Google’s AI surfaces are changing how shoppers find products. AI Overviews and AI Mode in Google Search are no longer future features. They are live, in use, and pull product data directly from your feed to generate answers for shoppers.
If your feed is incomplete or poorly structured, your products will not show up in those answers. Your competitors’ will.
I have spent years optimizing product feeds for merchants across different industries and verticals. One of my clients saw an 80% improvement in Google Ads performance after we improved their feed data. Not their bids. Not their creative. Just the data. Their cost per click dropped, too.
That is what clean, complete, well-structured feed data does.
In this article, I am going to walk you through the 8 attributes Google explicitly connects to AI-driven results. I will show you how to structure each one correctly, the mistakes to avoid, and what to focus on first if you are starting out.
The 8 Attributes Google Explicitly Links to AI Overviews
Google has been clear about which attributes feed into its AI-driven surfaces. Here they are:
- Product Highlight
- Product Detail
- Variant Option
- Item Group Title
- Related Products
- Question and Answer
- Document Link
- Popularity Rank
Let me go through each one.
1. Product Highlight
Think of Product Highlight as your product’s bullet points. The key selling points. The reasons a shopper should buy this product over anything else.
Google explicitly states that submitting this attribute helps customers discover your products on AI-driven surfaces, including AI Mode in Google Search.
What to include:
Focus on selling benefits, not technical specifications. Ask yourself: what problem does this product solve? Why would someone choose this over a competitor’s product? What is the single most compelling thing about it?
Good product highlights sound like this:
- Stays readable in direct sunlight with an anti-glare display
- 10-year battery life with no charging required
- Waterproof up to 50 meters, tested for open-water swimming
Bad product highlights sound like this:
- Material: Titanium
- Weight: 320g
- Dimensions: 15 x 8 x 4 cm
Those last three belong in Product Detail, which I will cover next.
The CSV comma problem:
This is where many merchants break their data without realizing it.
If your highlight contains a comma, and most sentences do, a CSV file will treat that comma as a column separator. So a highlight like “Features a titanium case, and a green nylon band” gets split into two values. The second value becomes “and a green nylon band.” That is broken data.
The cleanest solution is to use a TSV file rather than a CSV file. Tab-separated files do not have this problem.
If you use Google Sheets, what we do at FeedArmy is add a backslash before any comma that is part of the sentence, not a separator. So the value becomes: “Features a titanium case\, and a green nylon band.” Google Sheets handles this correctly.
2. Product Detail
Product Detail is where your technical specifications live. This is the specs table you would see on a well-built product page: materials, battery type, display type, dimensions, connector types, and so on.
This attribute has three sub-attributes, and you need to use all three together:
Section Name is the category or grouping for the specification. For a watch, this might be General, Battery, Display, Design, or Lighting.
Attribute Name is the specific property within that section. For example: Material, Brand, LED Count, or Modes.
Attribute Value is the actual value. For example: Titanium, 8 Total, or 5 Different Settings.
For a tactical watch, the Product Detail data might look like this:
| Section | Attribute Name | Attribute Value |
|---|---|---|
| General | Material | Titanium |
| General | Brand | MyWatchBrand |
| Lighting | LED Count | 8 Total |
| Lighting | Modes | 5 Different Settings |
| Battery | Type | Rechargeable Lithium Ion |
| Battery | Life | 10 Years |
| Display | Illumination | Blue LED |
| Design | Style | Tactical |
When someone asks Google AI Mode “what is a good tactical watch with a long battery life,” it is this structured data that allows Google to surface your product with a confident, specific answer.
The rule is simple: highlights are for benefits, product details are for specs. Do not mix them up.
3. Variant Option
Google already has standard variant attributes: color, size, material, pattern, age group, and gender. Those cover the most common use cases.
But what if your products have variants that do not fit into those categories?
Say you sell laptops, and the difference between your variants is the graphics card. One comes with a GeForce 4070, the other with a GeForce 5070. There is no standard “graphics card” attribute in Google’s feed specification. That is exactly what Variant Option is built for.
Variant Option has two sub-attributes:
Name is the type of variant. In this case: Graphic Card.
Value is the specific option for that variant. In this case: GeForce 4070 or GeForce 5070.
So for the first laptop variant, you submit:
Graphic Card : GeForce 4070
And for the second:
Graphic Card : GeForce 5070
One important requirement: Variant Option is designed to work with the item_group_id attribute and item_group_title attribute. If you are not already using item_group_id to group your variants, set that up first. Variant Option is built on top of that structure.
4. Item Group Title
If you are grouping product variants using item_group_id, you should also add Item Group Title as an attribute. Google explicitly recommends it.
Here is the concept. Each variant has its own product title. But Item Group Title gives the entire group a shared, overarching title that sits above all the individual variant titles.
For example, your individual variant title might be “Flip Phone 23rd Gen Ultra Max 128GB.” That is specific and variant-level. The item group title would be “Flip Phone 23rd Gen.” Shorter, cleaner, no variant-specific detail.
This matters for AI surfaces because it helps Google understand your product range as a structured group, not a collection of loosely related items.
