August 7, 2026
Your product detail page has always had one job: convince a human shopper to buy. Now it has a second audience deciding whether ChatGPT, Gemini, and other AI chatbots ever show that product to a human shopper in the first place.
According to Bazaarvoice’s research of over 3,000 shoppers, 83% have used AI in the past six months, and 43% used it specifically for shopping. That second audience is already here.
AI is changing how shoppers discover products, and it’s opening a new front door for every brand, one that runs on different rules than traditional search.
So here’s the question worth asking. Not “is my site optimized for AI search?“, but something more concrete: “can ChatGPT find enough accurate product data and real customer evidence to represent my product properly?“
That question applies whether ChatGPT is answering a head-to-head comparison or a needs-led query like “what’s the best power bank for a long hike?” Winning here has little to do with keyword density. It comes down to data clarity and customer evidence, the two things an AI-first web rewards most.
How ChatGPT Shopping finds products
When a shopper’s prompt suggests shopping intent, ChatGPT Shopping pulls product information from a few different places: direct product feeds submitted by merchants, brand and retailer product listings found through web retrieval, plus structured metadata and third-party content like reviews.
OpenAI’s own guidance confirms this. When ChatGPT picks which products to show, it draws on a mix of sources: structured product data supplied by brands and retailers, third-party content like reviews, and its own safety standards — weighed alongside what the model already knows before it goes looking for anything new.
There isn’t one fixed formula. Which of these matters most shifts with the shopper’s prompt and context: a budget-conscious query pulls price data forward, a needs-led query pulls use-case content forward. Your product needs to hold up inside AI-generated answers under a range of prompts, not just rank for one target keyword.
Ratings and reviews carry more weight in that mix than most teams expect. A Yale and Columbia research into how AI shopping agents actually choose products found that a small uptick in a product’s star rating or review count swayed an agent’s decision as much as a meaningful price cut.
Product feeds and PDPs play different roles
Profound’s June 2026 analysis of roughly one million ChatGPT Shopping offers gives the clearest public picture yet of how these two retrieval routes perform. It also settles the feeds-versus-PDPs debate in a useful way: the answer is both. This is observational research from Profound, so it shows what currently correlates with strong results — a useful signal, even though (as with any single study) it isn’t a guaranteed formula.
Product-feed retrieval means ChatGPT is pulling product details straight from a structured feed a merchant submits — similar to the feeds used for Google Shopping or other retail media — rather than crawling and interpreting a webpage on its own. And it’s growing fast, performing disproportionately well: feed-derived citations grew from 4.3% to roughly 20% of shopping retrievals over the study period, and around 99.9% of feed-derived citations landed in the first offer position.
A direct feed gives merchants more control over what ChatGPT sees — accurate pricing, availability, and structured detail, refreshed on your schedule rather than whenever a crawler next visits your web pages. Setting this up isn’t automatic: a merchant needs to build and submit a feed, then keep it updated, but doing so gives you more control over what ChatGPT sees — accurate pricing, availability, and structured detail, refreshed on your schedule rather than whenever a crawler next visits your web pages.
PDPs still carry most of the volume, though. Put simply: of every time a product actually showed up as an offer in ChatGPT Shopping, 88.29% of those instances came from a web PDP rather than a feed. That holds true even for merchants who already have a feed live — 75.81% of their offers still came from PDPs. In short: feeds tend to win the top spot when they’re used, but PDPs are still doing most of the heavy lifting.
The takeaway isn’t that feeds make PDP work optional. A feed helps you win the top slot for the SKUs you’ve structured. Your PDPs are still doing the job of representing your full catalogue, including products you haven’t fed directly. And because ChatGPT also pulls third-party content like reviews from around the web, your PDPs are carrying the customer evidence that structured product data alone can’t provide.
One more note on the numbers above: they come from a single 30-day snapshot in June 2026, so treat the specific percentages as directional for your longer-term content strategy rather than fixed targets to hit.
What this means for brand and e-commerce teams
Reading those two data points side by side shows AI visibility isn’t one team’s job.
E-commerce teams carry the structured side of the problem: product-feed accuracy, PDP accessibility, and consistent identifiers and pricing that stay current everywhere ChatGPT might look for them.
Brand teams carry the context side: making sure product positioning, audience, and differentiators come through clearly enough for a model to use them — not buried in a hero image or a product description built entirely from adjectives a crawler can’t parse.
Neither team solves this alone. Clean structured data with no context reads as accurate but generic. Rich brand storytelling with no accessible structure may never get retrieved at all. This is where Ratings & Reviews sits across both sides: it gives e-commerce teams structured, schema-ready data ChatGPT can parse, while giving brand teams the authentic customer voice and context that data alone can’t carry.
