If a listing service still pitches you on optimizing for Amazon's old chat-drawer assistant, you have learned everything you need to know in the first five minutes.
Evolve Media Agency provides Amazon listing optimization. I have written this as the evaluation framework I would want if I were buying, including the sections where the honest answer is that you should do it yourself or buy the cheaper tier.
2026 has been an unusually disruptive year for Amazon listings, and that is genuinely good news for buyers of this service. Not because change is pleasant, but because it makes vendor quality legible. A service that has adapted talks about different things than one that has not, and the gap is obvious once you know what to listen for.
The changes are substantial: titles capped at 75 characters, a new Item Highlights field, Premium A+ Content becoming free for eligible brands, and most significantly, the AI discovery layer moving from an optional chat drawer into the main search bar where every shopper encounters it whether they want to or not.
This guide is about buying the service well. If you want the vendor-by-vendor comparison instead, our review of ten Amazon listing optimization services covers the individual providers.
Amazon listing optimization — the discipline of structuring titles, bullets, descriptions, backend fields, structured attributes and supporting content to maximize discoverability, click-through and conversion. In 2026 it spans three audiences simultaneously: the traditional ranking algorithm, the AI assistant reading the listing to answer conversational queries, and the human deciding whether to buy.
What Changed in 2026
| Change | What it means | Why it filters vendors |
|---|---|---|
| 75-character title cap | Titles substantially shorter than the old limits | Keyword-cramming approaches no longer fit |
| Item Highlights field | New structured field for key product facts | A vendor unaware of it is not reading release notes |
| Premium A+ free | No longer gated behind spend for eligible brands | Changes what should be in scope at every price tier |
| Alexa for Shopping | AI moved from optional drawer into the search bar | Optimization is now for a default audience, not a niche one |
| Chat drawer retired | Predecessor assistant discontinued May 13, 2026; tech absorbed | Anyone still selling optimization for it is stale |
The Alexa for Shopping shift in plain terms
Amazon's previous AI shopping assistant launched in 2024 as a chat drawer shoppers had to deliberately open. It reached over 300 million customers during 2025, which sounds enormous until you remember that was still an opt-in fraction of Amazon's traffic. Plenty of sellers reasonably treated it as experimental.
On May 13, 2026, Amazon retired the standalone chatbot and launched Alexa for Shopping, which sits directly inside the main search bar, generates AI overviews above search results, and runs side-by-side product comparisons within the results page. There is no opt-in. Every signed-in US shopper encounters it.
The underlying technology and the optimization actions did not change. What changed is the stakes, because the audience went from a self-selecting minority to everyone.
Alexa for Shopping does not replace the traditional ranking algorithm. They operate simultaneously and you have to satisfy both. A listing that ranks well on keyword relevance but cannot answer a conversational question will lose an increasing share of discovery, and a listing built purely for the AI layer will not rank. Any vendor framing this as a migration rather than an addition has misunderstood it.
What the Service Actually Includes
Scope varies enormously between vendors at similar prices, which is why price comparison without scope comparison is meaningless.
Usually included
- Keyword research and a reverse-ASIN pass on competitors.
- Title rewrite, within the current character limit.
- Bullet point rewrite.
- Product description or A+ copy.
- Backend search terms.
Frequently excluded, and worth confirming
- Structured attribute completion. The unglamorous field-by-field work that feeds AI comparisons. Rarely quoted, increasingly the highest-value item on the list.
- Image and infographic production. Most listing services are copy-only.
- A+ module design and build, as distinct from writing the copy that goes in it.
- Item Highlights population. New field, frequently missed entirely.
- Implementation. Some vendors hand you a document and you do the uploading.
- Split testing and post-launch iteration.
- Variation and parent-child structure review.
Every empty attribute field on your listing is a question the AI cannot answer about your product. It is the least glamorous work in listing optimization and increasingly the most valuable.
The attribute point deserves emphasis because it is where the market has not caught up. Complete attribute data feeds AI overviews and side-by-side comparisons directly. A listing with beautiful copy and half-empty attribute fields will lose comparison placements to a plainer listing with complete data.
The Four Service Tiers
Keyword research and rewritten text. Often offshore, often templated. Fine if your copy is genuinely bad and your images are already good.
Better research, category-aware copy, usually A+ copy included. Attribute work and imagery still typically separate. Where most brands should start.
Integrated copy, attributes, A+ design, imagery, and post-launch iteration. Justified on high-revenue ASINs, wasteful on long-tail SKUs.
Monthly rather than per-SKU. Right when the catalog changes constantly or when listing work is part of broader account management.
The tier decision in one line
Match the tier to the ASIN's revenue, not to your catalog average. Premium work on a listing doing 300 sessions a month is a donation. Budget work on your top revenue ASIN is a false economy. Most catalogs justify a mix, and a vendor unwilling to price by tier is optimizing for their invoice rather than your outcome.
How AI Discovery Changed the Job
Understanding this section is what lets you tell a competent vendor from a confident one.
COSMO, and why you cannot optimize for it directly
COSMO is Amazon's commonsense knowledge layer — it connects products to situations, needs and contexts rather than just to keywords. You cannot target it the way you target a search term, and any vendor claiming a COSMO optimization service is overselling.
What you can do is produce the inputs it works from: complete structured attributes, natural-language copy that describes situations and uses rather than only features, and content that answers the questions shoppers actually ask. That is the practical output of understanding COSMO, and it is a legitimate scope item.
What the AI layer actually reads
- Structured attributes, which feed comparisons directly. Empty fields are unanswerable questions.
- Bullet structure, where leading with the specific feature then the benefit parses more cleanly than benefit-first marketing language.
- Natural-language description covering use cases, materials, dimensions, compatibility.
- Review and question content, which the assistant draws on for context you did not write.
- A+ content, including text within modules.
The honest position on external traffic
Amazon has not confirmed that external traffic directly improves AI recommendation ranking. The defensible framing is that external traffic builds sales velocity, reviews and conversion signals, which are confirmed ranking inputs for Amazon's performance systems generally. A vendor promising that driving outside traffic will improve your Alexa for Shopping visibility is asserting something Amazon has not said. Treat that as a flag.
The algorithm detail is covered in our COSMO vs A9 vs A10 breakdown, and the copy-level technique in noun-phrase optimization for Amazon's AI layer.
The 10 Questions
On method
- How has your process changed since May 2026? The single best filter in the list. A vendor who has adapted will mention Alexa for Shopping, the title cap, or Item Highlights unprompted. One who has not will talk about keywords.
- What is your keyword research source, and how do you validate it? Looking for named tools and, ideally, Search Query Performance data rather than third-party estimates alone.
- Do you complete structured attribute fields, or only write copy? If attributes are out of scope, you are buying half the job.
On scope
- Is imagery and A+ design included, or copy only? Most are copy only. That is fine if you know it.
- Who implements the changes — you or us? Handover documents that sit unused are a common and expensive outcome.
- What happens after the initial optimization? A one-shot rewrite with no iteration plan is a project, not a programme.
On accountability
- What do you measure, over what window, and what counts as failure? A vendor unwilling to name a failure threshold has committed to nothing.
- Do you split test, and who pays for it? Testing is how you learn what worked rather than assuming.
- Who actually writes it, and are you using AI to generate the copy? Not a gotcha — AI-assisted is fine and normal. Undisclosed and unedited is not.
- Can I see before-and-after data from a brand in my category? Anonymized is fine. Refusal is informative.
Question one. Amazon changed enough in 2026 that a vendor's answer immediately separates people running a current process from people running a 2024 template with a new date on the deck. You do not need to know the answer yourself to evaluate theirs — you only need to notice whether they volunteer specifics or reach for generalities.
