Google Shopping Feed Optimization with AI: The 6-Step Product Feed Playbook
Fix the feed fields that decide who sees your products. See what 12,375 products revealed, where titles get cut, and how to review AI drafts before upload.
- 12,000+PMax campaigns audited
- 200+Live ecom clients
- €200M+Tracked sales
Google Shopping feed optimization means rewriting the product data you send Merchant Center. Google uses that data to match each item to the searches that sell. Start with missing identifiers, then titles. Put buyer words in the first 70 characters. Draft with AI, read every line, and ship through a supplemental feed so the storefront stays clean.
In the Product Feed Quality Index we counted 12,375 products across 251 stores on 28 July 2026. 81.3% had no GTIN or MPN. 68% of titles ran past 80 characters while Shopping ads show about 70. Identifiers and useful title text needed the most work in this sample.
Products sampled
12,375
Stores measured
251
No GTIN or MPN
81.3%
Titles past 80 chars
68%
How to Optimize a Google Shopping Feed
Shopping ads have no keyword list. Google reads the feed and decides which searches get your product. A bid change on a weak feed rarely saves the account because the match can be wrong before the bid runs.
Before copy work, fix any missing ID, price, availability, product link, image, brand, and GTIN or MPN. A GTIN is the number under a product barcode. An MPN is the maker's part number. Then check sale price and item_group_id. Check Google product category, shipping, and extra images when they apply.
After that, fix titles and descriptions. Use product type and custom labels when they help separate products that need different targets. This order stops you polishing copy on a product Google cannot identify.
Rank products by revenue. If the store has no sales, rank them by impressions. In the jewelry case, we chose the top 80 by impression share from a 480-product catalog. Treat 80 as that store's cutoff, not yours.
ROAS
Before
1.8
Week 0, blended
After
2.6
Week 9, same spend
On a jewelry account spending €120k a month, we rewrote 80 titles. We added custom labels for margin band, 30-day sales speed, and metal type. We also split the catalog. We reset target return on ad spend (tROAS) from 340% to 220%, then moved it to 260% as results improved. At week 9, CPA was down 38%. ROAS rose from 1.8 to 2.6, and conversions were up 63% while spend stayed flat. This case does not isolate feed work from bid changes.
Fill Identifiers Before You Rewrite Titles
An identifier is how Google knows what the product is. With a GTIN or an MPN, Google can match your offer to its catalog and put you in the auctions that product belongs in. Without one, it reads your title and guesses. Your product can enter fewer auctions and get matched from weaker signals.
The hole is large. In that 12,375-product sample, 81.3% carried neither a GTIN nor an MPN. This was not a few messy stores dragging the average down. 168 of the 251 stores (66.9%) had no identifier on any sampled product at all. Many of those feeds were measured at onboarding, built by apps on default settings.
The decision rule is simple.
- If the product has a real barcode, put that GTIN in.
- If it does not, send brand plus MPN.
- Never invent a GTIN to fill the column. Google's Shopping ads policies treat inaccurate product data as a violation. That is a fast path to a Merchant Center suspension.
Brand is not the gap. Brand was filled on 97.5% of sampled products. Condition sat at 96.2%. Apps already write those. Identifiers are the field a human has to add on purpose.
Do this before you spend a week on titles. A keyword-rich title on a product Google cannot identify is still a guess.
How Long a Google Shopping Title Should Be
Merchant Center accepts 150 characters in the title and rejects anything longer. Not one title in our sample broke that cap, because the cap is hard.
The number that actually matters is the display cut. Most Shopping placements show about 70 characters. In the same sample, 68% of titles ran past 80 characters. The average store's average title was 91 characters. Only 4.2% of titles sat under 50. Stores are not writing titles that are too short. They are writing endings that no shopper ever sees.
We want to have an accurate and descriptive keyword-rich title. That is the whole goal. Because in Shopping, unlike Search, you do not set keywords manually. The targeting mainly happens via your product data feed.
Use an include-or-cut rule.
Include a word if a converting search term uses it, or if it names the product type or brand. Cut a word if it is promo text, a repeat of a structured attribute nobody searches, or a guess the model made. If the product type is missing from the first 70 characters, rewrite. If it is already there and the converting modifiers are there too, stop.
