Why Rufus Quotes Your Competitor and Not You: A Buy Box and PDP Diagnostic for Amazon Brands

By Stephen Honight, Founder of Lmo7

When a shopper asks Rufus “what’s the best electrolyte drink for cycling” and it answers with a competitor’s name, the reflex inside most brand teams is to look at price or to look at ads. Both are usually wrong.

Rufus, Amazon’s AI shopping assistant, is now connected to Alexa for Shopping. It does not pick brands the way a human shopper picks brands. It reads the catalogue. It pulls structured attributes. It extracts from bullets, A+ Content, Q&A and reviews. It looks at Buy Box posture and account signals. Then it picks the brand that is easiest to recommend confidently.

If you are not the brand it picks, the diagnosis is almost always one of three things. The listing is hard for Rufus to read. The Buy Box is unstable. Or the catalogue underneath is messier than you think. The work to fix this is unglamorous, methodical and high-leverage. It is also a long way from running another ads campaign.

This is a diagnostic piece for founders, marketplace managers and ecommerce directors trading on Amazon. It walks through how Rufus actually reads a listing, how to diagnose a Buy Box that keeps slipping, what an SPN-mapped PDP rewrite looks like in practice and how to prioritise the fix list.

How Rufus reads a PDP

Worth being precise about the mechanism, because most “Rufus optimisation” advice on the internet is theoretical.

Rufus does not parse your listing the way a human shopper does. It builds an internal answer to the question it has been asked, then picks the products and the source content that best support the answer. The pieces it weights most heavily are structured attributes, product title, bullet points, A+ Content, Q&A and reviews. The signals it considers around those pieces are Buy Box status, account health, in-stock posture, review density and category mapping.

That changes what the work looks like.

A title written for SEO keyword stuffing reads poorly to Rufus. It is hard for an AI engine to extract a clean product fact from a 200-character string of pipe-separated terms. A title built around the actual product, its category and the most important attribute reads cleanly.

Bullets written as marketing copy underperform. Bullets written as structured facts, each one mapped to a Subjective Property Need (SPN) facet that Amazon shoppers care about in the category, do the work.

A+ Content built as a brand campaign with image-heavy banners and minimal extractable text underperforms. A+ Content built around comparison points, clear benefit statements and structured modules reads well.

Q&A is the section most brands ignore. Rufus reads it heavily because it is shaped like the questions Rufus is being asked. Seeding accurate, well-written Q&A on a listing is one of the cheapest fixes available and one of the most overlooked.

Reviews matter, but not in the way most brands think. Rufus pulls themes and language from review content. A listing with a high review count but generic five-star content gives Rufus less to work with than a listing with fewer reviews and more specific buyer language. Encouraging detailed reviews, even at the cost of star-rating averages, is often the right trade.

A five-point Buy Box diagnostic

Buy Box slippage rarely has one cause. The honest answer when a client asks why theirs is unstable is usually three or four overlapping things. Worth working through the list.

The first check is suppression. Variants suppressed for missing attributes, image quality flags or compliance issues quietly remove themselves from the Buy Box and shoppers never see them. We have audited accounts where 15% of the live catalogue was technically suppressed and the brand had no idea. Seller Central does not always make this obvious. Pulling a clean suppression report and triaging it is week one work.

The second is parent-child integrity. Parent-child relationships in the catalogue tell Amazon what variants of a product belong together. When the relationships are broken, the wrong variant gets the Buy Box, reviews fragment across child ASINs that should be consolidated, and the listing reads as messy to both shoppers and to Rufus. Fixing parent-child structure is some of the highest leverage work on an Amazon account. Most brands underestimate how much of their visibility problem is downstream of broken catalogue structure.

The third is stock posture. Going out of stock, even briefly, costs Buy Box position and recovery is not instant. Forecasting and inbound shipment hygiene sit underneath this. If the operations cadence is loose, the Buy Box will be unstable regardless of what the content team does.

The fourth is content quality. A title that is too long, bullets that are too thin, missing structured attributes, no A+ Content, weak Q&A or thin reviews all reduce the listing’s credibility score in Amazon’s internal model. The Buy Box favours listings the algorithm trusts. Content quality is a Buy Box signal.

The fifth is account health. Late shipments, customer service flags, return rate spikes and policy compliance issues all degrade Buy Box eligibility. If trading metrics are slipping, the Buy Box gets harder to hold even when everything else is right.

In the worst cases all five are happening at once. The fix is sequenced, not parallel. Suppression first. Parent-child second. Stock and account health in parallel. Content quality fourth. Once those are stable, the work that moves the visibility needle on Rufus actually has a chance of compounding.

SPN-mapped bullets, before and after

This is the work most clients see first when we run a PDP rework. Worth a worked example.

Imagine a hydration powder sold to cyclists. The original bullet reads something like this.

“Premium quality electrolyte hydration drink with natural ingredients, refreshing taste and made with care in the UK. Trusted by athletes everywhere for performance and recovery.”

The problems are visible to a model. There is no specific claim. No category placement. No comparison point. No measurable attribute. No buyer-specific benefit. It is marketing copy without product information.

A rewrite mapped to the SPN facets a cyclist actually cares about reads more like this.

“Endurance electrolyte mix designed for rides over 90 minutes, with 1,000mg sodium per serving to replace the salt lost in sweat without the sugar crash of standard sports drinks.”

The difference is mechanical. The category placement is clear (endurance electrolyte). The use case is specific (rides over 90 minutes). The functional attribute is measurable (1,000mg sodium). The comparison is named (without the sugar crash of standard sports drinks). A model reading this can extract a clean fact and place this product confidently in an answer to “what is the best electrolyte drink for long rides.”

