The biggest change on Amazon heading into 2027 is that AI now sits on both sides of the marketplace. Alexa for Shopping, the assistant formerly called Rufus, answers shoppers with about five recommendations instead of a page of 50 results. Seller Assistant helps run store operations, and Ads Agent works on the advertising side.
For brands, that means product data, product quality and unit economics now decide more of the outcome than keyword rank alone. That was the core message of The 2027 Amazon Playbook, a Cart.com webinar held November 18, 2026, where Chris Mehrabi, President of Commerce Solutions at Cart.com, spoke with Jessica Binns, Director of Content at Berns & Co. and a contributing writer for Vogue Business.
Chris summed up the fall in one line: "The tools are getting a lot smarter, the shelf is getting significantly smaller, and the fees are going to go up." Here is what brands should take from the conversation.
Bar chart comparing 400 million Alexa for Shopping users in the past 12 months with 315 million unique Amazon customers per year, 250 million Prime members worldwide and 215 million Prime members in the US
"For the last 20 years, winning on Amazon has been about winning and dominating a keyword," Chris said. "Keywords are still critical, but the future that we're going to is much more about winning a conversation, a conversation with an agent that you're not there for, about your product."
The usage is already there. About 40% of Amazon app sessions touched Alexa for Shopping during last year's peak, according to Sensor Tower, and Amazon reports that about 400 million people used the assistant in the past year. Chris put that in perspective: Amazon has about 215 million Prime members in the US and roughly 250 million worldwide, and it gets orders from about 315 to 350 million unique people a year. People who aren't Prime members are coming to Amazon and testing these tools.
When the shelf shrinks, generic products suffer first. "AI is essentially raising the cost of being a generic product," Chris said. Products built for a specific person, use case or niche can stand out, as long as the data foundation is there for the AI to match them with the right shopper.
Brand size shapes how companies are reacting. Chris said smaller brands are excited, because they see AI as a way to run their store without adding cost. Larger brands are more cautious and want to protect the keyword rank they spent years building. They want humans in the loop, and almost all of them say the same thing about AI creative: don't use it. Shoppers can still tell when imagery is AI-generated, and it turns them off.
Amazon SEO isn't dead, but it is changing shape. Chris described three phases. Keyword SEO came first. Brands are now in a product data SEO phase, where the job is making sure Amazon's systems know everything about a product. The next phase is what Cart.com calls key answer SEO.
Key answer SEO is the practice of making sure a product's content answers the questions shoppers are likely to ask an AI shopping assistant. On Amazon, those answers need to live in three places: the product detail page copy, the Q&A on the listing, and uploaded product user manuals. Together they give AI agents what they need to match the product to the right shopper and use case.
AI can help write that content, but only as a first draft. "I would use it as a first draft, not a final copy," Chris said. Getting a good draft requires the right inputs, including the audience and intended use, not just product attributes. Every output needs checking, because a hallucinated attribute can sit unnoticed for a long time while the product quietly loses AI recommendations.
Brands are used to tracking keyword rank, and there isn't much good tooling yet for tracking how they show up in AI answers. Chris's advice is to measure it by hand. Go to Alexa for Shopping, move through the experience like a customer would, ask the top questions for your category, and note where you show up, where you don't and what the assistant gets wrong about your product.
It's also worth asking Alexa for Shopping what it knows about you. Chris tried it himself and found the assistant had built a detailed profile of his household, interests and brand preferences, including brands he had only ever bought at Whole Foods. That profile is what products are being matched against as Amazon moves from keyword search to matching.
Leading brands treat this as a leading indicator. Sales, ROAS and new-to-brand customers are lagging indicators. Visibility in AI answers shows where those numbers are heading.
Chris laid out a framework for deciding what to hand to Amazon's AI. First, can you set top-down guidance on what matters, such as a spending limit or a focus on new customer acquisition? Second, is the task high-frequency, so automation saves real time? Third, is it easily reversible if something goes wrong? Fourth, what is the cost of an error? And fifth, does Amazon have more data than you do, so its AI can actually drive better results?
Good candidates are account health monitoring, listing suppression alerts and inbound shipment planning, where an always-on system adds value and mistakes are easy to fix. Audience discovery is another, because Amazon has data brands don't. Chris pointed to DVA+, Amazon's rebranded DSP, where Amazon uses a brand's past customer data to find adjacent audiences. For one beauty brand Cart.com works with, that approach increased new-to-brand acquisition 50% year over year in its DSP campaigns and raised click-through rates 6x, while serving fewer impressions.
What brands should keep is just as clear. "I wouldn't give Amazon control over price right now," Chris said. "The cost is too high, and the incentives diverge." Amazon's AI is set to maximize revenue, which is how Amazon makes money, while most brands are trying to maximize profit.
Not every AI rollout is helping yet. Amazon now uses AI to handle seller support cases, and Chris said contacts needed to resolve policy, account health and listing suppression issues have increased sharply. "In some areas, it feels like we've taken a bit of a step back right now, hopefully in order to take two steps forward."
Full-Funnel Campaigns are Amazon's version of what Meta did with automated audiences and Google did with Performance Max. Chris called it "the final boss in advertising evolution." Cart.com tested Full-Funnel Campaigns across six clients and is still running them for one. Every campaign hit its ROAS targets, and new-to-brand acquisition looked good. But the increase in total ad spend was outweighed by a decrease in organic sales.
