Amazon Listing SEO: How to Write Listings That Rank (and Rufus Reads)

Amazon is a search engine for products, with one difference from Google that explains almost everything about ranking: Amazon only gets paid when somebody buys. Google can rank a page because it is relevant. Amazon ranks a listing because it is relevant and because it sells to the people who searched that term. Every piece of listing advice worth following comes back to those two words, relevance and sales.

What changed is the layer on top. Amazon’s shopping assistant, launched as Rufus and since renamed Alexa for Shopping in the US, now answers questions and narrows choices before a shopper scrolls a results page. And Amazon cut product titles to 75 characters. A listing written for the old keyword-matching game is now both too long and too vague.

Below: how the system works, what each field is for under the current rules, and a rewrite checklist. Where Amazon documents something we say so. Where it is folklore or vendor marketing, we say that too.

How Amazon search actually decides

Think of it as two gates.

Gate one is indexing. Your listing can appear for a search term only if Amazon associates that term with your product. That association comes mostly from text you control: the title, the new Item Highlights field, bullets, description, backend search terms, and the structured attributes in your product template. If a word is nowhere in the listing, you are usually not indexed for it, however relevant the product is.

Gate two is ranking. Among indexed products, order is driven by how each one performs for that query: clicks, add-to-carts, and above all purchases from people who searched it. Amazon does not publish the formula. “A9” was the name of Amazon’s search technology subsidiary and “A10” is a label sellers invented; neither is a spec you can read. What every rank tracker shows is the loop: sales on a keyword move you up on that keyword, the higher spot earns more sales, and the cycle repeats. A listing that indexes for a term but converts badly on it slides down.

The new layer is meaning, not just matching. Amazon researchers have published COSMO, a system that mines search-and-purchase behavior, has a language model propose why people bought what they bought, and stores the answers as a commonsense knowledge graph. Their example links slip-resistant shoes with pregnant women, a relationship no listing needs to state. Amazon describes it as deployed in search features. The assistant, separately, is documented as drawing on the product catalog, customer reviews, community Q&A, and information from across the web.

Claim	Status
Title, bullets, description, and backend terms are indexed	Amazon-documented
Item Highlights is a search input alongside the title	Amazon announcement and FAQ
Sales on a term drive rank on that term	Consistent with rank data; formula unpublished
The assistant reads catalog data, reviews, and Q&A	Amazon-documented
COSMO-style intent matching runs in search	Amazon research; deployment details sparse
A+ body text is indexed for Amazon search	Unconfirmed; seller tests disagree
“COSMO penalizes keyword density”	Vendor speculation

Amazon also says shoppers who use the assistant are 60% more likely to complete a purchase. That is a company-reported, attribution-based number: a sign of where Amazon is investing, not a forecast for your listing.

Every field under the current rules

Field	Limit	What it is for
Title	75 characters including spaces (media excepted)	Brand, product noun, two or three top terms
Item Highlights	125 characters	Materials, use cases, comparison terms
Bullets	Usually five; length varies by category	Verifiable benefits, Tier 2 and 3 keywords
Description	Hidden when A+ is live	Answers and specifics that do not fit the bullets
Backend search terms	250 bytes, not characters	Synonyms, spelling variants
A+ Content	Module-based	Conversion, not keyword storage

Titles. Titles still over 75 characters are being replaced, gradually, with Amazon’s AI-written recommendation. Brand Registry owners get 14 days to review and edit those rewrites; other sellers do not. The older rules still apply: no word more than twice (articles and prepositions excepted) and no promotional characters such as “!” or “$”. If you have not rewritten your titles, assume Amazon will, and that its version will be generic.

Item Highlights catches what the title lost: 125 characters shown beneath the title in search results and on the product page. Amazon frames it as the place for materials and recommended use cases, and its follow-up FAQ, as reported by sellers, says the item name and Item Highlights are both search inputs with neither prioritized. Some sellers report the field failing to save, so check that yours actually published.

Backend search terms cap at 250 bytes. English letters are one byte each; accented and non-Latin characters take two to four. Amazon’s guidance: no repeats of words already on the page, no competitor brands or ASINs, no punctuation. What happens over the limit is disputed (truncation or the whole field ignored), so stay under.

A+ Content. Amazon has never confirmed that A+ body text helps you index, and seller-written alt text on A+ images is being replaced by AI-generated descriptions, a rollout reported as complete in Europe. Write A+ to convert. A keyword that matters belongs in a field Amazon says it indexes.

Keyword research, in the order we do it

Our listing rewrites start as a Data Dive dive, with Helium 10 supplying the raw keyword data.

1. Pick true competitors. Five to ten listings selling the same thing to the same buyer at a similar price, not the best-seller that happens to share a keyword. Every later step inherits this choice.
2. Pull everything they rank for. Cerebro reverse-searches each competitor ASIN. Data Dive does it across the whole set and scores each keyword by relevance across all of them, a far better signal than search volume.
3. Fill the gaps. Run your core terms through Magnet for phrasings competitors missed, then add the words customers use in reviews.
4. Cut on relevance, not volume. A high-volume term for a different product earns impressions and no sales, and the loop punishes that.
5. Tier the survivors. Tier 1 defines the product and goes in the title. Tier 2 goes in Item Highlights and the first two bullets. Tier 3 fills the remaining bullets and description. Variants and leftovers go in the backend.

