Product Recommendation Emails: How to Make the Batch Send Feel Hand-Picked, Not Generated

Every day, retail brands send millions of product recommendation emails to drive repeat purchases. These emails rely on sophisticated logic, what you bought, browsed, or left in cart, to surface relevant items. But when subscribers open them, something feels off. The hero image is on-brand, carefully art-directed. Then beneath it? A grid of product thumbnails rendered in Arial or Helvetica, with raw PDP images and cramped spacing that looks like the ESP built it. Because it was.

The irony is staggering: the most personalized part of the email, the product recommendation emails, is also the most off-brand. We’ve spent years optimizing what products to show, while leaving how they look structurally broken. The result? A 5-7% conversion gap between generic HTML cards and composed, on-brand images, identical logic, different render layer. That’s not a test result. That’s a missed revenue stream.

What Shoppers See When They Open Your Daily Send

Imagine opening an email from your favorite apparel brand. The header shows a model in a new seasonal collection, shot on location, styled with intention. The typography matches the website. The tone is consistent. Then, just below: a block of product thumbnails with no margin, rendered in Roboto or Arial, showing cropped PDP images with inconsistent lighting and backgrounds. The transition feels jarring, like stepping from a boutique into a warehouse.

This isn’t a design oversight. It’s a technical constraint baked into how most ESPs handle dynamic content. Product recommendation emails are typically built using HTML cards, each product tile generated in real time with raw data from the product feed. That means:

• No control over image composition (PDP shots weren’t made for email)
• No custom fonts (Gmail only supports Roboto and Google Sans)
• No spacing or layout adjustments at the block level
• Image size capped at 102KB in some clients, forcing compression

Even if your brand uses custom fonts elsewhere in the email, they won’t apply to these cards in Gmail, used in 27.03% of opens (Litmus, July 2026). Apple Mail supports them, but Gmail doesn’t offer fallback to your brand’s typeface. It falls back to Arial or Helvetica. That means for a third of your audience, the personalization feels impersonal.

Why HTML Product Cards Can’t Look Like Your Brand

Let’s be clear: the problem isn’t the data. It’s the delivery. Your ESP pulls the right products, Browsed Together, Best Sellers, Cart Favorites. The logic is sound. But the rendering is stuck in 2012.

Here’s what happens in practice:

Typography fails. You’ve chosen a distinctive font for your brand. In Gmail, it disappears. As Email on Acid confirms, Gmail only supports two web fonts: Roboto and Google Sans. Anything else reverts to system fonts. So your elegant serif or modern sans-serif? Gone. Replaced with generic Arial. That’s not just a visual downgrade. It erodes brand recognition.

Imagery fails. PDP images are optimized for product pages, not email. They’re shot against white or gray backgrounds, often inconsistently lit or cropped. When grouped in a grid, they look like a data dump, not a curated collection. There’s no art direction, no color harmony, no lifestyle context.

Layout fails. HTML cards are rigid. You can’t adjust padding between items, align prices with descriptions, or create visual hierarchies. The result is cluttered, hard to scan, and visually disconnected from the rest of the email.

And because this block is dynamically generated, it’s often exempt from design reviews. The creative director signs off on the hero. The CRM team owns the logic. But no one owns the look.

The Creative Director Test: What an On-Brand Recommendation Feels Like

Here’s a simple benchmark: if your creative director can’t tell whether a product grid was dynamically generated or manually designed, it passes. That’s the standard we use with Zembula clients.

When recommendations are rendered as a composed image, a single, high-fidelity image built at open time with consistent typography, layout, and art direction, the experience shifts. The grid feels intentional. The spacing breathes. The fonts match the brand. The images are cropped and color-balanced to work together.

This isn’t about aesthetics for aesthetics’ sake. It’s about conversion. Product recommendation emails with on-brand rendering consistently outperform HTML cards by 5-7%. That’s not a one-off test. It’s a structural advantage.

Why? Because subscribers notice quality, even when they can’t name it. A polished grid signals care. It tells the shopper: We thought about this for you. In contrast, a raw HTML card says: The algorithm picked this.

And in a world where 71% of consumers expect personalization (McKinsey, Jan 2025), and 76% get frustrated when it feels generic, that subtle signal matters.

