I predicted videos would appear in Amazon’s AI shopping chat back in January 202
By Andrew Bell · October 16, 2026 · Curated by George's Blog
I predicted videos would appear in Amazon’s AI shopping chat back in January 2025, when it was still called Rufus. Over the last few weeks, I’ve consistently seen them in Alexa for Shopping.
There’s some history here. We were already developing strategies around images in Rufus chat in 2024, and by November, images were appearing consistently. By January 2025, we were looking toward video, a possibility Amazon’s patents had already described. Recent patents reinforce what I’ve long called the
𝐅𝐢𝐯𝐞 𝐕𝐬 𝐨𝐟 𝐈𝐧𝐩𝐮𝐭 𝐚𝐧𝐝 𝐎𝐮𝐭𝐩𝐮𝐭
- 𝐕𝐞𝐫𝐛𝐚𝐥: Conversational text
- 𝐕𝐢𝐬𝐮𝐚𝐥: Images and product cards
- 𝐕𝐢𝐝𝐞𝐨: Moving product demonstrations
- 𝐕𝐨𝐢𝐜𝐞: Spoken conversations
- 𝐕𝐢𝐫𝐭𝐮𝐚𝐥: Augmented reality and spatial shopping
I’m talking exclusively about chat in Amazon’s app and on desktop, not Echo devices or tablets. Today, input includes verbal, visual, and voice. Output includes verbal, visual, and now video.
𝐌𝐲 𝐧𝐞𝐱𝐭 𝐩𝐫𝐞𝐝𝐢𝐜𝐭𝐢𝐨𝐧? 𝐕𝐨𝐢𝐜𝐞 𝐨𝐮𝐭𝐩𝐮𝐭, 𝐭𝐡𝐞𝐧 𝐩𝐞𝐫𝐡𝐚𝐩𝐬 𝐕𝐢𝐫𝐭𝐮𝐚𝐥.
Imagine asking Alexa to furnish your living room and having it build the room, place furniture, and let you explore configurations. The AR and room-visualization technology already exists. It hasn’t reached this chat yet.
At the beginning of this year, I also predicted screen control and highly agentic shopping activities. Think of capabilities similar to Muse, at Amazon’s enormous shopping scale.
Of course, the complexity is enormous. Amazon already has one of the most personal kinds of information about you: 𝐰𝐡𝐞𝐫𝐞 𝐲𝐨𝐮 𝐬𝐩𝐞𝐧𝐝 𝐲𝐨𝐮𝐫 𝐦𝐨𝐧𝐞𝐲. Purchases reveal priorities, relationships, and needs. Giving an agent greater autonomy inside that environment raises profound questions of trust, privacy, permissions, and financial authorization. I suspect Amazon is working through these challenges as it expands what Alexa can do.
And naturally, I have another early optimization strategy: 𝐕𝐢𝐝𝐞𝐨 𝐅𝐫𝐚𝐦𝐞 𝐎𝐩𝐭𝐢𝐦𝐢𝐳𝐚𝐭𝐢𝐨𝐧 (𝐕𝐅𝐎). Yes, another acronym for the e-commerce EIEIO.
Many multimodal AI systems sample video into a montage of frames to interpret alongside other context. A frame showing garment fit, material texture, scale, or how something operates might become useful product evidence for an agent.
𝐂𝐨𝐮𝐥𝐝 𝐛𝐫𝐚𝐧𝐝𝐬 𝐨𝐩𝐭𝐢𝐦𝐢𝐳𝐞 𝐯𝐢𝐝𝐞𝐨𝐬 𝐟𝐨𝐫 𝐭𝐡𝐞 𝐟𝐫𝐚𝐦𝐞𝐬 𝐚𝐧 𝐚𝐠𝐞𝐧𝐭 𝐜𝐚𝐧 𝐮𝐧𝐝𝐞𝐫𝐬𝐭𝐚𝐧𝐝 𝐚𝐧𝐝 𝐮𝐬𝐞? That’s an intriguing extension of ASIN Answerability Engineering: making a product findable, understandable, provable, and selectable.
I wouldn’t put your bottom dollar on VFO yet. We don’t know how Alexa processes these videos for sure or whether information extracted from frames influences recommendations. It’s a hypothesis, not an established tactic.
Still, the direction is hard to ignore: 𝐕𝐞𝐫𝐛𝐚𝐥 → 𝐕𝐢𝐬𝐮𝐚𝐥 → 𝐕𝐢𝐝𝐞𝐨 → 𝐕𝐨𝐢𝐜𝐞 → 𝐕𝐢𝐫𝐭𝐮𝐚𝐥.
Perhaps soon, your next Lifelong Shopping Buddy.