2025 Holiday Discovery: AI Agent Feeds Redefine Retail

The 2025 holiday season was a problem for retailers. The digital marketplace is so fragmented now that you have to scream to be heard, and traditional ad buys weren’t cutting it. Audiences are too smart, and platform algorithms are too weird. We had to find a way to get to consumers earlier, so our whole focus shifted to holiday discovery. We decided to use AI agent feeds to put products right into personalized shopping assistant interfaces. The goal was to catch people’s attention while they were just starting to think about gifts, long before they’d decided on a specific brand or product. We wanted to see if AI-driven recommendations, popping up in conversational agents, could actually change how people shopped for the holidays.

Campaign Teardown: “Gifted Insights” by Aura Retail

The “Gifted Insights” campaign was Aura Retail’s big bet for the 2025 holiday season. They’re a mid-sized e-commerce brand in electronics and home goods with an average order value (AOV) of $180, and they wanted to grab a bigger piece of the early holiday shopping pie. Their main goal was to drive more sales by getting their products to show up in AI assistants, targeting people using Google Assistant and Amazon Alexa for gift ideas. The plan was to plant their product ideas in shoppers’ minds early on, not to chase direct sales through voice commands right out of the gate.

Strategy and Implementation: Beyond Traditional Search

Aura Retail’s strategy went way past just buying keywords. They saw that as AI assistants got smarter, people were going to use them for really broad, open-ended questions like “What are good gifts for a teenager?” or “Show me popular smart home gadgets for the holidays.” The entire campaign was built to make sure Aura’s products were the top answers for those kinds of questions, which meant they had to seriously invest in structuring their product data so an AI could understand it. We dug in with their product teams, enriching their existing feeds with really granular details, including who the gift was for, how it might be used, and holiday-specific keywords. So a smart speaker wasn’t just a “smart speaker” anymore. It became a “smart speaker for music lovers,” a “gift for college student,” and a “home automation gift.”

A huge part of this was linking their product catalog to the shopping graph APIs for both Google and Amazon. For Google Assistant, we had to optimize their Google Merchant Center feed with these new, richer attributes and also use structured data markup on the product pages themselves so everything was crystal clear, product type, availability, holiday relevance, all of it. For Alexa, the focus was building a custom “skill” that served as a curated discovery channel. It wasn’t a full store, but it let the assistant recommend Aura’s products when users asked for gift ideas in their categories. This wasn’t just a file upload. It took constant work, tweaking the product categories and the language people actually use when they’re thinking about gifts.

The campaign ran from October 15th to December 24th, 2025. The budget for just the AI feed optimization, content, and development was $120,000. That might seem small, but it was a dedicated 15% of their total marketing budget for the holidays. Our forecast was a 15% lift in conversions that weren’t last-click, because we knew this channel would start a much longer customer journey.

Creative Approach: Conversational and Contextual

Writing creative for an AI agent feed is a completely different world from typical ad copy. You can forget about flashy visuals and catchy headlines. Your whole world becomes conversational product descriptions and trying to guess what the user is really asking. For instance, a normal product description like “XYZ Smartwatch with advanced fitness tracking” just doesn’t work. The AI-optimized version we wrote was something more like, “Consider the XYZ Smartwatch, an ideal gift for fitness enthusiasts, offering precise heart rate monitoring and GPS tracking for outdoor activities. It’s a thoughtful present for someone looking to upgrade their health tech this holiday season.”

We built out a whole library of natural language prompts and matched them with product recommendations, which involved a lot of internal testing where we pretended to be holiday shoppers. We also had to write specific calls-to-action that made sense on a voice interface, like “Would you like me to add the XYZ Smartwatch to your holiday gift list?” or “Tell me if you’d like to see more tech gifts under $200.” The idea was to nudge the user toward taking some kind of action, even if it was just adding an item to a list, instead of expecting them to buy right away.

Visuals were still part of the equation, just not in the way you’d think. When an AI agent recommended a product, it would often push a rich snippet or a link, so it was critical that the destination page had great, holiday-themed photos and a clear pitch. We definitely saw that products with lifestyle shots showing them being used during the holidays (like a smart home gadget making a party better) got more clicks from the AI recommendations.

Targeting: Intent-Driven and Predictive

Targeting with AI agent feeds is all about intent-driven targeting, where the AI platform itself does most of the filtering, which is nothing like how you’d target a display ad. Our job was to give the AI agents the right signals to connect our products with the right questions. This boiled down to a few things:

• Extensive keyword research: We had to figure out all the long-tail, conversational questions people would ask an AI about gifts.
• Product attribute optimization: We tagged products with tons of detailed characteristics like “eco-friendly,” “luxury,” “budget-friendly,” “for gamers,” and “for cooks” to match them up with all the different gift-giving situations.
• Behavioral data integration: For people who were already customers, the AI platforms could use their past purchases and browsing history to give super-personalized recommendations which we helped along by making sure our customer data platforms (CDPs) were cleanly connected with the right data-sharing APIs.