Five Ways to Make Product Recommendations Feel Hand-Picked

It’s possible to scale this level of quality without manual design work. Here’s how leading retailers are doing it:

1. Composed image grids. Replace HTML cards with a single image block that renders at open time. The grid is built using brand-approved templates, fonts, and spacing. Each product tile uses lifestyle-cropped images, consistent typography, and aligned pricing. The result? A block that looks like it was designed, not generated.

2. Last-order personalization. Instead of showing generic bestsellers, use the recipient’s last purchase to inform picks. If they bought women’s sneakers, show complementary styles in the same category and price range. This creates continuity in the experience.

3. Recipe rotation. Avoid repetition by rotating the logic behind recommendations, Browsed Together one day, Cart Favorites the next. This keeps the content fresh across sends, even for inactive users.

4. Eligibility and hero-protection rules. Automatically exclude out-of-stock items, enforce price floors, and prevent brand conflicts (e.g., don’t show Nike next to Adidas). These rules protect margins and brand integrity.

5. Editorial alignment with the hero narrative. If the hero is promoting ‘Summer Essentials’, make sure the recommended products align, lightweight tees, sandals, sunglasses. This creates a cohesive story, not a disjointed collection.

These tactics work together to make the batch send feel curated, not automated.

What Personalized Recommendation Content Actually Earns

Let’s talk numbers. In Q2 2026, product recommendation emails on the Zembula platform drove $2.10M in attributed revenue from 527.7M impressions, with a normalized RPM of $7.16, an increase of $0.47 quarter over quarter. The click-to-conversion (CTC) rate was 8.9%, roughly 3.6x the 2.5% baseline for standard daily batch emails.

That’s significant, especially when you consider scale: this content appears in nearly every broadcast email, making it the highest-revenue block in most sends. For comparison, triggered Abandoned Cart banners have higher intensity (19.7% CTC, $125.51 RPM) but much smaller reach. Recommendations trade per-impression intensity for universal, every-open presence.

These figures are from our 2025 email performance benchmark report, aggregating cross-vendor retail data with standard exclusions. The trend is clear: when recommendations are rendered well, they convert.

Where Recommendations Sit on the Personalization Ladder

Most brands think of personalization in phases:

1. Smart Banners: Dynamic headers based on location, weather, or loyalty status. Low lift, easy to implement.
2. Triggered Heroes: Behavior-based images (e.g., abandoned cart, browse recovery). Higher lift, limited to specific segments.
3. Product Recommendations: Full dynamic grids. Highest complexity, highest volume impact.

The shift to product recommendation emails as a performance channel requires treating them with the same rigor as paid ads, testing, attribution, optimization. But unlike ads, email leverages owned audiences and first-party identity, making it privacy-durable and cost-effective.

One test we recommend: run a two-step grid experiment. First, A/B test HTML cards vs. composed images (same logic, different render). Then, test different recommendation recipes (e.g., Bought Together vs. Trending). Most brands see the biggest lift in the first test, proving that how it looks matters more than what’s shown.

Shipping It Without Rebuilding Your Template

The best part? You don’t need to overhaul your ESP or email design system to start. By replacing the HTML product grid with a single Smart Block image that renders at open time, you can deliver on-brand recommendations using your existing template.

Zembula’s Image-based Product Grid does exactly this: it pulls product data, applies your brand rules, and composes a high-fidelity image just before the recipient opens. No new design work. No template rebuild. Just better performance.

And because it’s decided at open, not send, the picks stay fresh, even for users who open days later. That’s critical, since over 10% of email revenue comes from opens more than a week after send.

Key takeaways

• The conversion gap in product recommendation emails isn’t about logic, it’s about rendering. Identical picks convert 5-7% better when shown as composed images vs. HTML cards.
• Gmail’s two-font limit (Roboto, Google Sans) means HTML cards can’t use brand typography for 27% of opens.
• On-brand rendering signals quality and care, which shoppers respond to even if they can’t articulate why.
• Q2 2026 data shows product recommendation blocks earning $2.10M revenue at $7.16 RPM and 8.9% CTC, making them the highest-revenue block in most emails.
• You can upgrade rendering without changing templates: use a single image block (Smart Block) that composes at open time.

For more on how to maintain brand fidelity in dynamic content, see Why Your Product Recommendation Email Doesn’t Look Like Your Brand and Your Email Design System Decides Whether Your Product Recommendation Email Stays On-Brand.