How to read this article based on your profile
- E-commerce SME owner: focus on §1 (the 3-in-1 feed asset) + §5 (12-18 month urgency) + §8 (audit)
- E-commerce marketing lead: focus on §2 (Performance Max depends on the feed) + §6 (the 7 rules) + §7 (the tools)
- E-commerce SEO lead: focus on §3 (structured data and AI) + §4 (the ChatGPT ads)
- Acquisition agency: focus on §5 (the urgency) + §6 (the 7 rules to audit at the start of a mission)
1 · The product feed, a 3-in-1 asset in 2026
Your product feed (the file that lists your products with their prices, photos and stock) is no longer just a technical file managed by the developer. It has become your shop's ID card across three distinct worlds:
World | What the feed powers | Consequence without a clean feed |
|---|---|---|
Paid ads (Google Ads + Meta Advantage+) | Performance Max + Shopping campaigns + Advantage+ catalog ads | 74-97% of the Performance Max budget wasted on broken optimization |
Organic search (Google, Bing, Google's AI summaries) | Rich result cards (price + stock + reviews) + Knowledge Graph + AI summaries | No organic visibility on product searches |
AI (ChatGPT, Claude, Perplexity) | Sponsored recommendations from the feed + the AI crawlers read your structured data | Total invisibility in searches done by chatting with an AI |
Three worlds, one asset. This is the best result / effort ratio in the entire 2026 e-commerce marketing toolkit. Yet, across HeySquad missions in 2023-2026, most e-commerce accounts arrive with a feed hand-cobbled by their developer, set up for Performance Max only, and completely ignored on the structured-data and AI side.
2 · On the paid-ads side: 74-97% of Performance Max depends on the feed
Performance Max became the e-commerce acquisition standard in 2024-2026. Nearly all of the optimization done by its AI rests on the quality of your product feed, which you send through Google Merchant Center (the Google space where you upload your catalog).
According to the 2026 Google Ads benchmarks:
Data point | Value | What it means |
|---|---|---|
Share of feed-based ads in Performance Max spend | 74-97% | The feed is literally lever #1 |
Shopping return for accounts > 50 k$/month ad spend | 5.2× | More volume = more data = an AI that performs |
Shopping return for accounts < 2 k$/month ad spend | 2.8× | Too little signal = the AI improvises |
Best title format | brand + product type + key attributes + size/color in 150 characters | Beats shorter titles consistently |
Barcode (GTIN) mandatory | Without it, your ad rank drops mechanically | Common basic mistake |
Source: Xictron Performance Max 2026, Search Engine Land Performance Max Tips 2026, AdNabu Performance Max Best Practices
On the Meta side, with the AndroMeda algorithm, the same principle applies. Advantage+ Shopping campaigns read your catalog feed and tune delivery product by product. Without a clean feed, the algorithm learns on noise and performance collapses.
Bottom line on paid ads: you can spend tens of thousands of euros per month on Performance Max and Advantage+ with a cobbled feed. You will spend. You will bring back little. The feed is your least-discussed profitability lever.
3 · On the search side: structured data is your passport to the AIs
For classic SEO, Google has recommended schema.org Product structured data since 2017 (the standardized product card that Google, Bing and the AIs know how to read, written in a technical format called JSON-LD). The recommendation hasn't moved in 2026. What has changed: this structured data no longer feeds only Google's rich result cards. It also feeds the AI summaries (the AI-generated answer blocks at the top of Google's results) and the AI crawlers.
Official Bing confirmation, from their product manager: structured data helps the AIs understand content and generate their answers (Leadgen Economy).
The AI crawlers that read this data today:
- ChatGPT (OpenAI's GPTBot crawler)
- Claude (Anthropic's ClaudeBot crawler)
- Perplexity (the PerplexityBot crawler)
- Google's AI summaries (the Google-Extended crawler)
Source: Product Schema for AI Commerce, ZipTie
The schema.org Product fields that decide your AI visibility:
- `name`, `description`, `brand` (the basics: name, description, brand)
- `offers` with `price`, `priceCurrency`, `availability`, `priceValidUntil` (the offer: price, currency, availability, validity)
- `sku`, `gtin13`, `mpn` (the product codes)
- `aggregateRating` and `review` (the average rating and reviews)
- `material`, `color`, `size`, `weight` (material, color, size, weight)
- `image` (several high-resolution images)
Without this structured data, your product exists for your shop but does not exist for the AIs. The most detailed sites are the ones that show up most in AI answers.
