Amazon is removing seller-written alt text from A+ Content modules in a phased rollout that began in 2025 and continues through 2026 - and has published nothing about it. That is one of the central findings in a detailed analysis published today by ZonGuru, the Amazon seller tools company. The change, combined with new research into how Amazon's COSMO knowledge graph actually works, reframes what A+ Content is for. The question was never whether Amazon indexes the copy. The question is what the copy makes shoppers do.

Amazon has built one of the largest product content systems in retail, yet the fundamental question that sellers ask about it - does A+ Content get indexed by Amazon's A9 search algorithm? - remains officially unanswered. What is now documented, through practitioner testing, Amazon's own published research, and observable platform changes in 2025 and 2026, is a more useful answer: A+ body text does not move keyword rankings, the alt-text field that sellers once controlled is being handed to AI, and the real mechanism through which A+ Content shapes Amazon's recommendation systems runs through shopper behavior, not text parsing.

The distinction matters because Amazon's AI shopping layer changed substantially on May 13, 2026. That day, Amazon retired the Rufus brand and launched Alexa for Shopping, a unified assistant combining Rufus's product reasoning with Alexa+'s personalization. Underneath that surface sits COSMO, Amazon's product knowledge graph, which Amazon described in its own SIGMOD 2024 paper. COSMO contains 6.3 million nodes, 29 million knowledge edges, and spans 18 product categories. How it gets built - and how A+ Content feeds it - is the part most optimization guides miss.

What A+ Content actually is in 2026

A+ Content is the long-form visual block - image panels, text modules, comparison charts, video - that appears in the product description area on brand-registered listings, according to Amazon's A+ Content Design Guide. Amazon offers it in two tiers: Basic A+, and Premium A+, plus the Brand Story format. The older term Enhanced Brand Content, or EBC, was retired around 2019. Basic A+ supports up to five modules from a library of 14, with images up to 970 x 300 pixels. Premium A+ supports up to seven modules from a library of 19, with larger images at 1,464 x 600 pixels, plus interactive hotspots, multiple video modules, and enhanced comparison charts.

Amazon makes its conversion case directly in its own published materials. Basic A+ can lift sales up to 8%, according to Amazon's claims. Premium A+ raises that ceiling to 20%. Neither figure comes with methodology notes, but they appear consistently across Amazon's help documentation and are the numbers most widely cited when sellers make the business case for investing in A+.

Eligibility is straightforward: a Professional selling account, Amazon Brand Registry, and publishing rights under the brand on the ASINs being worked on. Sellers build, preview, and submit content through the A+ Content Manager inside Seller Central.

The indexing question Amazon will not settle

No Amazon engine - not A9, not COSMO, not Alexa for Shopping - has ever been confirmed to read A+ body text. What is settled through practitioner testing rather than official documentation is that A+ body text does not move A9 keyword rankings. Major Amazon agencies have split-tested this for years, and results consistently agree. The only official confirmation on record comes from a staff reply in Amazon's own Seller Central forums - a response from a moderator identified as Glenn - which confirms that image alt text "helps your products in search results" and says nothing at all about body text.

There are structural reasons Amazon keeps the indexing question vague, according to the analysis published by ZonGuru on June 22, 2026. Confirming that any field is indexed invites keyword gaming at scale. Amazon's internal philosophy also separates the listing into two distinct jobs: discoverability (titles, bullets, backend search terms) and conversion (A+ Content). By declining to confirm body text indexing, Amazon pushes sellers toward writing A+ for human readers - which, as it turns out, is exactly the kind of content that produces the behavioral signals COSMO is built from.

The Google dimension adds a fourth structural reason. Google crawls and indexes A+ body text as part of the product detail page, according to ZonGuru's documentation. A+ copy drives external traffic to Amazon via Google Search. If Amazon officially announced that A9 does not index A+ text, some brands would remove or thin out the rich copy in their modules - cutting off outside traffic that Amazon wants.

The alt-text removal: what changed in 2025 and 2026

The part that changed under sellers' feet is the alt-text field. In a phased rollout that began in 2025 and expanded through 2026, according to ZonGuru's analysis, Amazon is removing the seller-written alt-text field from A+ modules and Brand Story. An AI now generates image descriptions instead of accepting seller input. Europe already has full removal as of 2026. A wider rollout is expected. Amazon has published nothing about any of it.

The practical consequence is significant. The single surface that Amazon staff ever officially confirmed as contributing to search results - image alt text - is now out of sellers' hands. Amazon's A+ Content Manager saw related changes earlier this year: on February 22, 2026, two simultaneous changes appeared as banner notifications inside the tool. Shoppable Collections Beta ended on February 27, 2026, with Brand Story replacing it. Content Quality Analysis launched in beta, evaluating A+ pages weekly across readability, information completeness, visual presentation, and conversion effectiveness. Neither change was communicated by press release or direct seller email.

The alt-text automation means the optimization lever shifts from metadata to image quality. If Amazon's AI is generating the description, the image needs to be unambiguous enough for that AI to describe it accurately. A short text overlay naming the benefit - say, "removes pet hair from car seats" - guides the vision model, the shopper, and Google's crawl of the rendered page simultaneously.

How COSMO builds and why A+ Content feeds it

COSMO is not built from listing copy. It is built from shopper behavior - specifically, two types with strong purchase intent: search-buy pairs (a shopper searches for a term and purchases a product within the session) and co-buy pairs (products purchased together in the same basket). According to Amazon's SIGMOD 2024 paper, Amazon sampled millions of these pairs across 18 product categories to construct the graph.

The mechanism runs in three steps, as documented in that paper. First, Amazon logs behavior from session data. Second, an LLM hypothesizes why each search-buy or co-buy pair occurred, and Amazon stores that hypothesis as a typed relationship. Amazon's own example: a "winter coat" query led to a Long Sleeve Puffer Coat purchase because the product "is capable of providing high-level warmth." Human annotators then filter each hypothesis for plausibility and typicality, and surviving knowledge becomes an edge in the graph. Third, when Alexa for Shopping handles a conversational query, the relationships COSMO has already stored determine which products appear in the answer.

Amazon's SIGMOD 2024 paper also reports a measurable outcome: deploying COSMO in Amazon's search navigation on roughly 10% of US traffic produced a 0.7% sales lift. For a company generating the volume Amazon does, that number represents hundreds of millions of dollars annually. COSMO is production infrastructure, not a research experiment.

The connection to A+ Content follows directly. Sellers do not control Amazon's session logs. But sellers do control what converts. When a shopper arrives from a specific query and A+ Content closes the sale, that creates a search-buy pair. When a lifestyle module shows the product working alongside a companion item and the shopper adds both, that creates a co-buy pair. The graph learns whatever the conversions teach it. As PPC Land has covered, COSMO's role in filtering recommendations is significant: research from Workflow Labs, published April 17, 2026, found that Amazon's AI compresses the effective product discovery space from roughly 50 results to approximately five, with COSMO's structured backend attribute fields determining which products enter the pool at all.

The fifteen COSMO relationships

Amazon's SIGMOD 2024 paper names 15 relationship types that COSMO stores. According to ZonGuru's analysis, these relationships form a content brief written by Amazon's own researchers - they identify which intents the graph carries and, therefore, which conversions sellers should engineer for.

Three clusters carry the most structural weight per A+ module slot.

Use-context relationships - used_for_event, used_in_location, used_on - live in lifestyle imagery and video. Each scene in a lifestyle module represents one use-context the listing can win conversions on. Amazon's own design guide cites brands leading with video to show products fitting into specific lifestyle scenarios.

Comparison and identity relationships - is_a, used_as - live in the comparison chart. A chart built against named competitors, with the use case each fits, answers the shortlisting question a shopper is weighing. That is a richer structured statement than a self-referential variant chart, and it is the format that maps most directly onto how COSMO categorizes products relative to alternatives.

Audience-fit relationships - used_for_audience, used_by, xIs_a, xWant - live in the Brand Story. Amazon's design guide cites Jackery, whose Brand Story explains the brand was founded "to help nature enthusiasts enjoy the outdoors sustainably" as an example of an audience-fit claim committed in a single banner-plus-narrative pair. According to ZonGuru's analysis, decorative A+ pages typically cover only two or three of the fifteen relationships, usually only the functional use relationship, with every gap representing a sale the listing cannot close.

ZonGuru's AI Readiness Score, run across more than 5,000 live Amazon listings, returns a median score of 65 out of 100 across two dimensions: COSMO Semantic Mapping and Alexa for Shopping Q&A Coverage. The COSMO half lands at a median of 70. The Q&A Coverage half lands at 54. Same listings, same seller effort, different results depending on which dimension is measured.

The hidden backend description and the indexed gap

There is a structural quirk in how A+ Content works that creates a common and costly mistake. When A+ Content goes live on a listing, it visually replaces the plain-text Product Description field on the front end. Shoppers never see the old text field. But that field stays in the listing's backend, and practitioner consensus is that A9 keeps indexing what is written there.

The failure pattern: brands stop maintaining the backend description field after publishing A+, reasoning that A+ has replaced it. The result is a listing with an empty or outdated text field in the one indexed description slot A9 actually reads - while the visible layer carries imagery and rich modules that A9 does not parse. With seller-written alt text now being removed by Amazon's AI rollout, the backend description is the last hidden metadata field sellers fully control.

Amazon's July 27, 2026 title character limit change fits the same structural direction. Amazon is capping all product titles at 75 characters across all categories except media, with content beyond that threshold redirected to a new Item Highlights field of up to 125 characters. Titles not updated by July 27 will be automatically rewritten by Amazon's AI. The change announced by Amazon on June 10, 2026, and updated on the Amazon Seller Forums, represents another step toward machine-readable structured data rather than human-optimized copy - the same direction COSMO's architecture reflects. Both changes narrow the surfaces sellers control and raise the stakes of what they write on the surfaces that remain.

What decorative A+ actually costs

ZonGuru identifies four concrete costs from A+ Content that does not connect to conversion intent.

The first is lost conversion against Amazon's own numbers. The 8% and 20% lift bands assume a coherent listing - A+ reinforcing titles and bullets, not contradicting them. A+ Content that contradicts the surrounding copy or mismatches the product's actual audience does not reach those bands.

The second is the neglected indexed description, covered above. A+ that replaces the visible description without maintaining the backend field surrenders the only A9-indexable description slot.

The third is corrupted training data. This is the behavioral cost. Contradictory A+ converts the wrong shoppers: a banner positioned for outdoor use closes sales from outdoor queries for a product that is actually a home unit. Those mismatched conversions become search-buy pairs that teach COSMO an intent the product does not serve. Returns and negative reviews that follow are additional negative signals on the same listing. Decorative A+ does not simply fail to train the graph; it can train it incorrectly.

The fourth is unreadiness for direct retrieval. Whether Alexa for Shopping retrieves A+ text directly is undocumented. Amazon's published training inputs for the Rufus system, as documented in Amazon Science and the AWS Machine Learning blog, include the full product catalog, customer reviews, community Q&A posts, and public web content. A+ Content is not named specifically, and the phrase "product catalog" is broad enough to include it at Amazon's discretion. Three of ten 2026 A+ guides claim the AI reads A+ in full; none of those guides cites an Amazon source. The behavioral path is documented; the direct-retrieval path remains unconfirmed.

Premium A+ and the eligibility gate

Premium A+ unlocks the Q&A module and higher-bandwidth comparison formats - the two module types carrying the most structured-knowledge weight per slot - and roughly doubles Amazon's published lift ceiling from 8% to 20%.

The eligibility gate contains a terminology trap. According to ZonGuru's June 22, 2026 analysis, Amazon's announcement phrases the requirement as "five approved A+ Content modules" in the trailing 12 months, but the counted unit is the submitted layout, not the design block. One full A+ page holding five design blocks logs as one submission. Sellers in Amazon's forums report sitting on seven approved modules for a single product with no Premium access. The full gate requires a Professional selling account, Brand Registry, a published Brand Story on every brand-owned listing, and five approved A+ Content project submissions in the trailing 12 months. The older threshold of 15 submissions is outdated - five is the current figure, and access is granted automatically via a banner in the A+ Content Manager.

Why this matters for the marketing and advertising community

The mechanics documented here intersect with several developments that the marketing community has been tracking closely. Alexa for Shopping launched on May 13, 2026, combining Rufus's capabilities with Alexa+'s personalization layer and becoming available to all US customers on the Amazon Shopping app, website, and Echo Show devices in the following week. Rufus had reached 300 million users and generated nearly $12 billion in incremental annualized sales during 2025, with monthly active users up 149% year-over-year and total conversational interactions up 210%.

For anyone managing Amazon advertising alongside organic listing strategy, the COSMO mechanism reframes the full-funnel question. Paid placements drive traffic. A+ Content converts traffic. Conversions train the graph. The graph determines which products appear in AI-mediated recommendations. Those recommendations now handle a growing share of product discovery: Rufus handled 38% of all Amazon sessions during Black Friday 2025, according to Workflow Labs research published April 17, 2026. Customers who engaged with Rufus during shopping sessions showed a 60% higher purchase completion rate, according to Amazon's Q3 2025 earnings call with investors.

Amazon's retail media advertising business reached $17.2 billion in Q1 2026, up 24% year-over-year. Retail media ROAS has climbed to roughly $5.80, the strongest reading in approximately 24 months, according to CommerceIQ data published June 18, 2026. Conversion rates across CommerceIQ's platform data averaged 29.8% through the first five months of 2026, up from 26% over the same period in 2025. That efficiency gain is directionally consistent with AI-assisted decision-making compressing the consideration phase - which is exactly what COSMO's behavioral architecture is designed to produce.

The product data question also extends beyond A+ Content itself. Research published April 17, 2026 by Workflow Labs, covered by PPC Land, found that most brands leave roughly half of their backend structured attribute fields empty. COSMO reads those fields more heavily than front-end copy. If those fields are blank, the product cannot be correctly classified, and Alexa for Shopping does not include it in responses to relevant queries.

The pattern Amazon is building - structured data over copy, behavioral training over text parsing, AI-generated metadata over seller-written fields - is consistent across the product title change, the alt-text rollout, the COSMO architecture, and the Alexa for Shopping launch. A+ Content sits inside that pattern as the highest-leverage conversion surface on the detail page, which makes it the highest-leverage point for influencing what the graph learns.

Timeline

  • Around 2019 - Amazon retires the Enhanced Brand Content (EBC) name, merging it into A+ Content with Basic and Premium tiers
  • February 2024 - Amazon launches Rufus in beta as a shopping AI for US customers
  • July 12, 2024 - Amazon expands Rufus to all US customers ahead of Prime Day
  • 2025 - Amazon begins phased rollout removing seller-written alt-text fields from A+ modules and Brand Story; AI generates image descriptions instead
  • September 2025 - Amazon announces four new Echo devices with AZ3 and AZ3 Pro chips engineered for Alexa+; Lens Live launches
  • November 2025 - Rufus reaches 250 million users; monthly active users up 149% year-over-year; total interactions up 210%
  • November 10-11, 2025 - Amazon launches unified Campaign Manager at unBoxed 2025, consolidating Amazon DSP and Ads Console
  • February 7, 2026 - Amazon reports Rufus reached 300 million users and generated nearly $12 billion in incremental annualized sales during 2025
  • February 22, 2026 - Amazon's A+ Content Manager displays two simultaneous changes: Shoppable Collections Beta ending February 27, and Content Quality Analysis launching in beta; neither announced by press release
  • February 27, 2026 - Amazon removes Shoppable Collections Beta from all product detail pages; Brand Story automatically replaces it
  • 2026 - Alt-text removal reaches full removal status in Europe; broader international rollout continues
  • April 17, 2026 - Workflow Labs CEO Justin Leigh publishes research finding Rufus compresses product discovery from roughly 50 results to approximately five; Rufus handled 38% of all Amazon sessions during Black Friday 2025; covered by PPC Land
  • May 1, 2026 - Amazon expands Rufus price history to 365 days; over 50 million customers have used the feature since launch
  • May 13, 2026 - Amazon retires the Rufus brand and launches Alexa for Shopping, merging Rufus and Alexa+ into a single AI assistant available on the Shopping app, website, and Echo Show
  • June 10, 2026 - Amazon posts and updates Seller Forums announcement capping product titles at 75 characters from July 27, 2026
  • June 18, 2026 - CommerceIQ publishes data showing Amazon retail media ROAS at roughly $5.80, strongest in 24 months; conversion rates at 29.8%, up from 26% in the same period of 2025; Amazon advertising at $17.2 billion in Q1 2026
  • June 22, 2026 - ZonGuru publishes analysis titled "Amazon A+ Content: The Layer That Trains Amazon's AI Without Being Indexed," documenting the COSMO mechanism, the 15 relationship types, the alt-text rollout, and the AI Readiness Score findings across 5,000+ live listings
  • July 27, 2026 - Amazon's 75-character title limit enforcement begins; listings above the threshold receive AI-generated rewrites gradually

Summary

Who: Amazon sellers and vendors using A+ Content on the Amazon marketplace, and the marketing teams and agencies that advise them on listing optimization and retail media strategy.

What: A detailed technical analysis published June 22, 2026 by ZonGuru documents how Amazon's A+ Content functions within the AI shopping layer in 2026: body text does not index in A9 and has never been confirmed to be read by COSMO or Alexa for Shopping; the seller-written alt-text field is being removed in a phased rollout across markets; and A+ Content's real mechanism for influencing Amazon's AI recommendations runs through conversion behavior, not text parsing. COSMO - Amazon's product knowledge graph with 6.3 million nodes and 29 million edges across 18 categories, per Amazon's own SIGMOD 2024 paper - is built from search-buy and co-buy pairs sampled from session logs, meaning every sale A+ Content closes on a specific query functions as a training signal for which products Amazon's AI recommends for that intent.

When: The analysis was published June 22, 2026. The alt-text removal rollout began in 2025 and continues through 2026. Alexa for Shopping launched May 13, 2026. The 75-character title limit enforcement begins July 27, 2026.

Where: The Amazon marketplace, primarily in the United States, with alt-text removal already at full coverage in Europe and continuing internationally.

Why: The marketing community's understanding of A+ Content has been anchored to the wrong question - whether A+ text is indexed - rather than to the documented mechanism: that A+ Content trains Amazon's AI through conversions, and conversions are the raw material of COSMO. The changes in 2025 and 2026 narrow the surfaces sellers control and raise the stakes of what is written on the surfaces that remain.