Across a five-bullet PDP, this is repeatable work. Map each bullet to a different SPN facet. Cover category placement, performance attributes, use case, comparison and credentials. Avoid stacking adjectives. Avoid claims you cannot back up. Aim for what a model can quote without needing to interpret.

The version-over-version impact of this work is measurable. We ran a similar rework across Veloforte’s Amazon catalogue. The dashboard now shows 198 bullet matches against the Lmo7 SPN framework, 42 titles live in the new format, 35 A+ Content pieces updated and 31 ASINs improved across three scrapes. Each scrape shows what has moved since the last one, which is the part that justifies the work commercially.

Three buckets: content fix, ops fix, authority fix

The single most useful frame we give Amazon clients is that their listing problems break into three buckets, and confusing one for another is the most common reason fixes do not stick.

Content fixes are the easiest to scope and the easiest to ship. Title rewrites, SPN-mapped bullets, A+ Content updates, Q&A seeding, structured attribute population, image and alt text upgrades. These move within weeks. They are the right entry point for most Amazon engagements.

Ops fixes are slower and they are usually owned by a different person inside the client business. Suppression triage, parent-child cleanup, stock posture, inbound shipment hygiene, account health metrics. These need an operations partner inside the client, not just a content brief.

Authority fixes are the slowest. Brand-level review density, off-Amazon citations that influence the way Amazon and Rufus understand the brand, retail media presence that compounds organic visibility, and external authority signals that travel back into the AI shopping flow through cross-engine signals. These pay off over months, not weeks.

A credible Amazon engagement does all three. The order is content first, ops in parallel, authority underneath. Trying to fix authority before content and ops is one of the most common failure modes we see, because the underlying listings will not hold the gains.

What we did with Veloforte

This is a real Lmo7 engagement worth grounding the principles in.

Veloforte is a founder-led sports nutrition brand. The starting problem was discoverability and conversion on Amazon, with a particular focus on whether their listings were doing the work they needed to do across both the shopper journey and the emerging Rufus AI surface.

The work ran across the three buckets. Content: an SPN-mapped rework of titles, bullets and A+ Content across the catalogue, with Q&A seeded for each major ASIN. Ops: a suppression and parent-child cleanup. Tracking: a dashboard that captured version-over-version change across scrapes so the gains were visible and so the team could prioritise the next cycle of work.

The internal dashboard now tracks 198 bullet matches, 42 titles live, 35 A+ Content pieces and 31 improved ASINs across three scrapes. That last number, the count of ASINs that have actually moved, is the one that matters commercially. Work that does not show up in version-over-version movement is work that has not landed.

This is a Challenger-tier engagement in the Lmo7 model. Modular delivery across Retail Ops, Content Optimisation and Ads. Monthly review calls. A working dashboard. The shape is repeatable across other founder-led brands and we are doing variants of it now with Pelotan and Symprove.

A buyer’s checklist for fixing your Amazon visibility

Five questions to ask before commissioning any Amazon agency work.

Ask them to pull a suppression report and a parent-child integrity check before they recommend any content work. If they want to write bullets without looking at the catalogue underneath, they are skipping the most important diagnostic step.

Ask how they map bullets to SPN facets. If the answer is vague, the rewrite will be generic. If they can show you a framework with named facets for your category, the work has a chance of moving the needle.

Ask whether they seed Q&A. Many Amazon agencies still treat this as a customer service surface rather than a Rufus extraction layer. The ones that don’t seed Q&A are leaving a cheap and obvious win on the table.

Ask what their version-over-version reporting looks like. One-off audits are useful. Tracking that shows what has moved across scrapes is where the work justifies itself.

Ask how they think about authority. If they tell you Amazon is a closed system and authority does not apply, they have not been paying attention to how Rufus pulls cross-engine signals. The brands winning in Rufus are also the brands that show up in ChatGPT, Gemini and Perplexity.

If a question gets a thin answer, you are about to pay for thin work.

So what should you do next

If your Buy Box is slipping, do the suppression and parent-child diagnostics this week. Both are free, both can be done inside the account, and both will give you a clearer view of what is actually broken. Most brands find more than they expect.

If your PDP content is older than a year and was written before Rufus launched, it almost certainly needs a rework. Start with the top five revenue ASINs. SPN-map the bullets. Rebuild the title around the product, the category and the most important attribute. Seed Q&A. Run a version-over-version comparison after 30 days.

If you have done the content work and still are not appearing in Rufus answers for your category prompts, the issue is usually authority. Off-Amazon citations, review density and category-level signals are slow to move but they are the lever that lifts the ceiling.

If you want help with any of this, the Amazon Stack at Lmo7 is built around exactly this work. Retail Ops at £750 a month covers the catalogue health and ops work. Content Optimisation for Search Visibility at £750 a month covers the SPN-mapped PDP work, including weekly ASIN keyword rank tracking. The Full Amazon Stack at £2,500 a month wraps in Ads. Most brands start with the Full Stack and feel the difference in the first eight weeks.

Rufus is reading your listings whether you optimise for it or not. The brands that put in the work now will set the recommendation default that the next year of shoppers will land on. The brands that don’t will keep watching competitors get quoted instead.

Stephen Honight is the founder of Lmo7, the AI-native agency helping challenger consumer brands win in AI-powered discovery and agentic commerce. Lmo7 works across Amazon, D2C and AI search surfaces with brands including Trip Drinks, Veloforte, Brown-Forman, Haleon, Pelotan and Symprove.