A few rules to keep in mind:
- The item group title must be the same across all variants that share the same item_group_id
- The item group title must be different from the title (normal title attribute)
- Keep it shorter and more generic than the normal titles
- Do not include variant-specific details like size or color in the item group title
- The character limit is 150 characters, but Google recommends putting the most important words first since some displays cut off around 70 characters
5. Related Products
This is one of the most important attributes in Google’s feed specification, and it is directly flagged as being primarily for conversational AI experiences like AI Mode.
The idea is straightforward. You tell Google what other products in your inventory are connected to this product, and what the relationship is.
Google supports six relationship types:
Part of set: A chair that belongs to a dining table set.
Required part: A battery that a battery-operated lamp needs to function.
Often bought with: A phone case that is commonly purchased alongside a phone.
Substitute: A comparable product from your range that could replace this one.
Different brand: The same product sold under a different brand name.
Accessory: Something that works with the product but is not required, like a webcam for a desktop computer.
Related Product has three sub-attributes that you need to include together:
Relationship type: Which of the six types applies.
Identifier type: Either “gtin” if you are using a barcode, or “id” if you are using the product ID from your own data source.
Identifier: The actual GTIN or product ID of the related product.
For a camera that needs a battery and a lens, and comes with an optional carrying case, you would submit:
required_part : id : AZ7A
required_part : id : AZ7B
accessory : gtin : 811571013579
You can submit up to 30 related product relationships per product. Do not comma-separate multiple identifiers inside a single entry. Each related product needs its own separate attribute submission.
When someone is in AI Mode and asks “what else do I need to use this product?” or “are there any alternatives?”, this is the data that drives those answers. If your competitor has this data and you do not, their products will show up in that conversation instead of yours.
6. Question and Answer
Google explicitly flags this attribute as being primarily for conversational AI experiences.
You provide pre-written questions and answers about your product. Google uses them to respond to detailed shopper questions in AI Mode conversations.
The attribute has two sub-attributes:
Question: Up to 1,000 characters.
Answer: Up to 1,000 characters.
You can submit up to 30 question-and-answer pairs per product, with a combined limit of 10,000 characters across all pairs.
The best source for your Q&A content is the questions your customers are already asking. Check your support tickets. Check the Q&A sections on Amazon for similar products. Check your reviews for the things people mention wanting to know before buying.
For a tea kettle, your pairs might look like this:
Q: What temperature settings does this kettle have? A: It has five settings: 175F for green and white tea, 197F for oolong, 200F for French press coffee, and 212F for boiling.
Q: How long does the Keep Warm function hold the temperature? A: It maintains the selected temperature for up to 15 minutes.
Two important rules:
First, do not duplicate content that already exists in your other attributes. The Q&A should add new, useful information, not restate what is in your description or product details.
Second, if you are also submitting the Document Link attribute and the same information is available in those documents, do not also submit it in Q&A. Google will extract the information from the documents directly.
7. Document Link
Document Link is one of the most powerful attributes on this list, and probably the one most merchants have never heard of.
Google explicitly marks it as primarily for conversational AI experiences.
You provide links to PDFs of your product’s user manuals, assembly instructions, training guides, and package inserts. Google crawls those documents and uses them to answer detailed shopper questions in AI Mode.
Think about what that means. If someone asks AI Mode “how do I assemble this furniture?” and you have linked your assembly PDF in your feed, Google can pull that information and give a full, structured answer. That is a major advantage in a conversational shopping environment.
The format is a URL pointing to a publicly accessible PDF. You can submit up to five document links per product.
Requirements to keep in mind:
The URL must start with http or https
The file must be a PDF
Maximum file size is 50 MB
The document must be publicly accessible so Googlebot can crawl it
The URL should be stable and not reused for different documents over time
The document should be about your specific product, not your company in general
Best document types to submit: user guides, assembly instructions, product specifications, manuals, and package insert information.
8. Popularity Rank
Popularity Rank tells Google how popular this product is relative to the rest of your inventory. It is a number between 0 and 100.
Google explicitly connects this attribute to AI Mode in Google Search, using it to help shoppers make more informed buying decisions.
Your best-selling product gets a high number, close to 100. A slow mover or older product gets a lower number. One decimal point is allowed, so 95.5 is valid. Do not add a percentage sign.
The ranking must be accurate and relative to your actual sales data. Do not put 100 on everything. That defeats the purpose and may work against you.
For merchants just getting started with this attribute, a simple approach works well:
- Top 10 best sellers: 90 to 100
- Mid-range performers: 50 to 80
- Slow movers and older inventory: below 50
Update this value when there is a meaningful shift in a product’s popularity. Running it through a supplemental data source monthly or quarterly is a practical way to keep it current.
The Most Important Thing I Want You to Take Away
These eight attributes are explicitly tied to Google’s AI surfaces. Start with them.
But do not stop there.
Every attribute you improve in your feed contributes. Your titles, your descriptions, your GTINs, your categories, your images, all of it matters. AI systems need complete, accurate, well-structured data across the board to accurately represent your products.
What I do for every client is improve everything. Do not cherry-pick a few attributes and move on. We go through the entire feed, find the gaps, fix the structure, and add what is missing.
That 80% performance improvement I mentioned at the start? That came from treating the feed as a whole. Not from ticking a few boxes.
Start with these eight. Then keep going.