Good Answers vs Worrying Answers
| Question | Good answer sounds like | Worrying answer sounds like |
|---|---|---|
| Changed since May? | Names the title cap, Item Highlights, Alexa for Shopping | "We always stay on top of updates" |
| Keyword source | Named tools plus Search Query Performance | "Proprietary methodology" |
| Attributes | Yes, field by field, with a completeness target | "We focus on the copy" |
| Implementation | We upload and verify, or a clear handover process | Unclear who does it |
| Measurement | Sessions, CVR, and a named window with a failure line | "You'll see improvements" |
| AI in copy | AI-assisted drafting, human editing, here is our process | Denial, or evasion |
| Case data | Anonymized before-and-after with real numbers | Testimonials only |
On the AI copywriting question specifically
Be pragmatic here. Nearly every vendor uses AI in drafting now, and there is nothing wrong with it — the tools are good and the efficiency is real. What matters is whether a human with category knowledge edits the output and whether the vendor is straight with you about the process. A denial is a worse signal than a yes, because it tells you they think you would object and they are willing to manage your perception rather than inform you.
Red Flags
- Optimizing for the old chat drawer. Amazon's previous assistant brand was retired in May 2026 and folded into Alexa for Shopping. A US-focused vendor still leading with it is quoting stale material.
- Guaranteed rankings. Nobody controls Amazon's algorithm. A guarantee means either meaningless conditions or manipulation you do not want associated with your account.
- Keyword density targets. A metric that stopped being the point some years ago and stopped working entirely once the AI layer started reading for meaning.
- Title recommendations over 75 characters. Either they have not read the current requirements or they are working from an old template.
- No mention of attributes anywhere in the proposal.
- Per-listing pricing with no tiering. The same price for your hero ASIN and a long-tail SKU means the process is identical for both, which means it is not tailored to either.
- Unwillingness to name a failure threshold.
- Pressure to buy the whole catalog at once. Optimizing five ASINs and measuring is the sensible sequence.
Comparing proposals?
Send us what you have been quoted. We will tell you what is standard scope, what is missing, and which questions to push on — whether or not you work with us.
Book a Strategy Call →The Ecom Profit Box
Eleven playbooks on listings, conversion, images, and email. Built for operators, no fluff, no email sequence.
Grab It Free →Pricing Benchmarks By Catalog Size
| Catalog | Sensible approach | Realistic spend |
|---|---|---|
| 1–5 ASINs | Mid-tier on all, premium on the top one | $2,000–6,000 one-off |
| 6–20 ASINs | Premium on top 3, mid-tier on next 7, defer the tail | $8,000–18,000 one-off |
| 21–100 ASINs | Tiered by revenue, phased over two quarters | $15,000–40,000 phased |
| 100+ ASINs | Retainer with a prioritized queue | $3,000–12,000/month |
| Constant new SKUs | Retainer plus a documented internal template | Retainer plus internal time |
The sequencing that saves money
Do not optimize the whole catalog at once. Take your top five ASINs by revenue, optimize them properly, measure for a full quarter, and use what you learn to brief the rest. You will discover things about your category — which claims resonate, which attributes matter, what the AI layer surfaces — that make the next forty listings materially better and cheaper.
The counterargument is consistency across a range, which is real for tightly related variations. Even then, phase by product family rather than doing everything simultaneously.
How to Measure Whether It Worked
Agree this before the work starts, because retrofitting a success definition after the fact produces arguments rather than answers.
The metrics that matter
- Sessions — did discoverability improve.
- Unit session percentage — did conversion improve. This is the cleanest signal that the copy and images did their job.
- Search Query Performance — observed impression, click and purchase data per query for your brand. Better than any third-party estimate.
- Query breadth — are you appearing for more distinct terms than before.
- Return rate — a listing that oversells raises returns, which is a failure disguised as a conversion win.
The timing rule
Give any single change a full fortnight before judging it, and a full quarter before judging the programme. Amazon's systems take time to re-evaluate a listing, seasonality confounds short windows, and the temptation to revert after eight days has ruined more optimizations than bad copy ever did.
What a fair failure threshold looks like
Something like: no measurable improvement in unit session percentage after 60 days on a listing with sufficient traffic to detect one. Note the qualifier — a low-traffic ASIN cannot produce a statistically meaningful read, which is another argument for optimizing your highest-volume listings first.
The reporting detail is covered in our guide to Search Query Performance and Brand Analytics.
