Amazon’s AI shopping assistant, Rufus, is not ranking products by keywords alone. Under the hood, it’s trying to understand what the shopper actually means and then decide which products are genuinely good for that context.
Amazon has been explicit about this. In its research paper “A Shopping Agent for Addressing Subjective Product Needs” (published by Amazon Science), the company outlines how large language models interpret shopping intent across a small number of structured “need types”, rather than raw search terms alone.
You can read the original paper [here](https://assets.amazon.science/34/fb/6df56530497084838cd28039e833/a-shopping-agent-for-addressing-subjective-product-needs.pdf)
This research matters because it maps almost perfectly to how Rufus now reasons over Amazon listings.
At Lmo7, we use this framework directly when optimising Product Detail Pages, especially bullet points, and we’ve embedded it into our Rufus Radar tooling so recommendations are built directly into bullet rewrite guidance.
Below is how to translate Amazon’s research into practical, Rufus-ready bullets.
**The 5 Rufus Need-Types → The 5 Bullet Angles**
Rufus is attempting to map intent → suitability, not just query → keyword.
Strong listings make that mapping explicit.
__1. Subjective property (how it feels / looks / seems)__
What shoppers ask:
“lightweight”, “sturdy”, “premium”, “gentle”, “quiet”, “powerful”
How to write the bullet:
State the subjective attribute, then prove it with something concrete.
Bullet pattern
[Subjective attribute]: [proof/spec] so it delivers [benefit].
Example
Lightweight, non-greasy feel: Fast-absorbing formula that won’t leave a heavy residue for comfortable all-day wear.
Why this works for Rufus: the model can safely recommend the product when asked about feel, comfort or preference-based qualities.
__2. Event (when they need it)__
What shoppers ask:
Christmas, holidays, festivals, school trips, weddings, pregnancy, “summer holiday”
How to write the bullet:
Anchor the product to a time-based or situational use-case.
Bullet pattern
Ideal for [event]: [why it fits] + any constraints (size, travel, clothing, photos).
Example
Holiday-ready protection: Great for beach days and city breaks, with easy reapplication on the go.
This allows Rufus to recommend confidently when the shopper prompt includes when rather than what.
__3. Activity (what they’re doing)__
What shoppers ask:
Running, hiking, gym, commuting, gaming, travel, swimming
How to write the bullet:
Make it explicit that the product performs well enough for the activity.
Bullet pattern
Made for [activity]: [performance claim] + [comfort/durability] + [supporting spec].
Example
For sport & outdoor days: Designed for long runs and hikes with durable coverage you can rely on.
This is critical for Rufus prompts like “Is this good for running?” or “Would this work for hiking?”
__4. Goal purpose (what outcome they want)__
What shoppers ask:
“protect skin”, “sleep better”, “stay organised”, “reduce pain”, “lose weight”
How to write the bullet:
State the goal, then connect it to a mechanism.
Bullet pattern
Helps you [goal]: [mechanism] so you get [outcome].
Example
Reliable sun protection: Helps protect skin from sun exposure with broad coverage designed for everyday use.
This supports Rufus when the shopper is outcome-driven rather than product-driven.
__5. Goal audience (who it’s for)__
What shoppers ask:
Kids, men, women, beginners, professionals, sensitive skin, gifts
How to write the bullet:
Declare suitability for a specific audience and explain why.
Bullet pattern
For [audience]: [reason it suits them] + [ease/fit/comfort].
Example
Great for sensitive-skin shoppers: Gentle-feeling wear and easy application for daily routines.
This allows Rufus to personalise recommendations instead of making risky assumptions.
**A Practical “Rufus-Ready” Bullet Checklist**
- When reviewing or rewriting bullets, we look for the following:
- One clear use-case per bullet (no multi-purpose dumping)
- Language that answers prompts directly (“Great for…”, “Ideal for…”, “Made for…”)
- Subjective claims paired with evidence (specs, materials, format, testing — where true)
Coverage across contexts:
at least 2 activities
1 event
1 goal
1 audience
Natural shopper language, not internal brand jargon. If a bullet wouldn’t clearly answer a Rufus question, it’s usually doing too little work.
**How Lmo7 Applies This in Practice**
We’ve embedded this exact framework into Rufus Radar. When you use the tool, bullet point recommendations are:
- mapped to Amazon’s own Subjective Product Need taxonomy
- structured to support Rufus prompt matching
- written to stay Amazon-compliant and conversion-first
In other words, you’re not guessing how Rufus might interpret your listing, you’re writing directly to the model’s decision layer.
If you want us to apply this to a live ASIN (or review existing bullets through Rufus Radar), get in touch.
Amazon has already told us how the system thinks.
The opportunity is in writing listings that finally reflect that.