Amazon listing optimization (2026): the complete guide, with real numbers

Short answer: optimize an Amazon listing in the order a buyer decides — main image and title first, then bullet points, the rest of the image set, A+ content, and backend keywords last. Every claim has to be provable, every field has to respect the current limits (titles have been capped at 75 characters since 27 July 2026), and any change you do not measure stays a guess. Optimization is not about fitting in more keywords; it is about raising click-through and conversion without dragging returns up with them.

This guide walks through each element, states the applicable limits with their dates, and closes with what a single A/B split test produced for one customer. Current as of August 2026.

What changed in 2026

Two changes affect almost every listing. If you last touched yours in 2025, both are new to you:

• Title length: per a Seller Central announcement (reported June 2026), titles in all non-media categories have been capped at 75 characters including spaces since 27 July 2026. Media categories are exempt, as are the Saudi Arabia, Egypt, Türkiye and UAE marketplaces.
• A new field, Item Highlights: 125 characters, searchable, displayed underneath the title. It absorbs some of what no longer fits in the shortened title.
• Rufus was retired on 13 May 2026 and folded into Alexa for Shopping — an agentic assistant sitting in the main search bar, on by default for signed-in US customers.
• COSMO, Amazon's semantic and common-sense layer, is publicly documented in a paper on Amazon Science. That Amazon now interprets the meaning of a query rather than matching strings is not speculation.

The title: 75 characters, and what belongs in them

The title is the only text field that appears in search results, in ads and in the cart. Since 27 July 2026 you get 75 characters including spaces in the affected categories — roughly half of what most sellers were used to. Keyword strings no longer work for the simple reason that they no longer fit.

Worth knowing: according to Amazon's announcement, non-compliant titles are gradually rewritten by Amazon's own AI. Brand-registered sellers get a 14-day review window. If you do not shorten your titles yourself, you eventually get a shortening you did not control.

• Build the title in this order: brand, product type, the one decisive attribute, then quantity or size if the purchase depends on it.
• Put the primary keyword up front — not because of a presumed position bonus, but because the tail of the title is truncated on a phone anyway.
• No repeated words. A term used twice costs characters and buys nothing.
• No promotional language (“bestseller”, “#1”, “sale”) and no decorative symbols. Both breach the guidelines and burn a budget you no longer have.
• Read the title out loud. If it does not sound like something a person would say, the semantic search layer understands it less well too.

Item Highlights: using the new field properly

Item Highlights gives you 125 searchable characters below the title. Do not treat it as an overflow bin for keywords that no longer fit above. It is the second line a shopper reads before scrolling, so it should carry the buying argument the shortened title could not: the use case, the audience, or the attribute that separates you from the three offers next to yours.

Because the field is searchable, the same rule applies as in the title: plain language, not a chain of terms.

Bullet points: five lines, two of which count

You get a maximum of five bullet points. In the mobile app, roughly the first two are shown before the rest collapses — and more than half of all Amazon purchases happen on mobile. In practice that means bullets one and two carry the purchase decision; three to five serve the shopper who is already interested and looking for detail.

• Benefit first, feature second. Not “420 g of stainless steel”, but what those 420 g mean for the buyer — with the number as the evidence.
• One bullet, one objection. The strongest bullets answer the concerns that show up in your competitors' reviews.
• Front-load each line. Mobile wraps early, and what sits at the end gets skipped.
• Do not repeat the title. Five lines are too few to say anything twice.
• No claim without a basis. Materials, certifications, effects and comparative claims must be provable — otherwise you risk suppression, not just weaker conversion.

Product images: the specs and the burden of proof

Images are the only part of a listing that works before the text does. The main image decides your click-through in search results; the rest of the set decides a large share of your conversion. The technical requirements are not negotiable — Amazon removes non-compliant images or suppresses the offer.

• Main image on pure white (RGB 255/255/255). No gradient, no staged background, no scene.
• The product fills at least 85 percent of the frame.
• At least 1000 pixels on the longest edge so zoom activates; 1600 × 1600 is the size worth shooting for.
• No logos, badges, borders, watermarks or Prime marks on the main image.
• For the remaining images: use case, scale reference, detail. A shopper who cannot judge the size returns more often.

The evidence rule applies to images too

What you write into an image needs the same backing as what you write into the copy. A certification badge you do not hold, a material claim that is not in the spec sheet, a “100% waterproof” without a tested standard — those are not design decisions, they are compliance exposure. Graphic text gets read and checked.

The practical handling is unglamorous: gather the evidence before you produce the images — spec sheet, test report, certificate — and drop every claim you cannot support. A weaker claim that holds beats a strong one that costs you the listing.

A+ content: free, but tied to Brand Registry

A+ content replaces the standard product description with modules of image and text. It requires Brand Registry and costs nothing. Amazon itself states that A+ content can lift conversion by “up to 20%” — that is Amazon's own claim and it is not independently verified. Treat it as an order of magnitude, not a promise.

