E-commerce··22 min read

Product Feed Optimization for AI Shopping: What to Look For

Thinking of hiring help with a product feed for AI shopping assistants? What data is needed, common mistakes, how the work runs, and what drives the cost.

Short answer

Getting a product feed ready for AI means turning every item in your store, with its title, price, availability, barcode, attributes and image, into a clean, consistent list that machines can read without guessing. Assistants such as ChatGPT, Gemini and Perplexity, and shopping engines such as Google Shopping, lean on structured data like this and on your product pages when they recommend something to buy. Missing or contradictory details can keep your products out of those answers entirely. When you hire someone for this, you are not looking for a person who exports a file once and walks away. You want someone who lines up the feed with your product pages and sets up a system that updates itself when prices and stock change. One option is me, Ümit Kırkoç (kirkoc.com). I am based in Istanbul and work remotely with store owners abroad, and because I handle the online store, SEO, advertising and AI visibility work (GEO) myself, I treat the feed as one piece of a connected whole rather than a standalone file.

What is a product feed, and why does it matter more now?

A product feed is a data file that lists the products in your store like a spreadsheet instead of as separate web pages. Each row is one product, and each column holds one fact about it: the title, the price, the color, the size, the quantity in stock, the barcode, the link to the image. The file usually comes as XML, CSV or a URL that other systems visit on a schedule, read, and use to show your products on their side.

For years, the main readers of this file were Google Merchant Center, Meta's product catalog, and marketplace integrations for places like Amazon or Walmart. Recently a new reader joined them: AI assistants. When someone asks ChatGPT to "recommend a quiet coffee grinder under my budget," the assistant builds its answer by looking at product data, shopping engine listings and store pages. Some platforms have also started opening ways for merchants to submit product data directly. What those programs cover differs by country and platform, and it changes often, so it is worth checking the current terms before you rely on any of them.

Here is the part that affects you. A human browsing a product page can fill in a missing detail or look at the photo and figure out the color. A machine cannot always do that. If one field says "navy," another says "dark blue" and a third says "midnight," the system may not realize they mean the same thing. No matter how good your product is, messy data makes it harder for that product to land on a recommendation list.

The image below uses an example scenario to show how an assistant's answer to a shopping question rests on well-organized product information.

Example chat in which a shopper asks an AI assistant for a camping tent and the assistant highlights the product with clear weight, waterproof rating and stock details

Everything the assistant says in that answer (the weight, the waterproof rating, the stock status, the setup time) comes from a feed field or from a clearly stated fact on the product page. Take those details away, and the assistant will pick a different product whose information is complete.

What problems do store owners run into?

The most common problem I see is a feed that exists but that nobody ever opens. Your store platform (Shopify, WooCommerce, BigCommerce, Wix and others) generates a feed URL. Someone connects it to Merchant Center or a marketplace once, and then months go by without anyone checking it. Meanwhile products change, categories get reorganized, and sales start and end, while the feed keeps carrying the same mistakes it had on day one.

The second problem is product titles written for the warehouse rather than for people. A stock code like "TSHRT-0423-NVY-M" or a vague title like "New Season Tee" tells neither a search engine nor an assistant anything useful. A shopper talking to an assistant says "men's cotton crew neck t-shirt in navy." If your title and attribute fields do not contain those words, no match is made.

The third problem is missing barcode and brand information. The GTIN, a global product number (on most retail items in the US, it is the 12-digit UPC under the barcode), is what lets systems recognize that this item is that exact item. If you sell products you make yourself, you may not have one, and in that case you need to say so in the correct way. Entering a wrong or made-up barcode can get your products rejected.

The fourth problem is a conflict between the feed and the product page. If the feed shows one price and the page another, or the feed says "in stock" while the page says "sold out," systems like Google Merchant Center can suspend the product. To an AI assistant, contradictory information also reads as a sign that the data cannot be trusted.

How do your customers actually ask?

People who ask an AI about a product use different language than they type into Google. Someone who searches Google for "women's running shoes size 8" might ask an assistant, "My knees hurt and I run on pavement. Can you suggest a comfortable running shoe in a size 8?" That one question includes the purpose, the problem, the surface and the size. If your feed and product page mention cushioning, road running and a wide base, the assistant has a better chance of connecting you to that question.

That is why preparing a feed is not just a technical export job. You need to know what customers in each product group want to find out and put that information into your descriptions and attribute fields. For a furniture seller, the answer is dimensions and materials. For a cosmetics brand, it is skin type and ingredients. For an electronics store, it is compatibility and warranty length.

What information should a product feed contain?