For fashion, brand and material often change who buys. For supplements, that might be dosage and serving size. For furniture, it might be dimensions in centimeters. The test is the same in every niche. If removing the word would attract the wrong buyer, keep it. If it would not, leave it out.
Apps love to stuff color, size, marketing phrases, and the store name into one long string. That is how you get a 91-character title whose selling words sit at position 90. Those words may as well not exist.
| Storefront Title | Feed Title (Sub-feed) | |
|---|---|---|
| What it looks like | Kept as-is, clean for shoppers | Product type, brand, converting modifiers first |
| Where it lives | Your live product page | Pushed to Merchant Center as a sub-feed |
| Job | Reads well on the page | Carries the Shopping targeting |
The title that works in the feed is often not the title you want on the product page. Keep the page readable. Push the Shopping version through a sub-feed. The full title playbook, with vertical examples, lives in Google Shopping product title optimization.
For most stores starting out, the fastest win is smaller than a full rebuild. Add the buyer keyword and the material to what you already have. Put those words up front. You do not need to rewrite the whole catalog on day one.
How to Use AI on Titles and Descriptions
Use AI to draft the copy. Check each fact against the product page. Upload only approved copy through a feed tool.
Pull 90 days of search terms from Shopping and Performance Max first. Separate terms that led to sales from searches that brought the wrong buyer. Use the wrong matches as your rewrite brief. Each one can show which word in the current title is causing a bad match.
Then paste the current title into an AI tool with that brief. Tell it to rewrite for a Google Shopping feed. List the attributes you want included. Run it once. Read the output. Fix anything that is off or invented.
Do the same for descriptions. The limit is 5,000 characters. That gives you room to explain real product benefits with proof. Use a feature-and-benefit structure, with bullet points for the selling points. Instead of "this jacket is made of cotton," write "made from 100% organic cotton so it keeps you cool all day." Ask the model for that shape. Then edit the tone so it sounds like your brand.
We do this for best sellers first. You are not going to prompt 5,000 products one by one. Pick the products that drive the most revenue and check each draft.
AI does not know the real size of your item or the exact fabric blend. It guesses. A wrong guess in the feed can trigger a Merchant Center misrepresentation flag. Check every claim before upload so the feed stays accurate.
If you want size and color added in a structured way across hundreds of products, you need a feed tool. AI is good at copy on individual products. A tool like DataFeedWatch is better at a template that applies the same order across the catalog. We compare the stack in the Google Shopping feed optimization tools guide. Shopify stores that want the install path can start with the Simprosys feed setup.
Product Type and Custom Labels
Product type is a category path you write. Performance Max can use that path to split products into listing groups. Those groups help you report on products that need different targets.
Product type had a value on 80.8% of products in our sample. Check whether each path separates products that need different targets. A broad path such as Supplements can hide an underperforming category. A path such as Apparel > Outerwear > Rain Jacket gives you a useful split when rain jackets need their own target. Add another level only when it changes how you report or bid. Merge duplicate names before you map the path to a Google product category.
Custom labels do not change who sees the ad. Use them to group products that should not share a target. On the jewelry account, we used margin band and metal type. We also used 30-day sales speed, which means units sold during that window. Metal type separated solid gold from plated stock.
Without that split, Google's automated bidding treats a thin-margin sale and a fat-margin sale as the same win. It goes after whichever one is easier to get.
In our study of 12,304 Performance Max campaigns, two groups beat one by 17% ROAS. Five groups fell back to about 6% above a single group. The split only works if the feed can name the difference. Custom labels are how you name it. The setup steps live in the custom labels guide.
On the jewelry account, the title rewrites were what stopped solid gold from matching plated-query searches. Metal type was the label that separated solid gold and plated stock into their own listing groups, so the algorithm stopped mixing two different buyer audiences. Margin and velocity were useful. Metal type was the one that changed how the campaign was structured.
Ship a Sub-Feed so the Storefront Stays Clean
A sub-feed is a second data source in Merchant Center. You can upload it as a Google Sheet. It overrides or adds fields on top of the main feed. Storefront titles stay as they are. Shopping gets the optimized ones.
