Alexa for Shopping and COSMO: 60 Amazon Criteria

On May 13, 2026, Amazon retired the Rufus brand and integrated its artificial intelligence engine into a unified assistant: Alexa for Shopping. Unlike Rufus, which operated in a separate chat window that few shoppers opened, Alexa for Shopping is now integrated directly into Amazon’s main search bar. Every shopper encounters it, whether they actively use it or not.

Under the hood, COSMO is still powering the experience. COSMO is the knowledge graph that understands the intent behind shopper queries. When someone searches for “shoes for a wedding,” COSMO does not simply look for the words “shoes” and “wedding” in product listings. It identifies the underlying intent—elegance, comfort for a long day, and a formal dress code—and surfaces products that match that intent, even when those exact words do not appear in the listing.

The central question for every brand selling on Amazon in 2026 is simple: how can a product be selected and surfaced by this AI?

AI-generated answers can display only three to five products, compared with 20 to 40 products in traditional search results. Competition for those positions has therefore become much more intense.

This article is a comprehensive guide. It covers the 60 criteria identified as being used by Alexa for Shopping and COSMO, grouped by category, with the exact thresholds verified as of July 2026.

Table of Contents

• The 2026 context: Alexa for Shopping, COSMO and agentic AI

• Product titles: 6 criteria

• Bullet points: 10 criteria

• Product descriptions: 4 criteria

• Backend keywords and Search Terms

• Visual content: 6 criteria

• Social proof: ratings, reviews and distribution

• Amazon badges: Amazon’s Choice, Prime and Best Seller

• The Buy Box and the commercial offer

• Availability and pricing

• Shopper guarantees

• What else COSMO analyses

• Taking action: where to start

Section 01

The 2026 Context: Alexa for Shopping, COSMO and Agentic AI

Amazon has fundamentally changed the way shoppers discover products. Before examining the individual criteria, it is important to understand the three concepts that now govern visibility on Amazon.

Alexa for Shopping, formerly Rufus

Alexa for Shopping is Amazon’s AI assistant. It answers shopper questions directly within the search experience.

When a shopper searches for “which headset should I buy for working from home with noisy children,” they no longer receive only a list of search results. They receive a conversational answer containing three to five recommended products.

This change is strategic.

On a traditional search results page, a shopper may scroll down to the 40th product. In an AI-generated answer, they see only three to five products. Appearing among those three to five recommendations is becoming one of the most important visibility challenges for brands.

COSMO: the semantic engine

COSMO, which stands for Common Sense Knowledge, is the knowledge graph powering Alexa for Shopping.

It does not rely solely on keyword matching. It focuses on understanding intent.

COSMO analyses 1.87 million search-to-purchase pairs and 3.14 million purchase-to-purchase pairs to understand what shoppers genuinely want when they enter a query.

In practical terms, COSMO rewards product listings that describe:

• Usage contexts

• User profiles

• Problems solved

• Compatible products

• Real-life situations

It gives less value to listings filled with disconnected keywords that do not form a coherent narrative.

A9 and A10: Amazon’s traditional search engine

A9, sometimes referred to as A10 by the Amazon community even though Amazon has never officially used the name “A10,” remains Amazon’s traditional search engine.

It indexes keywords and analyses factors such as:

• Keyword presence

• Keyword position

• Click-through rate

• Conversion rate

• Sales performance

A9 remains active in 2026. Most shoppers who enter a simple query such as “smart scale” still use the traditional search results page.

COSMO is an AI layer that operates on top of Amazon’s traditional search system.

What this means in practice

A product listing must now appeal to two different systems:

• A9, to appear in traditional search results

• COSMO, to be recommended by Alexa for Shopping

These systems do not operate according to exactly the same logic, which makes product listing optimisation more complex.

Amazon’s Listing Quality framework

Since 2026, Amazon has used its Listing Quality Dashboard as an active ranking signal.

A product with a listing quality score below 65% may experience a decline in organic visibility.

This dashboard is available in Seller Central, but its calculation remains opaque and can silently penalise millions of listings.

Section 02

Product Titles

The product title is the first element read by every system:

• A9 uses it for indexing

• COSMO uses it to understand the product

• Shoppers use it to decide whether to click

Six main criteria govern product titles.

Title length

The current recommendation in 2026 is between 180 and 200 characters for an optimised title.

