ChatGPT Shopping Ads: Product Feeds, Targeting & Bidding
ChatGPT Shopping Ads have caught my attention, particularly when it comes to product feeds and how much control advertisers will actually have over the products being promoted. We've spent years working with Google Shopping, Microsoft Shopping and Performance Max, learning how product data, bidding, targeting and campaign structures influence performance, so naturally I've got quite a few questions about how all of this translates into ChatGPT.
I've been digging through OpenAI's documentation, looking at the feed specifications, exploring the advertising libraries that have started appearing and trying to understand what advertisers can actually do within the platform. There are some familiar elements, but there are also differences that I think we need to understand before recommending it to clients, especially when we're talking about moving budgets away from channels that are already performing.
Things are moving pretty quickly with ChatGPT Ads. In fact, I've already had to update this article twice while writing it! We've had changes to bidding, geographic targeting and product selection, and some of the Help Centre documentation hasn't quite caught up with the announcements.
Rather than simply listing what's available, I wanted to explore how the platform works, what we can take from our existing Shopping experience and, more importantly, where we still have questions.
What are ChatGPT Shopping Ads, and how do they work?
ChatGPT Ads allow advertisers to appear in sponsored placements associated with relevant conversations, with product-feed campaigns providing a way for retailers to advertise individual products using structured catalogue information. The interesting difference compared with traditional Search is the amount of context ChatGPT can potentially use to understand what someone is looking for.
Think about how someone might shop for a sofa. They could start by asking which type would suit their living room, explain that they have pets, mention their budget and eventually narrow their search to a particular colour or style. Rather than treating each question as a completely separate search, ChatGPT can use the wider conversation to understand how their requirements are developing.
That opens up some interesting possibilities for advertisers, but it also raises questions about how products are matched to those conversations and what influence we actually have over that process. Having more context available doesn't automatically mean advertisers have more control.
It's also important to distinguish between ChatGPT's organic product recommendations and paid Shopping ads. ChatGPT can recommend products, display images, show prices and link to retailers without those placements being paid advertising. Sponsored ads should be clearly labelled, and we shouldn't assume that every product appearing in a shopping conversation has come from an advertiser's paid product feed.

Who can actually see ChatGPT Ads?
One of the first things I'd consider is the audience that's actually available to advertisers. ChatGPT Ads aren't shown across every account type, with advertising currently associated with eligible Free and Go users in supported markets. Plus, Pro, Business, Enterprise and Edu accounts aren't included, and Free users can also choose an ads-free experience with reduced usage limits.
This is particularly important when we're talking about reach. We can't take ChatGPT's overall user numbers and assume that every one of those people is available to advertisers, and I'd be especially cautious about doing that for B2B campaigns where the people we're hoping to reach may be using paid or business accounts.
The other consideration is that being eligible to see ads doesn't mean an ad will appear in every conversation, even when someone is clearly researching a product.
I tried finding a live Shopping ad myself
Naturally, I wanted to see what a sponsored product-feed ad actually looked like in the wild, rather than relying entirely on OpenAI's examples or third-party advertising libraries. So, I logged into a separate Free ChatGPT account and started testing shopping-related questions, beginning with a fairly broad request for corner sofa recommendations before narrowing the conversation down to blue corner sofas.
ChatGPT returned a selection of products from UK retailers, complete with images, prices, descriptions and links to their websites. What I found particularly interesting was how the recommendations changed as I refined my requirements, because it demonstrated the conversational shopping experience we're talking about. What I couldn't get it to show me, however, was a single sponsored product-feed ad!

Now, that doesn't mean the ads aren't running, and I'm certainly not suggesting that my handful of searches represents the entire platform. But it does highlight something worth remembering when researching ChatGPT Shopping Ads: seeing products in a conversation isn't proof that paid advertising has been triggered, and asking a shopping-related question doesn't guarantee that an ad will appear.
I've also been looking through third-party advertising libraries to find examples of what retailers are promoting, which I'll come back to later. For now, I'm still trying to catch a genuine sponsored product-feed ad within a live ChatGPT conversation!
Product feeds: how different are they from Google Merchant Center?
This was probably the first area I wanted to investigate properly because, for most retailers, the product feed is already doing a considerable amount of work across their existing advertising channels. Whether it's managed directly through Google Merchant Center, Microsoft Merchant Center, Shopify or a platform such as Channable, the last thing I'd want to recommend is building and maintaining another completely separate catalogue unnecessarily.
