Product Feeds Are Rising. Your PDP Still Decides Whether You Show Up in AI Shopping

By Stephen Honight, Founder of Lmo7

There is a quiet worry going around ecommerce teams right now. ChatGPT has started pulling products from direct merchant feeds, and the assumption is that the product page is on the way out. Get your feed plumbed in, the thinking goes, and the listing work stops mattering.

My sense is that read is half right and the wrong half is the bit people are acting on.

New research from Profound, published 24 June, looks at how ChatGPT actually retrieves product information when its shopping mode triggers. The headline most teams should take from it is the opposite of the worry. Feeds are rising fast. The product page is still doing most of the work. Around 88% of the product offers ChatGPT shows still come from a crawl of the web product page, not the feed.

So before anyone reallocates their content budget toward feed integration alone, it is worth understanding the mechanism underneath. That is where the real decision sits.

How ChatGPT actually finds products

When you ask ChatGPT something with buying intent, shopping mode can kick in and it returns a small carousel, usually three to five products. To fill that carousel it pulls product information from two different places.

The first is a web crawl of product detail pages. ChatGPT reads the live PDP the same way it reads any page, and lifts what it can from the content there. This is the surface most brands already have, whether they have thought about AI shopping or not.

The second is a direct product feed. Through the OpenAI partnership, merchants can pipe structured product data straight into ChatGPT. No crawl needed. The data arrives clean and labelled.

These are two very different inputs. A crawled page gives ChatGPT whatever it can interpret from your live content. A feed gives it a tidy, structured record with fields already named. That difference matters more than it first looks, and I will come back to it.

What Profound found

Profound ran this across serious scale. Their 30-day snapshot covered roughly one million shopping product offers across ChatGPT sessions, with about 6,000 cited product URLs scraped for extra PDP detail. They also ran an eight-month look-back across around 548 million shopping offers to track the trend in retrieval source. It is observational platform data, a snapshot in time, so treat the numbers as Profound’s findings rather than fixed laws. They are still the clearest read we have on how this works today.

A few things stand out.

Feed retrieval is climbing quickly. Profound report feed share rising every month since late November 2025, from 4.3% to about 20% of all shopping retrievals in the most recent six weeks. That is roughly 15x growth in a year. The direction of travel is not in doubt.

When a feed is present, it tends to win the top spot. Profound found that around 99.9% of citations that came from a direct feed appeared as the first product offer in the carousel, across more than 100,000 instances in the month. Structured feed metadata is a strong signal for rank. If you are in the feed, you are usually at the front.

That sounds like a clean case for feeds and nothing else. Here is the part that complicates it.

PDPs are still where most offers come from. In Profound’s month snapshot, 88.29% of all product offer instances were sourced from web product pages. Even among the roughly 150 merchants who have already integrated a feed with ChatGPT, 75.81% of their offers were still derived from the PDP crawl.

Read that again, because it is the whole point. Brands that have done the feed work still see three quarters of their ChatGPT shopping appearances coming from their product page. The feed is additive. It is not a replacement.

Why ChatGPT sees a different brand depending on the source

The reason both surfaces matter comes down to what each one actually exposes. Profound compared how often specific product features were present, split by whether the offer came from a PDP crawl or a feed. The gap is large.

From a feed, brand name was present 100% of the time. From a PDP crawl, it was present 0% of the time in their sample. Checkout image URL and merchant subtitle were the same story, 100% on feeds and 0% on crawled pages. ChatGPT’s own ‘best price’ tag, which is the machine-assigned label for the lowest price it can find for a product, showed up on 100% of feed offers and only 21% of PDP offers.

It runs the other way too. Delivery information was present on about 75% of PDP offers and only 4% of feed offers. Online availability showed on 79% of PDP offers and 0% of feed ones. Free delivery appeared on half of PDP offers and none of the feed offers in the sample.

So the feed gives ChatGPT a clean, branded, price-tagged record. The PDP gives it richer signals around availability and delivery that the feed often misses. Neither surface tells the full story on its own. A brand that only feeds, or only crawls, is handing ChatGPT a partial picture either way.

This is why the either/or framing is the wrong one. The question is not feed or PDP. It is how much of each, and in what order.

The two-track read

At Lmo7 we keep coming back to a two-track way of looking at AI search work. There are the things you can fix quickly that you control directly, and there are the things that take longer and compound. This Profound data maps onto that almost perfectly, so it is worth saying plainly.

Track one is the feed. If you can integrate a product feed with ChatGPT, do it. It wins rank when present, it locks in your brand name and your best-price eligibility, and it gives you a clean structured record that does not depend on a crawler interpreting your page correctly. For brands on Shopify in particular, where Profound found feed offers concentrated, this is the more controllable surface.

Track two is the PDP. This is the surface most of your AI shopping appearances are actually coming from right now, feed or no feed. It is also the surface that carries the delivery, availability and trust signals the feed tends to drop. And it is the same content substrate that Amazon’s Rufus reads, that Google’s AI Overviews read, and that every other AI shopping surface reads. Work done on the product page pays off in more than one place at once.

