Amazon Rufus Has Four Modes. You're Optimizing for One.

When a shopper asks Rufus "Is this durable?" it doesn't look at your bullet points. It reads your reviews.

That's not a small detail. Amazon brands are spending real money on listing copy and A+ content — and for a large slice of Amazon Rufus queries, that work is invisible to the AI. Not because it's bad. Because Rufus isn't reading it for that type of question.

Amazon Rufus doesn't run on one signal. What it reads depends on what someone asks. Analysis from Hashmeta, cross-referenced with Amalytix's 1,300-product behavioral study, documents four distinct retrieval modes — each pulling from a different layer of your Amazon content stack.

The Four Query Types

That last row is where most brands are caught off guard. Your listing could claim "easy cleanup" three times in the bullets. If your reviews tell a different story, Rufus surfaces the reviews. We documented this pattern in detail in last week's post on review signals — Rufus treats listing copy as claims and reviews as evidence. For sentiment queries, it goes straight to the evidence.

The taxonomy here isn't Amazon's official technical documentation. It's behavioral inference from systematic testing. But it holds up consistently across categories, and it matches what practitioners are seeing in the field.

Why Your Optimization Playbook Is Probably Wrong

Most brand teams treat Rufus optimization as one task: improve the listing. Better title, stronger bullets, more complete A+ content. That's the right fix for feature/use-case queries. It's the wrong fix for review-sentiment queries. And it's partially wrong for specific product questions, where backend attributes — item specifics, compatibility fields, package quantity — carry more weight than bullet points.

The mismatch gets expensive once Sponsored Prompts billing starts March 25. Every active Sponsored Products and Sponsored Brands campaign is auto-enrolled. You'll be paying for Rufus impressions without knowing which query types are generating them.

If your category skews toward "Is this safe for my toddler?"-type questions and your reviews contain consistent safety concerns, fixing your A+ content before March 25 doesn't solve the problem you're actually paying for.

What to Fix Per Mode

Feature/use-case queries are the ones most brands are already optimizing for — and they're right to. Match language to the actual use-case queries shoppers run. A stroller listing that doesn't mention "travel" in the relevant copy won't surface for "best stroller for travel" because Rufus is doing semantic matching against the query. Keyword density matters less than use-case completeness.

Specific product questions live or die on backend attributes. If "Is this compatible with iOS?" pulls from item specifics fields and those fields are blank, Rufus either guesses or surfaces competitors whose data is complete. For cameras and photography gear, compatibility fields are frequently the deciding factor — shoppers ask specific technical questions before buying. The same applies to dimensions, materials, certifications, and package quantity.

Review-sentiment queries don't care about your listing at all. The fix is review strategy: understand what language clusters are showing up in your positive reviews and reinforce it through post-purchase outreach, product inserts, and follow-up sequencing. For supplements, the dominant queries are sentiment-based — "effective," "side effects," "worth it" — and health claim language in listing copy gives Rufus nothing citable anyway. What works there is certification data and real customer language from verified purchases.

For health monitors, both query types appear in volume. Accuracy and ease-of-use are sentiment questions (reviews dominate). Compatibility and measurement range are product-specific questions (backend attributes dominate). Two different optimization tracks, same product.

The Part Nobody Can Measure

Here's the honest problem: Amazon doesn't report any of this.

The Ads Console doesn't tell you which query types your Sponsored Prompts impressions came from. It doesn't break out Rufus visibility by query type. It doesn't show you your "share of answer" — the percentage of relevant conversational queries where your product appears in the AI response. There's no native tool that lets you confirm whether your optimization actually moved your Rufus visibility.

Third-party AI visibility tools exist for general LLMs (Peec, Profound, Akii). None have Amazon Rufus integration with query-level data. The gap is real. Brands are making optimization decisions on inference and anecdote, then paying for impressions on March 25 with no way to verify the baseline.

That's what AgentBuy tracks — what Rufus says about your brand across query types, over time, so you can see what changed and whether your optimization moved anything.

---

The diagnostic you can run today: open Rufus and ask about your product in each mode. "What's the best [your product type] for [your primary use case]?" "Is [your product] [your primary quality claim]?" "Does [your product] work with [common compatible system]?"

Compare the answers to your listing copy. Where they diverge is the priority list.

Amazon Rufus doesn't have one algorithm. It has four access points. Most brands are only covering one.