A lot of the core product information is familiar, although the field names and specifications aren't identical to Google Merchant Center.
ChatGPT field | Google Merchant Center equivalent | Purpose |
item_id | id | Unique product identifier |
title | title | Product title |
description | description | Product description |
url | link | Product landing page |
image_url | image_link | Main product image |
brand | brand | Product brand |
price | price | Product price |
availability | availability | Product availability |
seller_name | No direct universal equivalent | Seller or retailer name |
is_ads_eligible | No direct equivalent | Indicates advertising eligibility |
These are the core fields I'd start by reviewing, although the specification contains additional attributes relating to seller information, product details, destination markets and returns. It's worth distinguishing between fields that are required, those that are recommended and those included in example feeds, rather than assuming everything shown in a sample file is mandatory.
For advertising eligibility, is_ads_eligible needs to be set to true for products intended to participate in paid advertising. That doesn't guarantee those products will appear, but it is an important distinction between having product data available and making products eligible for advertising.
Would I simply reuse my existing Google Shopping feed?
I'd certainly start with it, but I wouldn't upload it without checking the differences first. We already know how much trouble inconsistent identifiers, missing attributes, incorrect availability and poor product information can cause in Shopping, so adding another destination without reviewing the data doesn't seem particularly sensible.
What I wouldn't do is start rewriting every title and description to make them sound more conversational simply because we're advertising on ChatGPT. We don't have enough evidence to say that approach improves paid product performance, and I wouldn't want to compromise an established Google Shopping feed based on assumptions about how another platform selects products.
I would, however, look at whether our existing descriptions provide enough useful information. If someone is asking about a product's suitability, dimensions, materials or particular features, is that information actually available in the data we're providing? That's a worthwhile question regardless of whether we're talking about Google, Microsoft or ChatGPT.
There's also the question of which attributes the platform uses when matching products to conversations. The feed specification tells us what information we can supply, but it doesn't necessarily tell us how much weight each attribute carries or how the system decides which product is most relevant. I'd be careful about making optimisation recommendations until we understand more about that process.
And what about product images?
This is another area I'm particularly interested in, especially given the changes we've been seeing across Google Shopping and Performance Max around how images are selected and displayed.
We know that product imagery is part of the ChatGPT feed specification, but I want to understand how much control advertisers have over which images are used, whether additional images can influence the advertising experience and what reporting is available to show which creative actually appeared.
I'd rather investigate those questions than immediately recommend a whole new image-testing strategy without knowing what the platform supports. For now, accurate, high-quality product imagery is an obvious starting point, but there's still more I'd like to understand about how image selection works within paid product placements.
How do we upload and manage product feeds?
Before getting into the different ways of submitting feeds for paid advertising, there's another development worth looking at because it gives retailers an opportunity to provide product information to ChatGPT without necessarily running ads.
What about ChatGPT Business Hub?
OpenAI launched ChatGPT Business Hub on 10 November 2026, giving businesses a way to claim and manage their information within ChatGPT. You can access it through https://chatgpt.com/business-hub, where retailers can set up their business profile, upload or connect their product feed and confirm their delivery and returns proposition as part of the setup process.
For retailers, I think the ability to provide accurate product information alongside delivery and returns details is particularly interesting, especially when shoppers are comparing products and deciding where to buy. The feed options include uploading a CSV file, providing a hosted feed URL or connecting a Shopify catalogue, depending on the options available to the business.
However, as Business Hub has only just launched, some businesses are finding themselves on a waiting list rather than getting immediate access, so don't be surprised if you can't complete the setup straight away.
The important part is that you don't need to be running ChatGPT Ads to do this. Business Hub is separate from the paid advertising platform, and uploading a feed doesn't guarantee that products will appear in organic recommendations, but it does give retailers another way to provide accurate product information to ChatGPT.
There's also an Insights section covering information such as business mentions, website clicks and discovery topics, although this shouldn't be confused with the campaign performance reporting we'd expect from an advertising platform.
For retailers already managing product feeds, I'd recommend investigating Business Hub regardless of whether they're planning to advertise. If we're already maintaining accurate product data, it makes sense to explore where else we can use it, particularly when there's an opportunity to do so without committing to an advertising budget.
Uploading feeds for paid advertising
For ChatGPT Shopping Ads, OpenAI supports several ways of providing product information, including manual CSV or TXT uploads, hosted HTTPS feeds, SFTP and API-based updates. Which approach makes sense will depend on the retailer's catalogue size, existing infrastructure and how frequently its product information changes.
For a small catalogue or an initial experiment, a manual upload might be manageable, but I wouldn't recommend relying on that for a retailer with hundreds or thousands of products where prices and stock levels change regularly. We've already learned from Shopping how quickly feed information can become inaccurate when updates depend on someone remembering to upload a file.
One detail worth highlighting is that products expire after two weeks without being refreshed. That isn't a recommendation to update your product information every fortnight, particularly if you're managing an active eCommerce catalogue, but it does mean that feed maintenance needs to be considered from the beginning. The Delta API also provides a way to update existing product information without repeatedly submitting an entire catalogue, which could be useful for retailers with more complex feed management requirements.
What about Channable and other feed management platforms?
Channable has announced support for ChatGPT Ads, which is encouraging because I'd much rather explore whether we can extend an existing feed management setup than introduce another disconnected process.
That said, I'd still want to check exactly what the integration supports, how the attributes are mapped, how frequently updates are submitted and what happens when products are rejected or information is missing. Having an integration available is one thing; understanding how it fits into the retailer's existing workflow is another.
I'd also want to establish who is responsible for monitoring the feed. If products stop updating, become ineligible or start returning errors, we need to know how those issues are surfaced and resolved. That's basic feed management, but it's also the sort of operational detail that can get overlooked when everyone is focused on testing a new advertising platform.
Can we choose which products to advertise?
Yes, product sets provide a way to filter which products are eligible for campaigns, and OpenAI has announced support for multiple conditions on the same field. This gives advertisers more flexibility when deciding which parts of their catalogue to include, although I'd still want to explore the exact filtering behaviour in Ads Manager before making assumptions about how more complicated conditions work.
For example, if I were managing a retailer with a large catalogue, I might want to focus on specific categories, price ranges or products with sufficient stock, rather than making everything eligible from day one. I'd also be thinking about profitability, seasonality and which products the business actually wants to prioritise, because a product being available doesn't automatically make it a good candidate for paid advertising.
The specification also includes ads_metadata, which is worth investigating for more tailored product information and selection. I'd want to understand how this interacts with product sets and what additional control it genuinely provides before building it into a wider feed strategy.
The distinction I'd keep coming back to is that we can influence which products are eligible, but that doesn't mean we can dictate which product appears in a particular conversation. That's an important difference, especially for advertisers who are used to creating tightly controlled Shopping structures.

Targeting: what control do advertisers actually have?
This is where I think PPC specialists need to be particularly careful about applying what we know from traditional Search.
We're used to working with keywords, match types, negatives, search term reports and bidding strategies, even if automation has changed how much direct control we have over some of those areas. ChatGPT has a different approach to relevance, and while the conversation provides additional context, it doesn't mean advertisers can target individual questions in the same way we'd target search queries.
Keywords and conversational context
OpenAI's advertising guidance describes context hints, which allow advertisers to provide topics, keywords and conversational context to help guide ad matching. That's useful, but these aren't traditional exact-match keywords, and including a particular term doesn't guarantee that an ad will appear whenever someone mentions it.
For example, a retailer selling outdoor furniture might want to provide context around garden furniture, outdoor dining or particular product categories. That could help communicate the types of conversations the advertiser is interested in, but I wouldn't treat it as equivalent to building a Google Search campaign around a set of keywords.
I'd want to understand how broad or specific those hints should be, what influence they have over delivery and whether the reporting gives us enough information to refine them. Otherwise, there's a risk of creating campaign structures that look familiar on paper without understanding how the platform is actually using the information.
This is also where search term visibility becomes important. If we're not receiving the same level of query reporting we're used to, we need to think about how we'll identify irrelevant traffic, understand what is driving results and make informed optimisation decisions.
Audience targeting and bid adjustments
Custom audience inclusion and associated bid adjustments require a minimum matched audience size of 25,000, while exclusions don't have the same minimum requirement. That's worth considering for advertisers with smaller customer lists or niche audiences, particularly if they were hoping to use highly specific first-party audience segments.
I wouldn't automatically start by creating lots of narrow audiences without first understanding whether they're eligible, how much reach is available and whether we're restricting delivery unnecessarily. We've seen similar challenges across other automated advertising platforms, where the desire for control can sometimes work against the amount of data available for optimisation.
It's also worth remembering that the audience eligible to see ChatGPT Ads is already smaller than ChatGPT's total user base, so targeting decisions need to take that into account.
Geographic targeting has already changed
This was one of the limitations that originally stood out to me, particularly when thinking about retailers or dealerships with a relatively small service area.
