Amazon Rufus and the sponsored prompt evolution

Rufus changes the unit of competition in retail media. The question is no longer only whether a brand can win an impression. It is whether the assistant has enough evidence to put that brand into a useful answer.

Launched in early 2024, Rufus reached 250 million monthly users by late 2025. Shoppers using Rufus demonstrate 60 percent higher purchase intent compared to traditional search. The assistant synthesises recommendations from product specifications, technical attributes, customer review sentiment, and comparative analysis.

In November 2024, Amazon introduced sponsored prompts into Rufus conversations. Product-level prompts surface when users discuss specific categories. Brand-level prompts focus on category discovery. The prompt sits inside the conversation, where the shopper is already asking for help choosing.

This is not a new ad format layered on top of search. It is a discovery surface where the commercial suggestion appears as part of the answer.

From assistant to ad platform

Rufus was not built as an ad platform. It was built as a shopping assistant trained on Amazon's product catalogue, customer reviews, community Q&A, and information from across the web. Amazon says shoppers use it for product research, comparisons, and recommendations. The important detail is the input. Rufus does not need another banner. It needs usable evidence.

The shift happened because the assistant reached scale. 250 million monthly users. 60 percent higher purchase intent. Amazon reports that shoppers using Rufus demonstrate stronger intent signals than traditional search users. The assistant became a discovery surface too valuable to leave unmonetised.

The sponsored prompt does not look like an ad. It looks like a helpful follow-up. "Would you like to explore Samsung's latest models?" feels like the assistant being helpful. The commercial intent is embedded in the conversational flow.

The mechanics of sponsored prompts

Amazon introduced two types of sponsored prompts in November 2024.

Product-level prompts

These surface when users discuss specific product categories. A query about televisions triggers: "Would you like to explore Samsung's latest models?" The prompts draw on product specifications, technical attributes, and customer review sentiment. The assistant is not just matching keywords. It is matching the conversation context to product attributes.

Brand-level prompts

These focus on category discovery rather than specific items. A user asking "What should I look for in a running shoe?" might receive a brand-level suggestion guiding them toward a specific manufacturer's range. The prompt draws on broader brand authority signals, category expertise, and comparative positioning.

Both prompt types share a critical dependency: the quality of the underlying product data. If specifications are incomplete, if review sentiment is negative, if attributes are missing, the sponsored prompt will not surface or will surface with weak evidence. The bid does not control the prompt. The evidence controls the prompt.

What this changes for retail media

The unit of competition has moved from the keyword to the conversation context.

In traditional retail media, you bid on "wireless earbuds." In Rufus, the shopper asks: "What wireless earbuds work well for Zoom calls?" The assistant synthesises a recommendation based on microphone quality, battery life, comfort for long calls, and review sentiment around call clarity. A sponsored prompt might surface: "Would you like to explore Jabra's latest models?" because Jabra's product data includes detailed microphone specifications and positive review sentiment around call quality.

The brands that win are not the ones with the highest bids. They are the ones with the most complete, accurate, and conversion-relevant product evidence.

This extends beyond Amazon. Dunnhumby's 2025 retail tech trends identify the birth of "shoppable AI search" where brands and retailers battle for "share of generated sentence." The focus is on summarising content for AI ingestion and the recommendation mechanisms needed for content-rich product discovery.

Preparing your catalogue for conversational discovery

The sponsored prompt does not appear because you bid higher. It appears because your product evidence matches the conversation context.

Start with a question audit. Map the questions shoppers ask in your category. Not keywords. Questions. "What wireless earbuds work for Zoom calls?" "Which running shoes suit wide feet?" "What television works in bright rooms?" Each question defines the evidence the assistant needs.

Then audit your product data against those questions. Do your specifications include microphone frequency response? Do your attributes include fit type for wide feet? Do your review summaries mention call clarity? The assistant cannot recommend what it cannot verify.

Review velocity matters. A product with 50 recent positive reviews about call quality will outrank a product with 500 old reviews about sound quality. The algorithm weights recency and sentiment alongside volume.

Availability and fulfilment feed the algorithm. A product that cannot ship Prime loses position even if everything else is in place. The assistant factors in delivery promises when synthesising recommendations.

Separate paid position from earned position in your reporting. A sponsored prompt that surfaces because of strong product evidence is different from one that surfaces only because of budget. Track both.

Amazon's advertising revenue reached $56.2 billion in 2024, pacing a $69 billion run-rate. The company's AI Creative Studio, layered with Nova foundation models in late 2024, converts a single hero image into dozens of video variations in minutes. Prime Video now has an average monthly ad-supported reach of more than 315 million viewers globally. The flywheel is clear: more advertisers create more data. More data makes the system smarter. A smarter system gets better results, which encourages more spend.

But the flywheel only spins if the product evidence is there. The brands that invest in complete, accurate, conversion-relevant product data will be the ones the assistant recommends. The brands that treat Rufus as another bid surface will be the ones the assistant skips.