That is the shipping layer. AI does not do it. AI also will not handle item-group IDs or category rules across a 5,000-product catalog.
item_group_id links variants of one product. Give its size or color variants the same value. Set this structural field before you chase copy wins.
Images tell the same "apps fill the easy field" story. Only 41% of sampled products carried two or more extra images beyond the main photo. Google pulls extra images into Shopping formats and AI results. Most stores give it one photo.
Once the sub-feed is live, check the product data in the page code, known as schema. The price and availability must match the feed. Use the same GTIN in both places. Most Shopify themes already output basic Product schema. A mismatch looks like bad data to Google. It can trigger a Merchant Center flag, and it can keep AI tools from recommending the product.
How the Same Feed Serves Performance Max and AI Mode
Performance Max runs on this feed. For an ecom store, PMax is one feed campaign spread across six places at once: Search, Shopping, YouTube, Discover, Gmail, and Display. The product data you just rewrote is what it reads to decide which searches your items chase.
A title that names the buyer keyword, brand, and the converting modifier gives PMax something to match on. A title whose selling words sit past the display cut gives it a guess.
The same feed now does a second job. Google's AI Mode shows shopping results. It picks those products from the Shopping Graph, the product database Google builds largely from Merchant Center feed data. When someone asks AI Mode for "polarized sunglasses under 100," the products it can show are the ones whose feeds it can read and trust.
There is no separate AI Mode feed. A missing attribute can keep AI Mode from matching the product to a detailed request. If your feed does not say polarized, or does not name the material, the system cannot tell if the product fits. It picks a different one instead. Price and availability have to be right. Stale data will not get recommended.
AI Overviews and AI Mode use different sources. The text answers in AI Overviews come from Google's web index, not your feed. Product results in AI Mode and shopping carousels come from the Shopping Graph, which your feed fills. Feed work can affect AI Mode. Page content can affect the answer box.
Gemini uses Google's Shopping Graph. ChatGPT can read product pages and merchant feeds. Keep the Merchant Center feed and page data accurate because these systems do not use one shared source.
The playbook for getting picked by those surfaces, attribute by attribute, is how to optimize your product feed for AI shopping engines. This page covers rewriting and shipping. That page covers what AI shopping tools need to recommend you.
For budget work after the feed update, use how to scale Google Ads.
If you want this done on your catalog, see how we work.
Frequently Asked Questions
How do you optimize a Google Shopping feed?
Pull 90 days of search terms. Fill a real GTIN or brand plus MPN. Rewrite titles so product type and converting modifiers sit in the first 70 characters. Add product type and custom labels so you can split by margin. Draft descriptions with AI, read every line, and upload the new values as a supplemental feed so the live store stays clean.
How long should a Google Shopping title be?
Merchant Center accepts 150 characters and rejects anything longer. Most Shopping placements show about 70. In ZenoX's Product Feed Quality Index, 68% of titles ran past 80 characters, so the tail never appears in the ad. Put the words that sell in the first 70. Add more only if they show up as converting search terms.
Is it safe to use AI in my Google Shopping feed?
As a draft, yes. As a paste, no. AI does not know your real fabric blend or size. A wrong guess can trigger a misrepresentation or website-needs-improvement flag. Use AI to write the first version. You correct it. A feed tool ships the approved line to Merchant Center.
Can I use a different title in the feed and on my store?
Yes, and you usually should. A feed title like "Women's Slim Fit Bamboo Leggings - High Waist, Black - Brand Name" can look messy on a product page. Keep the storefront title readable. Push the Shopping title through a sub-feed in Merchant Center. The page stays clean. The feed carries the targeting.
Do I need a separate feed for Google AI Mode?
No. AI Mode picks products from the Shopping Graph, which is built largely from your Merchant Center feed. Gemini uses that same graph. ChatGPT reads the open web and merchant feeds. There is no second upload. Complete attributes, titles a machine can parse, and page schema that matches the feed.
Which feed fields should I fix first?
The empty ones that get you into more auctions. In 12,375 products across 251 stores, 81.3% had no GTIN or MPN, and material sat at 5%. Fill identifiers first, rewrite titles second, then one manual attribute across the catalog. Product type and custom labels come after that so you can split campaigns.
