack why.

Before we dig in, let’s start with this chart.

Take a look at ChatGPT’s total visits compared to Amazon, X, and IG. It’s been skyrocketing at a pretty insane rate. 

Out of the 5.7b inbound visits to ChatGPT last month, 3.1b were referred out, and 59 million of those referrals were to eCommerce sites. 

77.6% of those referrals went to marketplaces. Amazon alone got 0.38% of all ChatGPT traffic. Then it’s a long tail: Etsy (0.19%), eBay (0.09%), Temu (0.08%).

So if you’re not seeing it in your analytics, that doesn’t mean it’s not happening.

ChatGPT isn’t a traditional channel. It’s an invisible layer shaping how people decide. It starts upstream, during discovery and education, and plants seeds that often bloom as branded search or direct visits days later.

Now let’s zoom in.

TL;DR:

• AI loves high-consideration products. Categories like electronics, beauty, and baby gear consistently outperform. Where shoppers need guidance, ChatGPT delivers.
• Low-intent categories lag. It’s not about product type. It’s about how clearly a brand explains itself to the AI.
• Apparel looks weak at first, but it’s split. Zara crushes while peers flail. Emerging brands can still win with the right structure.
• Most brands are invisible to AI. Over 40% get less than 2% of their traffic from ChatGPT. But a few exceed 6%, showing what’s possible.
• Winning brands nail the backend. Rich metadata, structured content, and crawlable pages decide your AI visibility.
• AI shopping is not a front-end game. Assistants are not impressed by a pretty design. They reward sites that are easy to read, parse, and quote.
• Even if ChatGPT isn’t your top channel today, it’s shaping how shoppers learn and decide. And you probably won’t see it in your attribution model.
• Your product page is now your pitch deck…to a bot.

This chapter builds on last week’s deep dive on Bing. If you missed that, I recommend starting there.  edition 02- Why You Should Care About Bing

Let’s zoom out first. No, not all categories benefit equally from this AI shift. There’s growing evidence that certain verticals, especially high-ticket or complex ones, see stronger performance from AI tools.

Others, mostly lower-cost or habitual purchase categories, lag behind. But the reasons might surprise you. We’ll discuss more in this newsletter. 

Category Differences in AI Shopping Performance

Early data shows clear gaps. High-consideration products outperform. These are expensive or technically complex items, like TVs, laptops, niche appliances, or luxury goods. Shoppers face decision fatigue in these categories. AI thrives here. It narrows options, explains trade-offs, and acts like a smart salesperson.

Adobe reports 87% of users prefer AI help for large or complex purchases. Experts estimate that AI can drive 2–5 times higher conversion in these verticals.

Take outdoor pizza ovens. If someone asks ChatGPT for the best compact pizza oven, and Ooni has strong reviews, it gets featured. That’s qualified traffic with high intent.

So the takeaway for brands in these categories is clear: structured specs, benefit-driven descriptions, and strong reviews are now your growth engine.

Commodity Categories and AI Blind Spots

On the flip side, apparel, home goods, and grocery underperform. Not because people don’t shop there, but because AI doesn’t see enough signal to guide decisions.

For things like paper towels or shampoo, there’s no need to ask a chatbot. These are habitual buys. For groceries, AI helps a bit with recipes, but not much else.

Fashion is more nuanced. It’s personal. AI can’t fully grasp taste or fit. People enjoy browsing. They don’t want a quick answer. So for now, AI drives inspiration, not conversion. But that could change fast with Google’s AI Mode and similar tools.

The Visibility Spread: Winners vs. Everyone Else

When we looked at ChatGPT traffic across brands, it wasn’t even close. Over 50% of brands get under 2% of traffic from ChatGPT. 20% get under 0.5%. Only 7% break 6%.

The gap is already wide. And visibility depends on how well AI understands you, not how well-known your brand is.

Based on the analysis we did, there is a clear “winners” and “losers” 

Top-performing categories share one trait: shoppers want guidance.

• Electronics leads with a 4%+ share. Shoppers ask AI to compare specs.
• Jewelry & Watches perform well due to emotion-driven purchases.
• Baby & Parenting stands out—one toddler sleep brand gets 8% of traffic from ChatGPT.
• Health & Beauty is strong too. Clear benefits and structured info win.

Where people hesitate, compare, or need reassurance, that’s where AI thrives.

Now the opposite. Sports & Fitness underperforms, despite brands like Nike and Adidas. Why? Their content isn’t optimized for AI, and they rely heavily on offline sales.