DIY vs Service: The Honest Breakeven
Do it yourself when
- You have fewer than five ASINs and time to learn. The knowledge compounds and you will brief vendors better later.
- Your category is one you know deeply and the competitive copy is weak.
- The gap is attributes rather than copy. Filling structured fields is tedious but not skilled work, and it is frequently the highest-value item available. Nobody needs to be paid $800 to do it for you.
- Budget is genuinely tight. A competent DIY listing beats no optimization, and beats a $200 templated rewrite.
Buy the service when
- Your catalog is large enough that time is the constraint.
- Your category is competitive and the incumbent listings are strong.
- You need imagery and A+ design, which is genuinely skilled production work.
- You have tried and conversion did not move, which usually means the diagnosis was wrong rather than the execution.
Before hiring anyone, open Brand Analytics and look at Item Comparison to see which ASINs shoppers actually weigh against yours, then read your competitors' one and two-star reviews. That research costs nothing, takes an afternoon, and improves any listing brief more than an extra $500 of vendor budget will.
What to Have Ready Before You Engage
Vendors produce dramatically better work when handed these, and the effort is entirely yours to control.
- Your Search Query Performance export for the ASINs in scope. Observed data beats any estimate a vendor will buy.
- Your Item Comparison report from Brand Analytics, which defines your real competitive set behaviorally.
- Your own negative reviews, sorted and themed. These are the objections the listing has to pre-empt.
- Competitor negative reviews for your top three rivals. These are the gaps you can claim.
- Any compliance constraints — claims you cannot make, certifications you must reference, regulated language in your category.
- Current performance baseline — sessions and unit session percentage per ASIN, dated.
- Your actual differentiator, stated in one sentence. If you cannot, the listing will not either.
- Photography assets and their status, so the vendor knows what they are writing around.
Items three and four are the ones that change outcomes most and the ones brands almost never bring. A listing written from review mining reads differently from one written from a keyword tool, and the difference shows up in conversion rather than in rankings.
The Decision Framework
| Your situation | Do this |
|---|---|
| Copy is weak, images are good | Mid-tier copy-only service, plus DIY attributes |
| Images are weak | Fix imagery first; copy will not rescue a bad image stack |
| Attributes are half empty | Do it yourself this week. Highest value per hour available |
| Ranking but not converting | Conversion problem, not a discoverability one. Prioritize images and objection handling |
| Converting but low traffic | Keyword and attribute breadth, plus advertising |
| Large catalog, no time | Retainer with a revenue-prioritized queue |
| Competitive category, hero ASIN | Premium tier on that ASIN only |
| Not sure what is wrong | Diagnose before buying. Sessions vs conversion tells you which half to fix |
The three things to settle before signing
- Exact scope, itemized: copy, attributes, imagery, A+ design, implementation, iteration.
- The measurement window and the failure threshold, agreed by both sides.
- Who implements, and who verifies it went live correctly.
Settle those and the price comparison becomes meaningful, because you will finally be comparing the same thing. For the underlying craft, our high-converting Amazon listing guide covers the work itself, and the listing image stack guide covers the visual half that copy-only services leave to you.
The Short Version
- Pricing runs $200 to $500 per listing at budget tier, $400 to $1,200 mid-tier, and $1,500 to $5,000 for premium boutique work. Scope differences between vendors are larger than price differences.
- Amazon changed substantially in 2026: a 75-character title cap, a new Item Highlights field, Premium A+ free for eligible brands, and Alexa for Shopping launched on May 13, replacing the previous chat-drawer assistant in the main search bar.
- Alexa for Shopping runs in parallel with the traditional ranking algorithm rather than replacing it. You have to satisfy both.
- Ask question one first: how has your process changed since May 2026. The answer separates current vendors from ones running a 2024 template.
- Structured attribute completion is the least glamorous and highest-value item, and it is excluded from most quotes. Every empty field is a question the AI cannot answer.
- You cannot optimize for COSMO directly. What you can do is supply its inputs: complete attributes, natural-language copy about situations and uses, and content answering real questions.
- Give any single change a fortnight before judging it and a quarter before judging the programme. Optimize your top five ASINs first and let what you learn brief the rest.