Five ways to improve ChatGPT Shopping visibility
1. Audit your current visibility
Action: Run a consistent set of discovery, comparison, and needs-led search queries through ChatGPT and other AI chatbots — “best power bank for a long hike,” “[your product] vs [competitor],” “affordable options for X.” Track whether your products appear, how they’re described, and which source — feed or PDP — ChatGPT is pulling from.
Why: You can’t fix what you haven’t measured. This is the fastest way to see whether you’re even in the running for the queries your customers are typing into these tools.
2. Improve product-feed quality
Action: Prioritize accurate identifiers, factual descriptions, current pricing and availability, and correct variants — updated on a regular cadence, not just at launch. Direct feed access and regional availability are still evolving, so confirm what’s actually open to your catalogue today.
Why: Feed-derived citations are landing in the first offer position at a striking rate. A clean, current feed is the single highest-leverage lever available if you can access it.
3. Strengthen PDP accessibility
Action: Use clear titles, explicit use cases, accurate specifications, and valid structured data — schema markup that tags Product, Review, and FAQ data helps ChatGPT parse your aggregate rating, review count, and pricing/availability accurately, rather than inferring them from prose. Work with technical teams to confirm crawler access and check that JavaScript-dependent content actually renders for a bot, not only a browser. Use internal links between your PDP and supporting content — buying guides, comparison pages — so both shoppers and crawlers can find the fuller picture.
Note: OAI-SearchBot and GPTBot are not the same control. OAI-SearchBot is what surfaces your pages in ChatGPT search and shopping features — that’s the one to allow if visibility is the goal. GPTBot governs whether your content may be used to train OpenAI’s models, a separate decision. Per OpenAI’s crawler documentation, the two settings are independent — allowing one doesn’t require allowing the other.
Why: PDPs still account for the large majority of product-offer instances, feed or no feed. If OAI-SearchBot can’t reach your pages, none of the rest of this matters — and if it can’t see your JavaScript-rendered content, ChatGPT effectively can’t see it at all.
4. Build useful customer trust surfaces
Action: Encourage descriptive reviews, customer Q&A, FAQs, and visual UGC — including social media posts and photos — that explain real use cases, fit, and limitations: the kind of detail a product description alone won’t cover.
Why: Reviews and Q&A are exactly the third-party content OpenAI’s own guidance says ChatGPT draws on alongside structured metadata. Bazaarvoice’s research found that 62% trust product recommendations most when they’re backed by reviews and photos from real, verified purchasers. Treat this content as valuable context that helps ChatGPT represent your product accurately — not as a guaranteed ranking factor. Neither the Profound research nor OpenAI’s documentation supports that stronger claim.
5. Check consistency across the digital shelf
Action: Compare your product information and customer content across your feed, your own PDPs, and the retailer sites where you’re listed. Review syndication — pushing your verified reviews out to retailer PDPs, not just your own — is one practical way to strengthen that off-site coverage and grow organic traffic from AI-referred sources.
Why: ChatGPT pulls from PDPs across the web, not just yours. A product that looks well-documented on your own site but thin or inconsistent everywhere else is still a product AI struggles to represent well.
What not to assume
- A product feed guarantees the top spot. Feeds correlate strongly with first-offer placement, but nothing in OpenAI’s own guidance describes a guarantee.
- You need to allow GPTBot to show up in ChatGPT Shopping. OAI-SearchBot governs search and shopping visibility; GPTBot governs training. They’re independent settings.
- Schema markup can compensate for thin product information. Structured data organizes what’s there — it doesn’t invent detail that’s missing.
- More reviews automatically mean better AI visibility. Descriptive, specific reviews function as useful context; volume alone isn’t documented as a ranking input.
- A live product feed makes PDP optimization optional. PDPs still account for the majority of product-offer instances, feed or no feed.
Winning the recommendation era
ChatGPT Shopping visibility isn’t a single technical fix you implement once and move on from. It’s ongoing coordination across product-feed accuracy, PDP content, and the customer evidence only real shoppers can provide — and it will keep changing as the underlying research and OpenAI’s own guidance evolve.
We think about this as three requirements working together, our Triple-A framework: Accessible (can ChatGPT technically reach and parse your data), Authentic (is your customer evidence genuine and specific enough to be useful), and Abundant (is there enough of it, consistently, across every place ChatGPT might look). Miss any one of the three and the others can’t fully compensate.
For related reading on how this plays out beyond ChatGPT specifically, see our blog post on how to improve brand visibility in AI search engines.
So ask yourself: if ChatGPT summarized your product pages right now, would it find the data and the customer evidence it needs to recommend you?
Audit your top 10 product pages today, and start with the ones you most want ChatGPT to surface.