Take an agricultural machinery dealership in North Yorkshire. If it only serves a defined part of the country, targeting the whole of the UK wouldn't necessarily be commercially sensible, regardless of how relevant its products might be. That's quite different from a national eCommerce retailer that can deliver products across the country.
OpenAI has since announced support for more granular geographic targeting within product-feed campaigns, including regions and markets within countries. That's a welcome development and potentially makes the platform more interesting for businesses that don't operate nationally.
There is a complication, though. The Help Centre documentation has still referred to country-level targeting, so I'd check which geographic options are actually available within the account before promising a client that we can reach a particular region.
I'd also want to understand the level of precision available. Regional targeting is an improvement, but it isn't necessarily the same as being able to target an individual dealership's exact service area.
Bidding: what's available, and how would I approach it?
Bidding is another area that's already changed while I've been writing this article. OpenAI's November update renamed Maximise Results (Autobid) to Intelligent Bidding and introduced Target CPA and Target CPC for new campaigns from 9 November 2026.
These options replace the previous Max Bid and Max CPC approaches, with API support scheduled for 18 November and the legacy options due to retire on 1 January 2027.
For anyone familiar with Google Ads, the terminology will feel reasonably familiar, but I wouldn't assume that the underlying bidding systems behave in exactly the same way. We need to understand what each strategy is optimising towards, which signals are available and what reporting we can use to evaluate the results.
Bidding option | What I'd want to understand |
Intelligent Bidding | Which result is being maximised and how performance is evaluated |
Target CPC | Whether the average click-cost target is realistic for the campaign |
Target CPA | Whether there's enough reliable conversion activity to support the acquisition target |
It's also worth remembering that Target CPC and Target CPA are average targets, not maximum amounts. Setting a target doesn't mean every click or conversion will come in at that figure, and that's something I'd explain clearly when discussing budgets and performance expectations with clients.
Would I start with Target CPA?
Personally, I'd be cautious about using Target CPA immediately unless we already had reliable conversion activity and a good understanding of what the campaign was generating.
I generally like to see around 30 conversions per month before becoming too dependent on a Target CPA strategy. That's my own working guideline rather than an official OpenAI requirement, and it isn't a magic number that guarantees the bidding will work. The quality of the conversion data and the campaign objective matter just as much as the volume.
With a new platform, I'd want to understand the traffic we're receiving, whether conversions are being recorded correctly and whether we have enough meaningful activity to establish a sensible target. Otherwise, we could end up repeatedly adjusting CPA targets without knowing whether the real issue is the available audience, the products, the campaign setup or the tracking.
Budgets and spending
OpenAI's average daily budget approach is another detail I'd make sure clients understand before launching. Campaigns can spend up to twice their average daily budget on an individual day, with spending averaged across a seven-day period.
That's not necessarily a problem, but it does mean we need to explain how spending can fluctuate rather than allowing a client to assume their daily budget is a strict daily cap.
For any initial test, I'd agree on the budget, the objective and the points at which we're going to review performance. I don't think we need to overcomplicate the process, but I also wouldn't want to launch campaigns without knowing what we're hoping to learn or how much we're prepared to spend finding out.
Conversion tracking and reporting
Before recommending ChatGPT Shopping Ads to a retailer, I'd want to understand exactly what we're going to be able to measure. That includes how purchases are recorded, whether revenue values are being passed correctly, what attribution information is available and how much visibility we have at product level.
OpenAI's documentation now describes product-level reporting, which is encouraging, but I'd still want to explore which metrics and dimensions are available and how useful they are for actual optimisation. Knowing that a product generated activity is helpful, but the questions I'd be asking are whether we can identify what is performing, understand where we're spending money and use that information to make better decisions.
I'd also want to compare the data with the retailer's own analytics and sales reporting rather than relying entirely on the advertising platform's attributed conversions. That's particularly important when we're testing a new channel and trying to understand whether it's bringing additional value or simply receiving credit for sales that might have happened elsewhere.
UTM tracking is an obvious part of that process, although there's a detail worth checking: UTM parameters configured within Ads Manager can override matching parameters already present in product URLs. If a retailer is already adding tracking information through its feed management platform, I'd review how those parameters interact before launching.
There's also a wider question about how shoppers use ChatGPT. Someone researching products through a conversation may be at a different stage of their decision-making process from someone searching Google for a specific product they already intend to purchase. That doesn't mean we should accept poor performance or abandon our usual measurement standards, but I do think we need to understand what role the channel is playing before deciding how it should be evaluated.