A title that is too short may miss indexing opportunities. A title that is too long becomes difficult to read, particularly on mobile.

Amazon has announced that from July 27, 2026, the limit may be reduced to 75 characters for most categories, excluding media categories. A new “Item Highlights” field with a maximum of 125 characters is expected to compensate for the shorter title.

This rule has been confirmed for the United States but is not yet fully active in Europe. We are monitoring its effective rollout.

Attribute structure

COSMO strongly values front-loading.

The first 80 characters should contain:

• The brand name

• The main keyword

• The most important differentiating feature

This is the section most visible on mobile. It is also the portion the AI analyses first to identify and classify the product.

Brand name at the beginning

The product title should begin with the brand name.

This is an official Amazon rule, and COSMO uses the brand as a semantic anchor for conversational recommendations such as:

“I am looking for a product from brand X.”

Permitted characters

Amazon formally prohibits the following characters in titles unless they are part of a registered brand name:

! $ ? _ { } ^ ¬ ¦

Other characters such as:

~ # < > *

are accepted only in specific situations, such as indicating a style number or measurement.

Characters such as the following are generally permitted:

+ = & ( ) / , . : -

No promotional wording

Promotional expressions are prohibited in product titles, including:

• Promotion

• Best price

• Best seller

• Number one

• Free

Amazon may remove a listing from search results if these expressions appear in the title.

These terms also provide no semantic value to COSMO. They consume valuable characters without improving product visibility.

Semantic richness for AI

This is one of the most important title criteria for COSMO.

An AI-friendly Amazon title contains six distinct elements:

1. The brand

2. The product name

3. The product’s main function

4. The target audience or use case

5. A differentiating feature

6. A format, colour, size, quantity or pack detail

A title containing all six elements can be considered for a wider range of conversational queries.

A title containing only two or three elements is more likely to appear only for very precise searches.

A common mistake to avoid

Keyword stuffing means adding multiple synonyms for the same product in the title, for example:

“Insulated Water Bottle, Thermal Flask, Hot and Cold Drink Container.”

This practice can reduce AI visibility.

COSMO can detect unnatural keyword repetition and may give less weight to these titles.

One main product keyword is usually sufficient. Additional synonyms should be placed in backend Search Terms.

Section 03

Bullet Points

Bullet points form the main body of a product listing.

This is where COSMO extracts a large proportion of its intent-related signals.

Ten main criteria govern bullet points.

Number of bullet points

Amazon recommends using at least five bullet points.

With fewer than three bullets, the listing loses structure and the AI has less content from which to extract useful signals.

Using more than five bullets is not necessarily penalised, but additional content may provide little extra visibility once the indexing limit has been reached.

Bullet length

Each bullet should ideally contain between 180 and 250 characters.

This range provides a good balance between:

• Mobile readability

• Semantic depth

• Indexing efficiency

• AI interpretation

Indexed content in bytes

Amazon indexes only the first 1,000 cumulative bytes of bullet point content.

Anything beyond this limit may remain visible to shoppers but become invisible to A9 and COSMO.

This point is particularly important for non-English languages.

A byte is not the same as a character. Accented letters such as é, à and ç may use two bytes, while emojis may use four bytes or more.

French content can therefore reach the byte limit faster than the character counter suggests.

Usage signals

This is one of the most important criteria for COSMO.

Each bullet should contain a signal answering at least one of these questions:

• Who is the product for?

• In what context is it used?

• In what situation is it useful?

A usage signal can be short:

• For creative professionals

• At the office

• While travelling

It can also be more detailed:

• Designed for adults who monitor their blood pressure at home

Both formats provide useful semantic information.

Complementarity signals

COSMO analyses purchase-to-purchase relationships.

A product that explicitly mentions compatibility with other products or services can enter Amazon’s recommendation graph.

Examples include:

• Works with the Health Mate app

• Compatible with iOS and Android

• Can be used with a sports armband

• Designed for use with USB-C devices

These signals can support recommendations such as:

• Customers also bought

• Works well with

• Frequently used together

A product with no complementarity signal is less likely to appear in cross-product recommendations.

Usage frequency and timing

COSMO values explicit time-based references, such as:

• Every morning before breakfast

• Before and after exercise

• During your monthly health check

• At the end of each working day

These references help the AI connect a product with a shopper routine.