4 · On the AI side: ChatGPT has been running ads since February 2026
The shift from search to the AIs is no longer theory. OpenAI officially launched sponsored ads in ChatGPT on February 9, 2026 (Digiday).
The figures:
Data point | Value | Source |
|---|---|---|
OpenAI ads launch date | February 9, 2026 | Digiday |
Annualized revenue in 6 weeks | 100 million $ | Search Engine Land |
Format | Card at the bottom of the answer, explicit "Sponsored" label | OpenAI |
Automatic hook-up to the product feed | OpenAI builds the ads itself from the product catalog | |
Data displayed | Brand logo + Sponsored + name + price + stock + delivery | OpenAI Official |
How it works: OpenAI plugged product feeds into ChatGPT. The platform builds the ads itself from product names, images and feed fields, instead of setting up campaigns by hand. See OpenAI's official shopping research.
This is the same movie as Performance Max in 2018-2020:
- 2018: Google launches Performance Max. The latecomers catch on in 2020. They lost 24 months of visibility.
- 2026: OpenAI launches ChatGPT ads from the product feed. The latecomers will catch on in 2028. They will have lost 18 to 24 months.
In parallel, the AIs are gradually adding payment right inside the conversation. E-commerce search is not moving checkout to Amazon. It is moving checkout into ChatGPT, Claude, Perplexity. Your product feed is your entry ticket to this new storefront.
5 · A 12 to 18 month window: why now
If you run an e-commerce SME or lead its marketing in 2026, your calendar probably looks like this:
- Q3 2026 (now): you read this article. You realize you're behind.
- Q4 2026: you audit your current feed. You spot the missing fields.
- Q1 2027: you fix it, you enrich your structured data, you update your stock daily.
- Q2 2027: you start to see the effect on Performance Max (and Advantage+ on Meta) and the first appearances in AI summaries.
- Q3-Q4 2027: you are visible in ChatGPT ads via your feed. Competition still sparse on the AI side.
Beyond Q1 2028, the window closes. The AI ads market will have saturated the way Performance Max did in 2020. Acquisition costs will rise, the spots will be taken.
12 to 18 months to make the switch. Consistent with article 4 on creative agencies that have 18 months to evolve. Same calendar, same logic of an opportunity window before saturation.
6 · The 7 rules of a product feed that scales
Best practices drawn from 30+ HeySquad e-commerce missions with Belgian SMEs in 2023-2026. Confirmed by the 2026 Google Ads benchmarks and the schema.org/Product structured data.
Rule 1: a structured title (brand + type + attributes)
The ideal formula: `[Brand] [Product type] [Key attribute] [Size/Color]` in 70 to 150 characters.
Example: `Levi's Jean 501 straight men's indigo blue W32 L34` (readable by ads, search and the AIs)
Avoid: `Awesome jeans unbeatable quality price 12345` (keyword-stuffed, no useful signal)
Rule 2: mandatory product codes (GTIN, MPN, SKU)
Three identification codes: GTIN (your international standard barcode), MPN (the manufacturer's reference), SKU (your internal stock reference). Without a GTIN, Google mechanically lowers your ad rank. Without an MPN or SKU, you can't match the same product from one channel to the next. These 3 codes are non-negotiable in 2026.
Rule 3: a square image, plain background, 800×800 minimum
Square format is mandatory for Performance Max and the ChatGPT ad cards. Plain background (snow white #FFFFFF is ideal). Resolution 800×800 minimum, 2000×2000 recommended for high-definition screens and AI visual analysis.
Rule 4: a label per margin level
The `custom_label_0` field is the most powerful sorting lever in Performance Max. Our recommendation: sort by gross margin percentage per product (e.g. `margin_high` > 40% / `margin_med` 25-40% / `margin_low` < 25%). That lets you push high-margin products without spending on low-margin ones.