The predictive part came from digging into historical holiday search data and what people were talking about online to get ahead of gift trends. For example, when we saw a spike in searches for “sustainable tech gifts” in early November, we immediately went in and tagged Aura Retail’s relevant products with those attributes to catch the wave. You have to be that quick to stay in the game during the holiday rush.

What Worked: Early Discovery and High-Intent Leads

The biggest win of the whole campaign was its ability to pull in high-intent leads way earlier in the shopping season than usual. By the middle of November, Aura Retail was already seeing a 22% jump in product page views coming from AI agent referrals compared to what they got from standard search and social the year before. This was a clear sign we were catching people before they’d even thought about what brand to buy.

Key Metrics:

• Impressions: 18.5 million (AI agent recommendations shown to users)
• Click-Through Rate (CTR): 3.8% (users clicking on recommended products from AI interfaces)
• Conversions (AI-attributed): 5,100 (purchases where an AI agent recommendation was the first touchpoint, even if not the last click)
• Cost Per Lead (CPL): $8.25 (defined as a user adding a product to a wish list or initiating a cart from an AI recommendation)
• Cost Per Conversion: $23.53 (for AI-attributed sales)
• Return on Ad Spend (ROAS): 7.6x (calculated based on AI-attributed revenue of $918,000 against the $120,000 budget for this channel)

The average conversion rate for people who came from an AI agent recommendation was 2.8%, a little bit better than the 2.5% we saw from our normal paid search traffic during that time. That told us the leads were higher quality, probably because the AI’s recommendations felt more personal.

One tactic that really killed it was suggesting “gift bundles” in the AI feeds. When a user asked for something like “gifts for a new homeowner,” the AI, using our structured data, could pitch a whole bundle of smart home gadgets and send them to a dedicated landing page. Those bundles had a 15% higher average order value than single-item sales from the same AI channel.

What Didn’t Work: Data Latency and Complex Queries

It wasn’t all perfect. We hit some big roadblocks, and the main killer was data latency. Keeping the inventory updated in real-time across all the AI platforms was a serious headache. A product would get recommended by an AI, the customer would get excited, and then they’d click through only to find it was out of stock. This stuff is poison. It led to a 7% abandonment rate for products that went out of stock within 30 minutes of the AI referral. When your data isn’t fresh, you’re not just losing a sale, you’re actively creating a frustrating experience for a potential customer.

The system also choked on really specific or weirdly phrased questions. The AI was great at general stuff, but if someone asked for something like, “Show me a durable, waterproof Bluetooth speaker that floats, in a bright color, for under $75, for my brother who loves kayaking,” it would often serve up garbage or just fail completely. That showed us the limits of both the AI’s language skills for these multi-layered requests and the fact that we hadn’t tagged our products with that level of insane detail. We also saw a lower CTR on recommendations where the AI couldn’t give a short, convincing reason for its pick.

Optimization Steps Taken: Iteration and Refinement

We didn’t just sit there. Based on the early numbers, we started making changes on the fly:

1. Daily Inventory Syncs: We forced the issue and pushed for more frequent, ideally hourly, inventory syncs between Aura’s e-commerce platform and the AI feed APIs. This alone dropped the out-of-stock abandonment rate from 7% down to 3% in just two weeks.
2. Refined Product Categorization: For products that kept tripping up the AI on complex queries, we went back in and added more specific, long-tail attribute tags. For that “kayaking speaker” problem, we added tags like “water-sports,” “floating design,” “rugged build,” and “IPX7 waterproof rating” to give the AI a better chance.
3. A/B Testing Conversational Prompts: We were constantly testing different ways for the AI to phrase its recommendations. For example, we tested “This XYZ Drone is perfect for aerial photography” against “Capture stunning aerial views with the XYZ Drone, a top pick for photography enthusiasts this holiday.” The second one, being more descriptive and focused on the benefit, got a 12% higher click-through rate.
4. Negative Keyword Equivalent Optimization: This was a bit of a hack, but we figured out which product attributes were leading to bad recommendations and told the feed to de-prioritize them for certain queries. For example, if “gaming headset” kept getting recommended for “gifts for dad” and no one was clicking, we’d add a negative signal to that specific combination.
5. Focus on “Assisted Conversion” Metrics: We had to change how we reported on this. We shifted the focus to assisted conversions and first-touch attribution for the AI channel. It gave us a much truer picture of its value at the start of the customer journey, getting us away from the last-click model that always undervalues this kind of discovery work.

By the time we were done, the “Gifted Insights” project showed that AI agent feeds are a real and powerful channel for holiday discovery, even if it’s still got some rough edges. Being able to get in front of shoppers at the very start of their process with a relevant, personal suggestion gave Aura Retail a real leg up in a noisy market.

The future of shopping is definitely going to involve more of this kind of intelligent, proactive help. The brands that get serious now about optimizing for these AI channels will have a huge advantage. It’s not just about showing up. It’s about being relevant, accurate, and knowing how to talk to people.