Without a time anchor, the product may be less relevant for queries such as:

• What should I use in the morning?

• What product should I use after exercising?

• What should I take on a weekly trip?

Usage context

Usage context describes where and under what conditions the product is used.

Examples include:

• Living room

• Office

• Outdoors

• While travelling

• At the beach

• In the mountains

COSMO can use these contexts to answer searches such as:

“Which product should I use outdoors?”

or:

“What is the best product for travelling?”

Anchored differentiation signals

A valid differentiating claim should be:

• Measurable

• Certified

• Comparative

• Connected to a real benefit

A generic adjective such as “high quality” or “very durable” provides little value.

A standalone specification such as “500 g” provides partial value.

A specification connected to a benefit is stronger:

“Only 500 g, making it comfortable to carry throughout the day.”

A comparative statement can be even stronger:

“30% lighter than previous models.”

This is what we call benefit-linked differentiation.

Technical specifications

Bullet points should include at least three precise technical specifications, such as:

• Dimensions

• Weight

• Material

• Compatibility

• Capacity

• Battery life

• Connectivity

• Certification

Amazon uses these specifications for long-tail indexing and to answer detailed shopper questions through Alexa for Shopping.

Problem-solution signals

COSMO gives particular value to listings that explicitly name the problem being solved.

Examples include:

• Prevents data loss

• Helps relieve back pain

• Protects against power cuts

• Reduces water leakage

• Limits background noise

• Avoids overheating

These direct formulations allow the AI to match products with problem-based queries such as:

“My back hurts.”

or:

“I keep losing my files.”

Section 04

Product Descriptions

The product description provides more space than bullet points.

It allows brands to develop a complete usage story and answer more detailed shopper questions.

Four main criteria govern product descriptions.

Description length

The optimal description contains between 1,000 and 2,000 characters.

Below 1,000 characters, the listing may lack enough semantic depth for COSMO to build a meaningful usage narrative.

Beyond 2,000 characters, the additional value tends to decrease and the AI may give less weight to the extra content.

Narrative structure

COSMO values descriptions following a clear narrative structure:

1. Usage scenario

2. Shopper problem

3. Product solution

4. Concrete benefit

A description consisting only of technical features is less effective.

A description showing how the shopper uses the product in real life is more useful for the AI.

Answering seven shopper questions

An effective description answers the seven main questions a shopper may have:

1. What exactly is this product?

2. Who is it designed for?

3. In which situations should it be used?

4. What problem does it solve?

5. What differentiates it from alternatives?

6. How is it used in practice?

7. What is included in the box?

xIntent: cross-category intent

COSMO identifies what can be described as xIntent, or intent that crosses multiple product categories.

A shopper searching for “how to improve my sleep” may be interested in:

• A mattress

• A weighted blanket

• An essential oil diffuser

• A smart lamp

• A sleep tracker

Descriptions that explicitly mention the shopper’s broader intention can appear for these cross-category searches.

Section 05

Backend Keywords and Search Terms

Backend keywords are one of the most underused indexing tools on Amazon.

There is one main criterion, but it is critical.

Search Term volume and format

Backend Search Terms are invisible to shoppers.

They are entered through Seller Central or Vendor Central and allow A9 to index the product for queries that do not appear in the visible listing.

In other words, brands can index a product for words shoppers use without having to include all of those words in the title, bullet points or description.

The absolute rule to remember

Amazon strictly limits this field to 249 bytes.

Exceeding the limit by a single byte may cause the entire field to be ignored, rather than only the excess words.

Best practices include:

• Separate words with spaces

• Do not use commas

• Do not repeat terms already included in the title or bullet points

• Do not include competitor brand names

• Avoid unnecessary punctuation

• Check byte length, not only character length

Section 06

Visual Content

COSMO is now multimodal.

This means it can analyse images as well as text.

Six visual criteria are particularly important.

Main image on a white background

The main product image must use a pure white background.

The product should occupy at least 85% of the frame and should not include:

• Promotional text

• Decorative logos

• Borders

• Watermarks

• Additional accessories not included with the product

This is an official Amazon rule.

A non-compliant main image may cause the product to be removed from search results.

Additional packshots from multiple angles

COSMO analyses multiple views to recognise the product from different perspectives.

The ideal image set contains at least three additional views beyond the main image:

• Three-quarter view

• Side view

• Rear view

Without these images, the AI can identify the product from only one angle, limiting its ability to describe and compare it.

Lifestyle images showing active product use

A lifestyle image shows a real person actively using the product in a recognisable environment.

It is not:

• A packshot on a white background

• A decorative image

• An infographic

• A person simply standing next to the product

The interaction between the person and the product must be visible.

This matters because COSMO analyses visual content to associate products with real usage situations.

Diversity of usage contexts

A product may be used in several environments:

• At home

• At the office

• Outdoors

• While travelling

• At the gym

• At school

• In a professional setting

Showing the product in multiple environments expands the range of queries for which COSMO can recommend it.

A product displayed in only one visual context may be limited to searches mentioning that specific environment.

Infographics and visualised benefits

Infographics can transform technical specifications into understandable visual information.

Examples include:

• Annotated dimensions

• Feature comparisons

• Compatibility tables

• Before-and-after results

• Installation steps

• Capacity demonstrations

COSMO can use these images to visually verify claims made in the product listing.

Product video

The presence of a product video is a strong signal.

Video content demonstrates the product in action and can increase time spent on the product detail page.

A9 may indirectly value this through stronger engagement and conversion signals.

Amazon does not make video mandatory, but listings without video appear to be underrepresented in Alexa for Shopping recommendations.

Section 07

Social Proof

Ratings, reviews and review distribution are trust signals used by COSMO to determine whether a product deserves to be recommended.

Average rating

The average rating is one of the most important factors, but the effective thresholds are often misunderstood.

For A9 and traditional search results

For Alexa for Shopping recommendations

Number of reviews

Review volume is a product maturity signal.

• Fewer than 10 reviews: the product appears insufficiently tested

• Between 10 and 49 reviews: credibility remains limited

• From 50 reviews: COSMO can identify recurring strengths and weaknesses

• From 500 reviews: the product benefits from a strong and defensible trust base

Distribution of positive reviews

The average rating does not show the full distribution.

A product with a 4.3-star rating could have:

• 80% five-star reviews and 20% one-star reviews

• A more balanced distribution producing the same average

COSMO uses review distribution, not only the average score.

A product with more than 75% four- and five-star reviews gives the AI enough evidence to mention high shopper satisfaction in conversational answers.

Negative review rate

The proportion of one- and two-star reviews should remain below 5%.

Above this level, COSMO may actively mention recurring product issues when answering shopper questions.

A high negative review rate is not simply a passive ranking problem. It can become information that the AI uses against the product.

What this means in practice

A product rated 4.3 stars with a 3% return rate may perform better than a product rated 4.8 stars with a 15% return rate.

The rating alone is not enough.

It must remain consistent with post-purchase behaviour.

Section 08

Amazon Badges

Amazon badges are external authority signals automatically assigned by Amazon.

In 2026, Alexa for Shopping appears to give significantly more weight to the Amazon’s Choice badge, making it strategically important.

The Amazon’s Choice badge

Amazon’s Choice is assigned automatically by an algorithm.

There is no application form.

The badge is query-specific. A product may receive Amazon’s Choice for “Bluetooth sports headset” without receiving it for “Bluetooth headset.”

Five conditions generally need to be met at the same time:

1. A product rating of at least 4.0 stars, with 4.3 or higher being preferable

2. Prime eligibility

3. Reliable stock availability

4. A return rate below the category average

5. A conversion rate of approximately 8% to 12% for the target query

The weakest condition determines the result.

A product with a 4.6-star rating, Prime eligibility, reliable stock and a 4% return rate may still fail to receive the badge if its conversion rate is only 6%.

Alexa for Shopping’s impact on Amazon’s Choice

Since May 2026, Alexa for Shopping appears to treat Amazon’s Choice as a stronger trust signal than before.

The observed weighting is approximately twice as strong in conversational answers.

When Alexa selects three to five products, it tends to prioritise Amazon’s Choice products first and expands to products without the badge only when the Choice products do not match the query.

The Best Seller badge

The Best Seller badge is assigned to the number-one product in a category or subcategory over a recent period.

It requires:

• Sustained sales velocity

• Continuous stock availability

• Strong conversion

• Stable category performance

A stockout lasting 72 hours may be enough to lose the badge, while regaining it may take four to six weeks.

The Prime badge

The Prime badge is automatically assigned to eligible products fulfilled through Amazon logistics.

For merchants using Seller Fulfilled Prime, performance requirements include maintaining approximately 93.5% on-time delivery in 2026, with a cancellation rate below 2%.

Prime eligibility

Prime eligibility allows a product to appear in Prime-filtered search results, which represent a significant share of Amazon searches.

Eligibility should remain above 90% over a 30-day period.

Intermittent eligibility may reduce rankings and disqualify the product from Amazon’s Choice.

Overall Pick, New Release and Small Business

Overall Pick is similar to Amazon’s Choice but is assigned to broader queries.

New Release is reserved for recently launched products with high sales velocity.

Small Business is a free badge available to qualifying companies with fewer than 100 employees. It may generate a conversion lift of approximately 13.5% without requiring additional advertising investment.

Section 09

The Buy Box and the Commercial Offer

Winning the Buy Box is an essential condition for selling effectively on Amazon.

Three main criteria govern it.

Buy Box ownership

The Buy Box, also called the Featured Offer, is the default “Add to Basket” or “Add to Cart” offer.

A product without the Buy Box becomes almost invisible because shoppers rarely click “See all buying options.”

A Buy Box ownership rate below 80% can significantly reduce performance.

Major change in 2026

In early 2026, Amazon removed certain seller-performance-based eligibility requirements for the Featured Offer.

Competition between offers has therefore intensified, with the algorithm switching between sellers more quickly based on pricing and availability.

Buy Box stability

Winning the Buy Box once is not enough.

The objective is to maintain it consistently.

A Buy Box that frequently switches to third-party sellers indicates commercial instability.

COSMO may include this instability in its recommendation assessment.

Buy Box stability should be measured over at least 30 days.

New condition

A product whose Buy Box is controlled by a seller offering a used item may be negatively assessed by A9 and COSMO.

This often suggests that the official offer is unavailable and a third-party reseller has taken over the main purchasing position.

Section 10

Availability and Pricing

A product that is unavailable or poorly positioned on price will not be recommended, regardless of the quality of its content.

Six criteria govern availability and pricing.

Immediate availability

The product should be in stock at the time of assessment.

Even a short stockout can cause an immediate decline in visibility and may result in the loss of the Amazon’s Choice badge.

Availability over 30 days

Amazon monitors inventory continuity over a 30-day period.

• Above 95% availability: optimal

• Between 90% and 95%: acceptable but not ideal

• Below 90%: the algorithm may identify operational weakness

Products with unreliable availability are less likely to be prioritised.

Price stability

A stable pricing history is a trust signal.

Frequent price changes may cause Buy Box suppression when Amazon considers the offer no longer competitively priced.

A useful objective is to keep the price stable over at least 30 days and avoid short-term changes greater than 10%.

Price consistency within the product range

Large price differences between products in the same range can create inconsistency for shoppers and the search algorithm.

A coherent pricing structure strengthens shopper confidence and helps Amazon understand the product hierarchy.

Prime eligibility over time

Prime eligibility should remain above 90% over a rolling 30-day period.

Maintaining stable eligibility is necessary to remain visible in Prime-filtered search results.

Section 11

Shopper Guarantees

Three main criteria determine whether the product can be purchased, returned and used without unnecessary friction.

Complete technical specifications

Amazon rewards listings that complete all relevant structured product attributes in the back office.

These may include:

• Dimensions

• Material

• Compatibility

• Weight

• Capacity

• Certifications

• Battery information

• Connectivity

• Safety information

• Included accessories

These attributes directly feed COSMO and help it answer detailed questions such as:

“Is this compatible with X?”

or:

“What are the exact dimensions?”

A listing with fewer than 10 completed attributes may lack sufficient depth for AI-based recommendations.

Category consistency

The product must be classified in the category that accurately reflects its primary use.

A miscategorised product may:

• Lose search visibility

• Appear for irrelevant queries

• Fail to appear for its real category

• Be excluded from Alexa for Shopping recommendations

Non-Add-on status

An Add-on item cannot be purchased on its own. The shopper needs a qualifying order, often worth at least €25, before adding it to the basket.

This restriction significantly reduces conversion and may exclude the product from most Alexa for Shopping recommendations.

Amazon can automatically classify a product as an Add-on item based on price.

In some cases, a small price increase may be enough to remove this status.

Section 12

What Else COSMO Analyses

Beyond the traditional criteria used by A9, COSMO uses signals that are more difficult to measure but important for Alexa for Shopping visibility.

External traffic converting on Amazon

COSMO values products receiving traffic from:

• Google

• Social media

• Newsletters

• Influencers

• Brand websites

• Affiliate content

However, traffic alone is not enough.

External traffic that leaves the page without purchasing may become a negative signal.

External traffic that converts provides strong evidence of market demand independent of Amazon.

Review recency

A product with 300 reviews from 2023 may be seen as a product that had traction several years ago.

A product receiving 15 recent reviews within the last 60 days may be seen as having current traction.

COSMO values recent review velocity, not only total review volume.

Brand Registry

Amazon Brand Registry membership is an important trust signal.

COSMO may give more confidence to registered brands when generating recommendations.

Brand Registry is not mandatory for recommendation eligibility, but it can provide a meaningful advantage.

Return rate

Amazon increased the importance of return rates in 2026 as part of its focus on profitability.

A return rate two to three times higher than the category average may lead to:

• Loss of the Amazon’s Choice badge

• Reduced visibility

• Lower recommendation priority

• Increased negative review summaries

The product subtitle

Amazon introduced a subtitle field displayed directly below the main title on mobile.

This provides an additional semantic surface without consuming characters from the title.

COSMO analyses the subtitle as a separate source of information.

This makes it a valuable but underused field.

Amazon’s review summary

Amazon generates an automatic review summary on many product detail pages.

This summary is created by Amazon’s own AI using recent shopper reviews.

It may mention both strengths and weaknesses.

The summary becomes an additional source of information for COSMO when evaluating the product.

The ease of the buying journey

COSMO considers friction throughout the purchasing journey.

A product with:

• No Add-on restriction

• Prime eligibility

• Reliable stock

• A stable Buy Box

• A competitive price

• Clear return conditions

can be recommended across a wider range of situations.

A product with significant purchasing friction may be deprioritised even if the listing content is excellent.

Section 13

Taking Action: Where to Start

We have covered 60 criteria.

Optimising all of them at the same time is unrealistic.

The following order of priority provides the greatest impact with the lowest initial effort.

Priority 1: Fix complete invisibility

Start by checking criteria that can entirely exclude the product from recommendations:

• Add-on item status

• Rating below 4.0 stars

• Lost Buy Box

• Unstable Prime eligibility

• Incorrect category

• Product out of stock

• Incomplete structured attributes

These are on-and-off criteria.

Until they are corrected, other optimisation work may produce limited results.

Priority 2: Optimise the title and backend keywords

These are among the fastest and most powerful indexing improvements.

The title should contain the six main elements:

• Brand

• Product

• Function

• Target audience

• Differentiating feature

• Format or size

Backend Search Terms should ideally contain between 200 and 249 characters without exceeding the byte limit.

Together, these changes may significantly expand the number of queries for which the product is indexed.

Priority 3: Enrich bullet points with usage signals

Each bullet should contain at least one COSMO signal:

• User profile

• Usage context

• Usage timing

• Problem solved

• Complementary product

• Technical specification connected to a benefit

This is one of the strongest levers for Alexa for Shopping visibility.

Priority 4: Add lifestyle visuals

A listing containing only white-background packshots provides little contextual information to COSMO.

Add two or three lifestyle images showing a real person actively using the product in a recognisable environment.

This is one of the visual investments with the strongest potential return.

Priority 5: Generate recent reviews

COSMO values recent review velocity.

Two available levers include:

• Amazon’s “Request a Review” feature after purchase

• The Amazon Vine programme for eligible products

The objective is to maintain a regular flow of recent and relevant reviews.

The Most Important Point to Remember

None of these criteria works in isolation.

A product with a perfect title but a 3.8-star rating is unlikely to be recommended.

A product with a 4.7-star rating but a title targeting no relevant queries may never be found.

A product with excellent content but no stock, no Prime eligibility or an unstable Buy Box will also struggle to appear.

Alexa for Shopping visibility depends on the combination of all these criteria, not on a single optimisation.