Rule 5: update stock daily, not weekly
The AIs and Google Merchant Center penalize stock marked out of stock but left unsynced. Update your stock at least every 24h via an automated link or a dedicated tool (see §7). Without it, your ads point to unavailable products, your ad rank drops and the customer experience breaks.
Rule 6: a deep category mapping
The Google Merchant Center category mapping goes down to 5 levels or more (e.g. `Apparel & Accessories > Clothing > Pants > Jeans > Skinny Jeans`). Most e-commerce SMEs stop at 2 levels and cap out. Going deep improves Performance Max targeting and the AIs' understanding.
Rule 7: structured data ready for the AIs
On each product page of the site, schema.org structured data with:
- `name`, `description`, `brand`, `sku`, `gtin13`, `mpn` (name, description, brand, product codes)
- complete `offers` with `price`, `priceCurrency`, `availability`, `priceValidUntil` (the detailed offer)
- `aggregateRating` + `review` if you have them (average rating and reviews)
- `material`, `color`, `size`, `weight` depending on the category (material, color, size, weight)
- several high-resolution `image`
This data is read by the crawlers of ChatGPT, Claude, Perplexity and Google's AI summaries (see §3). Consistent with server-side tracking, which keeps measurement clean on the server side.
7 · HeySquad's tools for the product feed
On e-commerce missions, we combine feed management tools, AI enrichment and the modules built into e-commerce platforms.
Multi-channel feed management
Channable (channable.com) is our primary recommendation for Belgian e-commerce SMEs. Powerful rules engine, direct links with Google Shopping, the Meta catalog, TikTok and marketplaces. Accessible pricing vs competitors.
Alternatives:
- DataFeedWatch (Cart.com): equivalent in features, more geared toward large accounts
- Productsup: enterprise scale, overkill for an SME
- Feedonomics (BigCommerce): direct link if you're on BigCommerce
The module built into Shopify
Shopify's built-in Google & YouTube channel and Facebook & Instagram channel update your feed automatically to Google Merchant Center and the Meta Commerce Manager. Enough for shops with fewer than 100 products without complex rules. Beyond that, Channable brings more finesse.
AI enrichment for feed visuals
Consistent with article 4 on creative and live data:
- Higgsfield Fashion Factory: generates a product photoshoot from a single image (-90% cost vs a traditional shoot). Ideal for creating image variants for your feed
- Pomelli (Google Labs): generates on-brand ad visuals, for free (in beta in Europe)
- Adobe Firefly or Photoroom: batch retouching of your product photos
Measurement and monitoring
- Google Merchant Center Diagnostics: tracks approved products, warnings and disapprovals
- Google Search Console: tracks impressions and clicks on your products in organic search
- Structured data validator (schema.org/validator): checks your structured data
8 · The 4 traps we see in audits at e-commerce SMEs
Trap 1: the broken sync with Google
The link between Shopify (or another e-commerce platform) and Google Merchant Center often breaks after a site migration or a theme update. Symptom: 20 to 40% of products get disapproved without a clear alert. Mandatory audit at the start of a mission: check Merchant Center Diagnostics at least 1× per month.
Trap 2: the keyword-stuffed title
A product title like `Super Promo Women's Jeans Blue Cheap Free Shipping ! ★★★★★`: Google penalizes this promotional overload. Your Performance Max ad rank drops. Your structured data loses clarity. A classic mistake seen in missions.
Trap 3: the default vendor image
Many shops reuse images straight from the supplier (B2B distributors in particular). Result: navy-blue pixelated 300×300 images on a dark gray background. Unusable for Performance Max, ChatGPT ads and AI summaries. This is exactly why AI enrichment matters (see Higgsfield §7).
Trap 4: the sloppy category mapping
A Merchant Center mapping at 2 levels (e.g. `Apparel > Clothing` instead of `Apparel > Clothing > Pants > Jeans > Skinny Jeans`). Diluted Performance Max targeting and weak understanding by the AIs. Simple fix, quick gain.