Chris couldn't say exactly why, because Amazon doesn't yet share which search terms trigger the ads or which placements the campaigns use. Full-Funnel Campaigns also aren't available programmatically yet, so for brands with an established advertising system, they sit in a separate box that can't be optimized alongside other campaigns. Large brands with efficient advertising should likely wait for more data. A new product or product collection that can be isolated from the rest of the catalog is a good way to test.
Sponsored prompts are less of a choice than they sound. Brands advertising on Amazon are part of sponsored prompts automatically, and Amazon writes the ads from the brand's listing copy, Q&A and user guides. Across the roughly 50 brands and more than half a billion dollars in annual sales Cart.com manages, sponsored prompts are about 1% of ad spend today. Ads Agent, meanwhile, describes itself as a research and analysis tool for now. Tasks like keyword harvesting and negation are coming, but they aren't there yet.
The holiday surcharge ends January 14, the permanent fee increase starts January 15 and storage fee changes follow on February 1. The average increase is about $0.19 per unit (PPC Land), but Chris said that average isn't meaningful. "Amazon is really telling us what type of inventory they want in their warehouse." Amazon wants smaller, high-velocity SKUs, though not items priced under $10. It doesn't want slow-moving inventory, so seasonal SKUs, high price point items and oversized, long or flat products need evaluating.
Chris called this Amazon's most aggressive fee move yet and its clearest statement about which business it wants to be in.
Brands should start by calculating profitability at the SKU level, allocating every Amazon cost to each SKU. Products that lose money after advertising are candidates to cut. Multipacks and bundles can raise the price point, and they've worked across apparel, beauty and men's grooming, especially for existing customers buying in larger amounts or through Subscribe and Save. Price points and packaging near Amazon's $10 and $50 thresholds or size tiers can sometimes move to a cheaper tier with a very small change. Oversized and slow-moving SKUs can shift to Seller Fulfilled Prime or regular FBM through a 3PL.
That last point is the hedge against future fee increases. A 3PL that serves DTC, Amazon and wholesale and retail distribution from one place lets brands meter small, high-velocity SKUs into FBA and fulfill everything else themselves.
Brands that lead with sales and ROAS aren't measuring wrong, Chris said, but they are measuring incompletely. Most brands want to grow in a specific way: acquiring new-to-brand customers without cannibalizing DTC sales, or selling multipacks to loyal customers instead of single units.
The single biggest waste Chris sees in Amazon ad spend is over-reliance on brand advertising instead of winning market share. Campaigns can show a strong ROAS and even plenty of new-to-brand customers while still overspending, if the brand isn't tracking how its Amazon ad products interact or the halo from off-Amazon advertising like TikTok, brand campaigns and PR.
For CFOs, Chris suggested two questions. What does the fee change do to our unit-level economics? And how will our team, or our growth partner, use AI to automate backend work like case handling, so people can focus on strategy and market share instead of triaging Amazon tickets?
"Peak is for execution, not for experimentation," Chris said. Brands that aren't already set up with Amazon's new automations should wait until January to deploy them.
Inventory is leaner this year. Cart.com has seen brands stock roughly 3% more holiday inventory than last year, compared with a more typical 10% to 15% increase. Two steps still fit before Black Friday. First, test the AI shopping experience yourself and note where your products show up and where they don't. Then run a short 48- to 72-hour sprint to fill the gaps: complete product attributes and copy, upload user guides and answer Q&A. Starting with the test keeps the sprint focused on the holes that matter.
Longer term, listing copy can't fix the product itself. "There's not a way to copy write yourself out of a defective product," Chris said. AI agents will spot recurring review complaints quickly, which tightens the cycle: brands will need to act faster, on less information, to stay in the game.
Chris closed with one recommendation. "Measure what matters. Get out there and interact with these agents, see where your product shows up and where it doesn't, and use that to create your checklist of what you need to fix." If every brand ends up with the same AI agents, designed to optimize the outcome for Amazon, differentiation comes back to who the product is for and how well it solves their problem.
Alexa for Shopping is Amazon's AI shopping assistant, previously called Rufus. Shoppers ask questions in natural language and it responds with a short list of recommended products, typically about five. Amazon reports about 400 million people used it in the past year.
No. Keywords are still critical, but Amazon SEO is moving from keyword optimization to product data optimization and, next, to key answer SEO, where listing copy, Q&A and product manuals answer the questions shoppers ask AI assistants.
For brands with an established, efficient advertising program, it's likely worth waiting. Cart.com's tests hit ROAS targets but showed organic sales declines that outweighed the added ad spend, and Amazon doesn't yet share search term or placement data. New products that can be isolated from the rest of the catalog are the best place to test.
Not yet. Amazon's AI is designed to maximize revenue, while brands are trying to maximize profit. Cart.com recommends keeping pricing decisions in the brand's hands.
FBA fees rise an average of about $0.19 per unit starting January 15, 2027, but the impact varies widely. Small, high-velocity SKUs fare best, while slow-moving, seasonal, oversized and sub-$10 items are hit harder.
Tooling is still limited, so the most reliable method is manual testing. Ask Alexa for Shopping the top questions for your category, track where your products appear and what the assistant says about them, and use the gaps as a fix list.
If you want a second opinion on your 2027 Amazon plan, Cart.com's marketplace team can review your product data, advertising and SKU-level economics. Get in touch to book a plan review.
Related reading: