Two audiences, one listing
Every listing now has two readers: the shopper looking at the page and Amazon’s AI discovery systems, including Alexa for Shopping (formerly Rufus) and the COSMO knowledge graph, which can shape what a shopper sees before they ever reach the page.
The shopper reads copy and images, while the AI systems read attributes, structure, and context, so a listing built around keyword repetition alone tends to satisfy neither one.
That is why keyword research still matters, but it has a different job: COSMO matches intent instead of exact phrases, so a listing for “running shoes” can miss a shopper asking about plantar fasciitis on pavement, no matter how often the phrase appears.
What Alexa for Shopping changes
Alexa for Shopping now sits between your listing and the buyer, and it does more than answer questions, because it can compare products, show price history, schedule reorders, and buy automatically at a target price.
Scheduled Actions re-evaluate the best matching product every time they run, so a competitor with cleaner listing data can win a reorder without the buyer ever choosing to switch, while Auto Buy fires on base price alone, applies only to FBA inventory, and ignores coupons.
Price history adds another wrinkle: buyers can see 30, 90, and 365 days of pricing, and a discount that repeats on a schedule reads as a pattern rather than urgency.
Your catalog data is now part of the sales pitch.
Keyword research that serves meaning
Keyword research is still the starting point, though the goal has moved from placing a term in every field to treating each term as an anchor for content that answers the question behind it.
Product Opportunity Explorer, Brand Analytics, and Search Query Performance show what buyers search for, and search bar suggestions add the phrasing buyers use in real life, giving you language to write around instead of a list to repeat.
From there, the work covers intent, who the product is for, what problem it solves, and when it's used, since those are the questions an AI assistant tries to match.
Titles and Item Highlights
Titles are where the newest rule lives because Amazon’s 75-character limit means a longer title can be replaced with an AI-generated version built from your existing listing data, keeping the listing active but handing one of its most visible elements to a system that prefers brevity over the keyword logic you spent months building.
A 75-character title has room for the brand, the product type, and one or two differentiators, and something like “Acme Pro Insulated Water Bottle, 32 oz, Stainless Steel” covers all of that in 55 characters.
Whatever no longer fits has a home in Item Highlights, a 125-character field that is searchable and visible in results and on the detail page, so it works best when treated as a fifth bullet instead of an overflow bin.
Brand owners also get a 14-day window under Review Listing Changes to approve or replace an AI suggestion before it goes live, which is worth knowing before anyone has to find out the hard way.
Bullets and descriptions
Bullets and descriptions look similar, but they work best when each does a separate job, since repeating the same signals in slightly different words makes the content feel redundant and gives an AI assistant less to work with.
Amazon allows five bullets, and each one reads well as an answer to a question a shopper might ask, such as what the product is made of, who it is for, what it solves, and how it differs from the alternatives, with a short label up front and roughly 100 characters of explanation.
Without Brand Registry, the description is a 2,000-character plain text field that should reinforce the main use cases in your own words, because copied manufacturer text performs worse, and with Brand Registry, A+ Content takes its place.
Images and A+ Content
Images carry the most weight for conversion, so the basics hold, a white background main image with the product filling at least 85% of the frame and at least 1,000 pixels on the longest side, followed by lifestyle images that show a clear use and infographics that call out dimensions, materials, and specifications.
A+ Content adds a part many sellers overlook: text typed into the module fields is indexed, and image keyword fields work like alt text, while text burned into an image isn't read at all, so a key benefit placed only in a graphic is invisible to search.
Mobile deserves the first look, because modules that read cleanly on desktop can stack into something unreadable on a phone, and compliance deserves the second, since superlatives, guarantee language, pricing, competitor names, and contact details are the usual reasons a submission gets rejected and sent back through a review that can take up to seven business days.
Backend keywords and the ranking system
Backend search terms are the one field built purely for search coverage, and the rules are strict, with a limit of 250 bytes instead of characters, which means accented letters from Spanish terms cost extra and one byte over the limit can stop the whole field from indexing without any error message.
Words already in your title are indexed, so repeating them wastes space, while synonyms, alternate names, use case phrases, and Spanish terms add reach, and a single brand name, superlative, promotional phrase, or unsupported health claim can suppress the field instead of just the offending word.
Once the string is saved, give it 24 to 48 hours and test a few priority terms by searching your ASIN followed by the keyword, since saving and indexing are two different events.
Beyond that, ranking tends to improve in stages: clean up search terms, refine copy, then focus on keywords between positions 20 and 50, where small gains move the most traffic.
Backend structure and implementation
Here is the part of optimization that rarely appears in a brief, which is making sure the changes take effect correctly inside Amazon, since a listing can have excellent copy and still underperform because of a wrong category, a broken variation family, or missing attributes that the AI systems rely on to match your product.
Category and browse node placement, variation relationships, and attribute completeness (a fill rate of 90% or higher is a good target) all shape visibility without ever showing up as an error, and flat file uploads add their own risks, because a blank cell in a full update clears what was there before, while PartialUpdate changes only the fields you fill in.
A short routine protects against most of it: start with a test batch of 5 to 10 products, then run a suppressed listings check within 48 hours of any bulk upload, and save a fresh Category Listing Report beforehand as a backup.
A monthly rhythm for your listings
With the pieces mapped, a monthly pass keeps optimization from becoming a one-time project, and it takes less time than it sounds when one person owns it.
Start with the top ASINs, checking titles against the character limit, Item Highlights, and whether bullets still answer the questions buyers ask, and then look at the search terms field for byte count and anything that could trigger a flag.
Next, check attribute completeness and variation structure, and after any bulk upload, look for suppressed listings, since a file can process with zero errors and still drop a listing from search within 48 hours.
Before scaling ad spend or running a promotion, confirm the listing is structurally sound and review the price history a buyer would see, because ads send traffic to a page and do not fix it.
The aim is recognition, spotting what needs a closer look without having to diagnose it yourself.
When something needs a closer look
Some findings from that pass take a minute to fix, and others open into a diagnosis that runs for days, like a listing with strong copy that still stalls or an upload that processed cleanly and then disappeared from search.
That is usually where more reading stops helping, because the words on the page and the data behind them are separate problems, and your team’s time is better spent on the product and the growth plan, which is the expertise you already have.
So if you would like a second set of eyes on a listing that is not behaving, reaching out is an easy place to start, and we will take the Amazon part from there.