On substance: A+ is where the things go that five bullets cannot hold — a comparison table across your own range, usage steps, brand context, answers to the second and third most common objections. What it is not: a second image gallery with the same shots.

Backend keywords: 500 bytes, not 500 characters

The Generic Keyword field is limited to 500 bytes, not 500 characters. The difference matters as soon as you leave plain ASCII: accented characters, umlauts and ß count as two bytes each. A field filled with German or Spanish terms is full earlier than your editor's character counter suggests.

And the limit is unforgiving. Exceed it and the entire field can drop out of indexing — not just the overflow. One word too many can neutralize every backend keyword you entered.

• Count bytes, not characters, and leave a buffer on non-English marketplaces.
• Do not repeat anything already in the title, bullets or description. It is indexed either way; the space is wasted.
• No competitor brand names. That is a guideline breach and a legal exposure.
• No commas needed as separators — spaces are enough. Every byte saved is another keyword character.
• Synonyms, spelling variants, colloquial terms and common misspellings belong here: everything a shopper might type but that would read badly in the visible copy.

Think mobile first

More than half of Amazon purchases happen on a phone, which reweights every element. The title truncates earlier, the shopper usually sees two of your five bullets, and an image whose text was designed for desktop zoom is unreadable on a handset.

Check every listing in the mobile app before you publish — not in the desktop preview. What sits in the first two screen-heights there is your actual listing. The rest is corroboration for people you have already convinced.

The AI layer: what Alexa for Shopping and COSMO actually change

Rufus was retired on 13 May 2026 and folded into Alexa for Shopping, an agentic assistant in the main search bar that is on by default for signed-in US customers. Alongside it, COSMO works as a semantic layer connecting queries to common-sense context. Together they mean that a growing share of shoppers no longer type keywords — they ask a question.

The consequence for your listing is concrete. An assistant asked “which pan works on an induction hob in a small kitchen?” can only answer from listings that state the use case and the audience explicitly. A title made of stacked nouns gives it nothing to work with.

• Spell out the use case: what for, for whom, in which situation — in a readable sentence, not by implication.
• State the limits and exclusions. “Not dishwasher safe” prevents returns and answers a question the assistant would otherwise have to guess.
• Write in complete statements rather than keyword fragments. What a person understands when read aloud, the semantic layer understands too.
• Stay with verifiable claims. An AI summarizing your listing will pass your exaggerations along — to a customer who settles the account at the returns desk.

What Amazon actually ranks

Search position is a behavioural contest, not a keyword contest. The two dominant levers are click-through rate (how many shoppers open your offer from the results page) and conversion rate (how many of those buy). Sales velocity compounds on top: selling more today means being found more easily tomorrow, which sells more again.

The third factor is routinely underestimated. Return rate carries increasing weight. A listing that lifts conversion with overstated promises and produces returns in exchange loses twice over. That is exactly why the evidence rule is not a compliance formality but part of the optimization.

Measuring and testing: Manage Your Experiments

Amazon gives brand-registered sellers its own A/B testing tool, Manage Your Experiments. It tests titles, main images, bullet points and A+ content against each other. It needs sufficient traffic on the tested ASIN and typically runs for several weeks, which makes it a tool for your top sellers rather than your long tail.

Test one variable per run. If you change the title and the main image at once and conversion rises, you know exactly as much afterwards as you did before.

A number from our own work: for one customer we ran the existing listing against a reworked version as an A/B split on Amazon. Conversion rate went from 2.1% to 4.2%, click-through was 38% higher between the variants, and returns fell by 27%.

That is a single test at a single customer — not an average, not a benchmark, and explicitly not a forecast for your product. What it does show is the order of magnitude that can sit between a weak and a carefully built version of the same offer. What your listing will do is something only your own test can tell you.

A working order that holds up

• Collect the evidence: spec sheet, test reports, your own reviews and those of your three strongest competitors.
• Rework the main image and the title — the two elements that decide click-through.
• Rewrite the bullets objection by objection, with the two most important ones on top.
• Extend the image set: use case, scale reference, detail.
• Build A+ content if you have Brand Registry.
• Fill the backend keywords last — only then do you know which terms the visible copy already covers.
• Let it run for two to four weeks, then measure; on high-traffic ASINs, test with Manage Your Experiments.

When a tool earns its place

For one or two listings you do not need software. The work is manual: read reviews, sort evidence, sharpen wording. The effort per listing is the same whether you have five or five hundred — and that is where the arithmetic breaks. Past a certain catalogue size the problem is not the quality of any single text, it is that half the catalogue has not been touched in two years.

A tool is worth it when it takes over three things: the evidence work, the format limits and the repetition. Torch Listings does exactly that — an ASIN becomes evidence-cited copy with its sources attached and TOS-conform images, and claims it cannot verify are blocked rather than invented. A+ content, variants and translations come out of the same system, though not out of the same run.

What no tool takes off your hands: deciding which claim your product can actually carry — and running the test that shows whether you were right.