Data Dive’s listing builder shows a “Ranking Juice” score as you place keywords. It is Data Dive’s model, not an Amazon number, but it is a quick check that your top terms landed in the heavy fields. Helium 10 has its own Listing Builder for AI drafts and Listing Analyzer for scoring an existing listing. SmartScout has keyword and AI listing tools too, but it is a wholesale research tool first.

Still deciding whether the product is worth launching at all? Start with how to validate an Amazon product.

A worked example: a garlic press

A made-up stainless steel garlic press, written the old way:

Garlic Press Stainless Steel Garlic Mincer Crusher Heavy Duty Professional Garlic Press with Silicone Peeler and Cleaning Brush, Rust Proof, Dishwasher Safe, Easy Squeeze Garlic Chopper for Kitchen

That is 197 characters with “Garlic” four times, so it broke the repeated-word rule even under the old limit. Rewritten in 66 characters:

Stainless Steel Garlic Press with Silicone Peeler, Dishwasher Safe

Item Highlights, 122 characters:

304 stainless steel, crushes unpeeled cloves, soft-grip handles that need less squeeze, peeler and cleaning brush included

“Mincer,” “crusher,” and “chopper” move to the bullets and backend, where they still index. The title now reads like a product name a person would say out loud. Then the bullets do the specific work, each leading with the attribute a shopper is checking:

• Crushes cloves without peeling them. The 1.2-inch chamber takes two medium cloves; skins stay in the basket.
• Fits a small drawer. 7 inches long, 4.8 ounces.
• Cleans in one rinse. The basket swings out, and the included brush clears the holes.

Dimensions, weights, and what is in the box are what reviewers complain about when they are missing, and what an assistant can quote when someone asks “will this fit in my drawer?”

Writing for Rufus (Alexa for Shopping)

Amazon has not published how the assistant chooses products or how it weighs listing text against reviews. Anyone selling a “Rufus optimization score” is guessing. What is documented is the input list, and that points to a few rules. These are our reasoning, not Amazon’s instructions:

• Answer the questions shoppers actually ask. Read the one- to three-star reviews on your listing and your top competitors’. Each recurring complaint is a question someone had before buying (“does it rust?”). Answer it in plain words in a bullet or the description.
• Name the use case and the user. COSMO-style systems connect products to situations and audiences. “For small kitchens” or “left-handed friendly” gives the system something to connect. Skip health claims you cannot support.
• Fill every structured attribute. Material, dimensions, capacity, included parts. The catalog is a named input, and blank attributes are blanks in the answer.
• Fix what the reviews say. Reviews are an input you do not write. If the assistant’s summary mentions a flaw, copy will not outrun it.
• Test it. Open your listing in the Amazon app and ask the assistant the five questions your buyers ask most. If it answers from reviews, hedges, or says it is unsure, the listing does not say it clearly enough.

None of this fights keyword SEO. A listing that answers real questions in specific language is also the one that converts, and conversion feeds rank.

Ranking follows sales, so fix conversion too

You can index for every term in your niche and still sit on page three if shoppers do not click or do not buy:

• Main image. It is the click. Secondary images and infographics are where Canva earns its keep.
• Price against the visible set. Shoppers compare you to the results page, not to your cost of goods.
• Reviews. Slow to change, which is why the copy should defuse the top objection first.

Brand Analytics’ Search Query Performance report (Brand Registry required) shows, per search term, your share of impressions, clicks, add-to-carts, and purchases. It tells you which gate you are failing: low impressions mean indexing or rank, low clicks mean image or price, clicks without purchases mean the page. Sellerise surfaces Search Query Performance on its Growing plan, and Helium 10 has a Search Query Analyzer.

How to check indexing and measure a rewrite

• The search-box check. Type your ASIN and a keyword into Amazon search together. If your listing appears, you are very likely indexed for that term. It is a practitioner method Amazon does not document, so treat a miss as a prompt to dig, not proof.
• Rank tracking. Start tracking your Tier 1 and 2 terms before you change anything. Data Dive’s Rank Radar, Helium 10’s Keyword Tracker, and Sellerise’s keyword tracker (20 keywords on Starter) all do it. Give a rewrite two to four weeks.
• One change at a time. Change the title and main image in the same week and you will never know which one moved the numbers.

The rewrite checklist

1. Pick five to ten true competitors and pull every keyword they rank for.
2. Cut on relevance, then tier the survivors.
3. Title of 75 characters or fewer: brand, product noun, top terms, no word more than twice.
4. Item Highlights in 125 characters: materials, use cases, comparison terms. Confirm it published.
5. Five bullets, each leading with a checkable attribute and a number.
6. Answer the top review complaints and buyer questions in plain sentences.
7. Fill every structured attribute in the product template.
8. Backend terms under 250 bytes: synonyms and variants, no repeats, no competitor brands.
9. A+ as conversion copy, not keyword storage.
10. Ask the assistant your buyers’ top questions and fix whatever it cannot answer.
11. Start rank tracking first, change one thing at a time, wait two to four weeks.

For the full set of tools we run the store with and what each costs, see the best Amazon FBA tools.

FAQ

Is keyword SEO dead now that there is an AI assistant? No. The assistant still needs to know what your product is, and the results page still runs on indexing and sales. What died is the 200-character keyword-stuffed title.

Should I let Amazon’s AI rewrite my title? Only after reading it. Brand owners can edit before it goes live, and a title written from a real keyword list will almost always beat a generic one.