For me, that means agreeing on the purpose of the campaign before we start. If we're testing direct sales, then purchases, revenue and acquisition costs are going to matter. If we're exploring whether ChatGPT introduces new shoppers to a retailer, we'd need to be clear about how we're going to assess that contribution as well.
Can we see what competitors are advertising?
Of course, one of the first things I wanted to do was find out which retailers were already advertising and what their ads looked like.
There are a few third-party advertising libraries starting to track ChatGPT Ads, including:
AirOps Ads Library: https://www.airops.com/ads-library
Focal Ad Library: https://www.tryfocal.com/product/ad-library
LLM Pulse ChatGPT Ads Library: https://llmpulse.ai/chatgpt-ads-library
I've been exploring these to look at the advertisers being recorded, including the Wayfair examples shown earlier. It's useful to see real product information and creative rather than relying entirely on platform illustrations, particularly when we're trying to understand how retailers are approaching the format.
However, I wouldn't treat an ad library as a complete picture of what's happening. Finding an advertiser doesn't necessarily tell us where the ads were shown, which conversations triggered them, how frequently they appeared or whether the campaigns were performing well. We also need to distinguish between an advertising library record and an actual screenshot of a paid placement within ChatGPT.
That's why I've been trying to trigger one myself. I'd like to see the actual shopping experience, including how the sponsored products appear alongside the conversation and what happens when someone interacts with them.
For now, I'd use these libraries as a research tool rather than a definitive source of competitor performance or targeting information.
Would I recommend testing ChatGPT Shopping Ads?
I think there's a case for testing, particularly for retailers that already have strong product data, reliable conversion tracking and some budget available for experimentation. But I wouldn't be recommending ChatGPT Shopping Ads to every client simply because it's a new advertising opportunity.
My starting point would be the existing feed and the commercial objectives of the business. If we're already dealing with inaccurate product information, tracking problems or campaigns that aren't performing as they should, I'd rather address those issues than introduce another advertising platform and hope it solves them.
For a retailer with a well-managed catalogue, I'd begin by reviewing the ChatGPT feed requirements against what we already have, identifying the products that make commercial sense to advertise and checking the available targeting options. I'd also want to understand whether our existing feed management platform can handle the additional destination, because I don't want to create unnecessary work maintaining multiple versions of the same catalogue.
The next question would be budget. If a retailer has profitable Google Shopping or Performance Max campaigns, I wouldn't automatically recommend taking money away from them to fund a ChatGPT experiment. We'd need to consider what we're trying to achieve, how much we're prepared to invest and whether the potential learning justifies the opportunity cost.
I'd also be careful about promising immediate performance. We need enough activity to understand how the campaigns deliver, which products receive exposure and whether the traffic is commercially useful. At the same time, testing shouldn't become an excuse to keep spending without reviewing the results.
What I'd really want to establish is whether ChatGPT Shopping Ads can reach shoppers we aren't already reaching effectively, whether we have enough control over the products we're promoting and whether the reporting allows us to make meaningful optimisation decisions.
Those are the questions that matter to me far more than simply being able to say we've launched campaigns on another platform.
🔹 Final Thoughts
I'm interested in where ChatGPT Shopping Ads go next, particularly around product feeds, image selection, targeting and the reporting available to advertisers. There are some familiar elements that should make the platform easier for experienced Shopping specialists to understand, but I think we need to be careful about assuming everything works in the same way as Google or Microsoft Ads.
The announcements are coming quickly, and I'm sure there'll be more functionality to investigate as the platform develops. I'll continue looking at the documentation, exploring what advertisers can actually control and testing the shopping experience wherever possible, because I'd much rather understand how something works before recommending it to a client.
And if you've managed to trigger a genuine sponsored product-feed ad inside ChatGPT, I'd love to hear about it! I've managed to get plenty of sofa recommendations so far, but I'm still waiting to catch a paid one in the wild.
And remember, Don't Panic! You don't need to jump onto every new advertising platform the moment it launches. Get your product feeds in order, understand what you're actually able to control and make sure any testing has a clear purpose. There's plenty to explore with ChatGPT Shopping Ads, but that doesn't mean we need to throw our existing PPC strategies out of the window!
4M Digital is a Google and Microsoft Ads Partner specialising in paid search and paid social, from Google and Microsoft Ads to Meta, LinkedIn, Pinterest and TikTok. Working in paid media since 2010, we help brands, retailers and agencies grow through campaign management, audits, training and white-label PPC support.