My recommendation would be to treat these as two tracks running together, not a choice. Integrate the feed where you can. Keep optimising the PDP regardless. The brands that win the AI shopping carousel over the next year will be the ones doing both, not the ones who treated the feed as a reason to stop improving the page.

The three PDP levers Profound calls out

Profound set out three practical ways to improve how a product page performs in ChatGPT shopping. They are sensible, and they line up closely with the content work we already scope for clients, so I will give you their version and then how I would apply it.

The first is to maximise your chance of the ‘best price’ tag. ChatGPT assigns that tag to the lowest price it can find for a product, and it leans on it. Profound suggest monitoring competitor merchants selling the same product, holding the lowest price where you realistically can, and adding clear best-price or value wording on the page. For most brands this is as much a pricing and ops question as a content one, so it needs the people who set price in the room.

The second is to build visible trust surfaces on the page. Descriptive, usage-based reviews. Embedded FAQs, Q&As, videos and images. The point is to give ChatGPT plenty of concrete, human-shaped evidence to read and quote. This is the same logic behind the Q&A and review work that lifts a product in Rufus. The machine is looking for clear answers to the questions a shopper would ask, and a page that already contains them is easier to cite.

The third is clearer product naming. Descriptive titles that name the common use cases and the real features, without getting so niche that the page only matches one obscure query. We have written before about moving from pipe-stuffed SEO titles to titles that read cleanly to both a shopper and a model. Profound’s data is another reason to do it. A clear name is easier for ChatGPT to match to a buying prompt.

None of these are exotic. That is the reassuring part. The work that earns a citation in ChatGPT shopping is mostly the same product content work that earns a recommendation in Rufus and a mention in an AI Overview. You are not building a separate AI shopping programme. You are making the product page genuinely good, and the page is now doing several jobs.

Where this shows up in the work we do

We see this pattern in live client work. On the Amazon side, the PDP content optimisation we have run for Veloforte, where the work is matching listing content to how shoppers actually search and how Rufus reads a page, is the same substrate this Profound research is pointing at on ChatGPT. Improve the product page so a machine can understand it, and you are improving your odds across every AI shopping surface at once rather than the single one you had in mind.

The same is true on the trust-surface point. Reviews, Q&A and clear claims do double duty. They reassure a human shopper and they give a model something concrete to extract. A page built only for a human skim reads thin to a machine. A page built only for a machine reads cold to a human. The pages that win are the ones that work for both, which has been the through-line of most of our content work for the past year.

The honest caveat sits alongside all of this. The Profound numbers are a 30-day platform snapshot, observational, and they will move. Feed share was 4.3% not long ago and is around 20% now. If that line keeps climbing, the balance between the two tracks shifts and the feed gets more important over time. The right posture is not to bet the budget on today’s split. It is to hold both surfaces in good shape so you are covered whichever way the retrieval mix moves.

What to do next

If you take one thing from the Profound data, take this. The feed era has not killed the product page. Around 88% of ChatGPT shopping offers still come from the PDP, and even feed-integrated brands see three quarters of theirs from the page. The product page is the surface you most control and it is still the surface doing the most work.

Here is how I would sequence it.

Start with the feed. Check whether a product feed integration with ChatGPT is open to you, especially if you are on Shopify, and get it in if it is. It wins rank when present and it locks in your brand and best-price eligibility. This is the quick, controllable win.

Then audit your top product pages against the three levers. Are you price-competitive enough to earn the best-price tag on your hero SKUs. Do your pages carry real reviews, Q&A and clear images. Are your titles descriptive and clean rather than keyword-stuffed. Fix the worst offenders ahead of everything else, starting with your highest-revenue products.

After that, hold the PDP as one content layer feeding several AI surfaces rather than a one-off task. The same page work lifts you in ChatGPT shopping, Rufus and AI Overviews. Build it once, properly, and review it on a cadence as the retrieval mix shifts.

If you want to run that PDP and feed work yourself, our DaaS tier gives you direct access to the underlying Amazon, Shopify and search data plus the upskilling to read it, from £250 plus VAT a month. If you would rather we did the hands-on content and feed work for you, that is the Challenger Content Optimisation and Agentic Ops modules, which is where most single-brand consumer businesses land. And if you are sitting on a multi-brand catalogue and need this assessed and prioritised across the portfolio before anyone touches a page, that is an Enterprise piece, and the place to start is a workshop that gets your stakeholders aligned on where the gaps are.

Wherever you start, the move is the same. Get the feed in where you can, and keep making the product page good. Right now the page is still deciding whether you show up.

Stephen Honight is the founder of Lmo7, an AI-native agency helping consumer brands win in AI-powered discovery and agentic commerce across Amazon and D2C. Lmo7 works with brands including Trip Drinks, Veloforte, Brown-Forman, Haleon, Pelotan and Symprove on AI search visibility, Amazon and marketplace AI, content for AI commerce, and agentic enablement.