In Arts & Crafts, only one brand breaks 3%. Most barely register. Weak structure and limited discoverability are likely reasons.

Even big marketplaces struggle. AI prefers individual product recs, not aggregated listings.

Apparel: A Split Story

Apparel looks weak overall. But when you zoom in, it’s a tale of extremes.

Among the top five apparel brands online, Zara dominates ChatGPT traffic with a 66.5% share. The rest, Gap (12.8%), Skechers (9.0%), Uniqlo (6.4%), and Victoria’s Secret (5.4%), lag well behind.

Why? Zara’s backend is clean. Metadata is structured. Descriptions are clear. Page layouts are simple. The site is LLM-friendly.

Others? Not so much. Clutter, heavy branding, inconsistent formatting—all confuse the AI. When ChatGPT has to guess, it skips.

Emerging DTC brands also pull the category down. Most aren’t playing the AI game yet. Schema is missing. Pages are unstructured. Traffic reflects that.

So while apparel’s median is low, Zara proves what’s possible.

Fashion and accessories were super interesting, as the theory was that we would barely see any significant income traffic from GPT. But I was surprised (and quite happy) that some of our strongest brands see 5%+ of their incoming traffic coming from ChatGPT, and for the best one, I noticed a remarkable 14.6% of their traffic.

So I had to dive deeper and understand why. What I noticed was that this brand couple of the key things I mentioned in the last newsletter that make the LLM crawler’s life much easier and ranking the page with high relevance/authority for EACH ONE of their products

• Size & Fit section built from user reviews

• Full size charts and measurement guides

• Detailed fabric and wearability descriptions

• Elegant, natural-language product copy

• And if all that wasn’t enough, they even added the model’s info

All of it structured and easy for the LLM to read. While browsing the super impressive product page, I couldn’t stop thinking about how all of the provided data is being used to answer any type of question along the customer journey. 

Then I wanted to challenge myself and look at the product page of the site that got low traffic from ChatGPT (0.81%)  and see if I could notice what they were doing wrong, or not as good.

I admit, it looked quite well, at least from the information POV it provides, they even provided an FAQ within the product page.

So… I did what everyone would have done, and I went to do a deep research on ChatGPT. I asked it to analyse both product pages, and give me its point of view on what is working for each product page and what’s not. 

To make it even more interesting, I asked it to guess which one got more traffic from it :) 

And the results were:

LULU won the fight because: 

• Structured data & metadata, which was easier for crawlers/LLMs because of its Static HTML title with full keyword string; descriptive image alt tags; price & rating visible in source HTML. Vs Spanx, who was indeed a Keyword-rich title/meta, but much of the page renders client-side behind JS
• Rich product content with depth & UGC – Story-style paragraph plus Size & Fit bullets and ≈ 800 reviews with user photos & body data
• Crawlability / AI access – Lulu with Fully readable HTML; no bot blocks; paginated reviews still text-based vs Spanx, which had an Anti-bot wall and heavy JS (“Unauthorized website” screen) that can hide content from non-Google crawlers
• AI-friendly highlights – Spanx won here due to compact benefit bullets perfect for LLM extraction vs Lulu that had Attributes dispersed; no “at-a-glance” bullet summary
• Natural-language keywords – was indeed a tie as I sensed, as both had super relevant content within the product page

And the last one was beyond the hands of Spanx but was a factor (not sure how big though) – Brand/domain authority, both are strong, but Lulu won here due to raw traffic that the site gets

So Lulus wins on crawl-friendliness, review depth, and overall traffic, making it an easy pick-up for LLMs when users ask for a “romantic midi dress for a wedding.”

Still worth taking everything with a grain of salt, as on average the overall referral traffic is ±5% and almost 95% coming from Organic or Paid.

Having said that, AI tools like ChatGPT may be quietly shaping your funnel

1. Users who click are often more qualified and closer to purchase. 
2. No one knows how much of the paid or organic traffic actually did an initial research via ChatGPT. A possible indication of that is that you’re seeing traffic rise in branded search.

Final Thoughts

Right now, ChatGPT traffic is still small for most, around 5% average. But it’s likely influencing far more than you can see. Shoppers who click through are usually closer to purchase. And branded search spikes might be the clue that AI planted the seed.

This chapter is getting long. I’ll cover the technical checklist for making your site AI-visible in the next one. But for now, remember:

You’re not optimizing for humans anymore. You’re optimizing for assistants that behave like trusted product advisors.

And most of them are blind unless you show them exactly what to see.

— Tomer