Key takeaways
- ChatGPT Search cited product pages in 32.8% of shopping-related responses by July 2026, up from roughly 18% in March, per Promptwatch's citation type data. Google AI Overviews crossed a similar threshold, with product pages overtaking listicles as the most-cited format in late July.
- ChatGPT cites only about 5 sources per response versus roughly 10 for AI Overviews and Perplexity, so the citation slots are scarce and competitive. Fight for AI Overviews as an easier entry point while you build authority for ChatGPT.
- Feed hygiene now matters as much as page copy. ChatGPT Shopping, Perplexity's Merchant Program, and Google's Universal Cart all want clean, real-time product data: GTIN, price, availability, and plain-language descriptions, not marketing prose.
- Since August 8, 2026, ChatGPT has been running
site:scoped searches at nearly 17% of its fanout queries, meaning thin, unindexed, or JS-rendered category pages can get skipped entirely. - Don't build your visibility strategy around a single third-party source. Reddit's share of ChatGPT citations collapsed from roughly 4% to under 1% in a single week in August 2026, which shows how fast a platform-side change can erase a channel you were leaning on.
Why this isn't a future problem anymore
If you run an e-commerce store, you've probably already felt this: someone lands on a product page from ChatGPT or an AI Overview, and they buy at a noticeably higher rate than your average visitor. Adobe's 2026 retail data backs that up. AI-referred traffic to US retail sites grew 393% year over year in Q1 2026, and by March it was converting 42% better than paid search and email. A year earlier, that same traffic converted 38% worse. That's not a gradual trend line, that's a flip.
The mechanism is simple. A shopper who asks an AI assistant "what's the best trail running shoe for wide feet under $150" gets an answer that has already done the comparison shopping for them. If your product shows up in that answer, they arrive at your page pre-sold. If it doesn't, you never entered the running, no matter how well your PDP converts once someone lands on it.
Generic GEO advice, written for blogs and SaaS companies, doesn't map cleanly onto this. E-commerce has its own mechanics: feeds, SKUs, schema, marketplaces, and now literal shopping checkout protocols baked into the AI platforms themselves. This guide is about those mechanics specifically.
What AI engines actually cite when they shop
The content-type data is the most useful thing to internalize here, because it tells you where to spend your time.
| Citation type | ChatGPT Search (July 2026) | Google AI Overviews (July 2026) |
|---|---|---|
| Product page | 32.8% | 16.3% (17.9% by month-end) |
| Listicle | 9.7% | 18.0% (16.2% by month-end) |
| How-to | 4.1% | 15.1% |
| News article | 5.2% | 13.5% |
| Comparison | 2.9% | 3.6% |
| Video | 0.1% | 5.9% |
Two things jump out. First, ChatGPT has shifted hard toward citing brand-owned product pages directly, nearly doubling its product-page citation share since March, according to Promptwatch's data on ChatGPT citation types. That's good news for retailers with clean PDPs, and bad news if yours are thin or JavaScript-heavy. Second, AI Overviews still leans heavily on listicles and how-tos, meaning comparison content and buying guides carry real weight there, based on Promptwatch's AI Overviews citation type breakdown. Video sits at just 5.9% in AI Overviews and effectively zero in ChatGPT, which is a format almost nobody is exploiting yet if you have the assets for it.
The practical takeaway: your product pages need to be citable on their own, and you need a layer of comparison and how-to content sitting above your catalog to catch AI Overviews queries that are more exploratory.
The citation slot problem
Here's something that doesn't get said enough: AI engines cite far fewer sources than a Google results page ever showed you. ChatGPT averages about 5 sources per web-search response, roughly half of what AI Overviews serves at around 10, and Perplexity sits close to AI Overviews at almost exactly 10, per Promptwatch's average sources per response data. Microsoft Copilot is erratic, sometimes swinging from under 2 to 17 sources within weeks.
That means