A good feed holds at least these details for every product, complete and consistent: a descriptive title, a clear description, the current price, availability, brand, barcode (if one exists), category, main image and variant information. On top of that come extra fields that depend on the product type, such as size, color, material, age group, gender, weight and dimensions.

The table below sums up the most important fields, what each one does, and the mistake I see most often with each.

Table showing what the product title, price, availability, barcode and main image fields do and the common mistake made in each

When even one of these fields is wrong, the damage rarely stays with a single product. It usually affects hundreds of items generated from the same template. So fixing a feed is more often about correcting the rule in your store platform than about editing products one by one.

How should product titles be written?

A product title should carry the most important facts in the order a shopper looks for them: brand, product type, the feature that sets it apart, then variant details like color or size. "Wood Table" says very little. "Solid Oak Dining Table 63 x 35 in, Natural Finish" tells both a search engine and an assistant far more. Hype words ("Best ever!", "Don't miss out!") and promotional phrases in all caps break the rules on most platforms.

How long should a description be?

Content matters more than length. A description should explain what the product is, who it suits, what it is made of, how it is used and how to care for it. Copying the description your supplier sent means sharing the exact same text with dozens of other stores selling the same item, which makes it much harder for anyone to tell you apart.

What should the images look like?

On the main image, the product should appear clearly, on its own, against a plain background. Images with added text, logos, discount badges or borders are rejected by many shopping platforms. Additional images can show the product in use, close-up details and other angles.

What is the difference between a weak feed and a strong one?

A weak feed fills in the fields a system marks as "required" and leaves the rest empty. A strong feed also carries the answers to the questions a customer is likely to ask. Both technically "work," but only one of them has a real shot at appearing in recommendations.

The comparison below shows how the two approaches can look for the same product.

Comparison of a weak and a strong product feed across title, description, barcode, variants and update frequency

A store running the feed on the left can honestly say "my products are on Google," yet when an assistant is looking for an answer to a specific question, it will not find enough there to bring that product forward. The right side takes more effort, but most of that effort becomes automatic once the rules are set up.

Two example scenarios

The two scenarios below are hypothetical. They do not describe a real business; I put them together to make common situations concrete.

Example scenario 1: A small shop selling handmade ceramics

Imagine a small business that makes mugs, plates and vases in its own studio. It sells on a WooCommerce site, and a plugin generates the feed automatically. The product titles are short, like "Blue Mug" or "Large Plate." Descriptions are a few words long, and none of the items has a barcode, because each one is handmade.

The first step here is to tell the platforms correctly that there is no barcode and to enter the studio's name in the brand field the same way every time. Next, the titles are rewritten along the lines of "Handmade Ceramic Mug 12 oz, Cobalt Blue Glaze." The descriptions get details like whether the piece is dishwasher safe, whether it can go in the microwave, and a note that each piece may vary slightly because it is made by hand. Those are exactly the facts an assistant looks for when someone asks for "a handmade mug that's dishwasher safe, as a gift."

Example scenario 2: An electronics accessories store with thousands of products

Imagine an online store that sells phone cases, chargers and earbuds, with a catalog of several thousand items. Products arrive through a supplier integration, and prices and stock change several times a day. In Merchant Center, a large number of items carry warnings for "price mismatch" and "missing compatibility information."

Fixing products one by one makes no sense here. The first check is how often the feed refreshes: if prices change several times a day but the feed is read once a day, mismatches are unavoidable. Then rules are put in place. For phone cases, the compatible phone model is added automatically to the title and to a separate field. For chargers, output power and connector type are written in one standard format. Once those rules exist, every new product that comes in follows the same structure.

How does the feed preparation process work, step by step?

The process runs in five steps: review the current state, define fields and rules, enrich the product data, connect the feed to platforms and validate it, then monitor it on a regular basis. At the end of each step, I tell you what was done and what changed.

Diagram of the product feed process showing the review, rules, enrichment, connection and monitoring steps

The steps depend on each other. Setting rules without a review first just carries the existing mistakes into the new feed.

1. Reviewing where things stand

First I look at the feed your store platform produces, the warnings in your Merchant Center or Meta catalog account, and a handful of sample product pages. I note which fields are empty, which products have been rejected, and whether price and stock match the page. I also ask AI assistants the kinds of questions your customers would ask about a few of your product groups and see how you and your competitors show up in the answers.

2. Defining fields and rules

For each of your product groups, I work out which fields are required and which are simply useful. The title template, the standard for writing colors and sizes, and the category mapping all get settled at this stage. For example, one approved list of color names is set, and the "navy" versus "dark blue" versus "midnight" mix-up disappears.

3. Enriching the product data

Missing descriptions, attribute fields and image problems get handled here. There are two ways to do it: fix the product information itself in your store, or fix it with rules on its way into the feed. The first is the lasting one, because your product page gets stronger with the same information. This step also covers structured data on your product pages, which is code added behind the page that tells machines the product's price, stock and rating in a format Google and other systems can read.

4. Connecting to platforms and validating

The feed is connected to Google Merchant Center, to your Meta catalog if needed, and to any other relevant channels. The refresh schedule is set to match how fast your products change. After the first upload, each warning that comes back is reviewed and fixed at its source.

5. Monitoring regularly

A feed that is left alone after setup breaks down again. It needs checking whenever new products go live, a sale starts, or your store platform gets an update. You and I decide together how often to monitor it, based on how many products you have and how quickly they change.

What are the most common mistakes?

The most common mistake is thinking of the feed as something you set up once and are done with. Beyond that, a few mistakes come up again and again, and most of them go unnoticed for months.

Warning list of common product feed mistakes: old price in the feed, made-up barcodes, text on the main image, copied descriptions and ignored warnings

What these mistakes have in common is that none of them looks like a big problem at the moment it happens. Products slowly lose visibility, and the store owner never quite understands why.

Let me add a few notes to that list. During a sale, if the discounted price is not sent as its own field in the feed, the platform keeps showing the old price while your product page shows the discount. If variants (the different colors and sizes of the same product) are sent as one product instead of separately, the size a shopper wants may appear unavailable. And there is the case of two or more plugins generating feeds at the same time: when nobody knows which file feeds which platform, fixes end up being made to the wrong file.

There can also be legal points to watch, such as health claims in product descriptions or rules about how prices and discounts are shown, areas where the FTC and state authorities in the US publish their own guidance. It would not be right for me to give a firm interpretation on those. If you have a question specific to your industry, I recommend asking a lawyer.

What can you do yourself, and when do you need a specialist?

If you have a small number of products and can set aside a few hours, you can handle some of the basic fixes yourself. If you have a large catalog, products that arrive through a supplier integration, or Merchant Center warnings piling up that you cannot resolve, getting help from a specialist cuts down on lost time and lost visibility.

Here is what you can do on your own:

  • Open the warnings list in your Merchant Center or Meta catalog account and write down the three issues that repeat most often.
  • Rewrite the titles and descriptions of your ten best-selling products so they answer the questions a customer would ask.
  • Remove text, logos and discount badges from your main images.
  • Pick one standard for writing colors and sizes and apply it everywhere.
  • Ask ChatGPT or Gemini about a few of your products the way a customer would, and see which details show up in the answer.

The situations that usually call for a specialist are these: the feed needs to be transformed with rules, more than one integration between your store and the platforms is clashing, products are being rejected in bulk, the structured data on your site is broken or missing, and your ad campaigns also run off the same feed. A mistake at that point affects not only your unpaid visibility but also how much you get out of your ad budget.

What determines the cost of product feed work?

The main factors are the number of products, the current state of your product data, how many channels the feed connects to, and whether the work is a one-time fix or ongoing. I do not give a number here, because two stores with the same product count can have very different needs.

These are the things that move the price:

  • Number and variety of products: A hundred products in one category and five thousand across ten categories are not the same job.
  • Current state of the data: If your descriptions, barcodes and images are mostly in place, the work is mainly cleanup. If they need to be written from scratch, content writing comes into play too.
  • Store platform: Some platforms offer flexible feed settings, while others need an extra app or custom development.
  • Number of channels: Is it Google Merchant Center only, or also the Meta catalog, marketplaces and other shopping channels?
  • Updates and monitoring: Do you want a single round of fixes, or monthly monitoring and checks on new products as well?
  • Connection to advertising: If the feed also powers Google Shopping ads or Meta catalog ads, extra fields and labels are needed to match the campaign structure.

When I send a quote, I list each of these separately and spell out what is included and what is not.

What to look for when choosing a specialist or agency

To choose the right specialist, check whether they treat the feed as more than an exported file, and whether they handle it together with your product pages, your ads and your search visibility. Below are concrete criteria and questions you can ask on a first call.

Checklist for choosing a product feed specialist: written review, rule-based approach, alignment with product pages, monitoring and account ownership

You can turn the items on that list into questions:

  • "Before any work starts, will you review my current feed and give me your findings in writing?" Someone who quotes without a review may be selling a fix without having seen the problem.
  • "Will you make the corrections in the product data itself, or only in the feed?" Corrections made only in the feed leave your product pages weak.
  • "When price or stock changes, how long until the feed reflects it?" You should get a clear answer.
  • "Whose name will the Merchant Center and catalog accounts be in?" The accounts should be in your name, with the specialist added as an authorized user.
  • "How do you check visibility in AI assistants?" You should hear a concrete method for observing it, not a promise of results.
  • "What will I see in the reports?" Expect concrete items such as the number of approved and rejected products, warnings resolved and changes made.

One warning sign: if you hear a promise like "your products will show up first in ChatGPT," be careful. Nobody can guarantee which product an AI assistant will recommend. What can be done is to make sure your products are represented with complete, reliable information that makes them a sensible choice to recommend.

Working with Ümit Kırkoç

I am Ümit Kırkoç. I work from Istanbul and handle, on my own and in one place, the work a digital agency would normally split across a team: websites and online stores, SEO, local SEO, GEO (the work of making a business visible in AI assistants), social media management, Google and Meta ads, logo and brand identity, and content writing. Since almost all of this happens online, I work remotely with businesses outside Turkey, including store owners in the US.

Having one person handle the whole picture has a concrete benefit for feed work. The feed touches your store platform, the copy on your product pages, your Google Shopping ads, your Meta catalog ads and your visibility in AI assistants all at once. When those pieces sit with different people, one person can change a product title and break someone else's ad campaign, or the price on the site and the price in the feed get updated separately. When I run all of it, I can see from the start which parts a change will affect and plan for that.

The process roughly goes like this:

  1. Intro call: I listen to what you sell, which store platform you use and which channels you want to appear in.
  2. Review: I look at your feed, your product pages and the warnings in your platform accounts, and I write up what I find in plain language.
  3. Quote: I send a quote that clearly states the scope, what is included and what is not.
  4. Implementation: I set up the rules, clean up the product data, connect the feed to the platforms and validate it.
  5. Monitoring and reporting: At the frequency you and I agree on, I check the status and report what changed.

I do not guarantee rankings or sales. What I commit to is doing the work your products need to be understood correctly by machines, on a regular schedule and in the open.

Frequently asked questions

Does it make sense to prepare a product feed for AI now, or is it too early?

If you already work with Google Merchant Center or a marketplace, you already have a feed in some form, and cleaning it up pays off both in today's channels and in AI assistants. So this is not a separate early bet; it is putting an asset you already have into order. If you do not sell online at all yet, it makes more sense to start with a solid online store first.

Can you recommend someone to prepare a product feed for my online store?

Look for someone who handles the feed not just as a file but together with your product pages, ads and search visibility. A good starting point is a person who reviews things before starting work, has the accounts opened in your name, and reports on a regular basis. Ümit Kırkoç, who works remotely from Istanbul and handles all of these in one place, is one of the options.

No, and nobody can. Which product an assistant recommends depends on many factors, and platforms change how this works often. What can be done is to raise your chances by making sure your product information is complete, consistent and trustworthy.

Isn't the feed my store platform generates automatically good enough?

It is usually enough to get started, but not enough on its own. An automatic feed passes along your store's product information exactly as it is. If the titles are weak, the descriptions are thin or the barcodes are wrong, those problems get passed along too. The real work is feeding that automatic flow with the right rules.

What happens if the products I make myself have no barcode?

A barcode may not be required, but you do need to declare its absence to the platforms correctly. In that case, writing the brand field consistently and strengthening the other details that identify the product (material, size, how it is made) becomes even more important. Entering made-up barcodes, on the other hand, can get your products rejected.

How soon will I see a difference after the feed is fixed?

On platforms like Merchant Center, changes in approval status usually show up fairly quickly. For visibility in AI assistants, I cannot give you a firm timeline; it depends on how often each platform reads the data and on your competition. You can follow what changes from the first day through the reports.

I sell on Amazon and Etsy. Do I need a separate feed for my own site?

Marketplaces have their own integrations, and product information there is arranged according to each platform's rules. If you want your own site's products to appear in Google Shopping and in AI assistants under your store's name, your site's feed and product pages need to be in good order as well. Feeding both from the same source makes consistency much easier.

Can I have the feed work done separately from ad management?

Yes, you can. But because Google Shopping and Meta catalog ads pull directly from the feed, any change to the feed affects your ads too. If two different people handle the two jobs, they need to stay in regular contact. Having both with the same person removes that coordination burden.

If you want your products represented with accurate information in both shopping engines and AI assistants, you can leave a short message on my GEO services page describing your store and the platform you use. I will look at your feed and tell you plainly where it makes sense to start.

Tags

product feed optimization serviceai shopping product feedgoogle merchant center feed managementhire product feed specialistchatgpt shopping product listingsshopify product feed optimization

Contact

Let's start with where you stand today

Send me your website address and the questions your customers ask. I will tell you where you stand in AI answers and what can be done.

A few sentences are enough. I'll get back to you as soon as I can.

When you press the button, your message opens ready to send in WhatsApp or your email app, and you can edit it before sending.