Amazon just quietly shelved most of its flagship AI models — the ones running behind Rufus and its AI seller tools — to focus resources on one model built to compete with OpenAI, Anthropic, and Google.
Meanwhile, ChatGPT turned brand mentions into clickable homepage links in May, and referral traffic is surging — but marketplace-only sellers with no homepage to send people to may be the ones left out.
Here's what changed this week, and what to do about it.
- 🤖 Amazon just shelved most of its flagship AI models — here's why sellers should care
- 🔊 Alexa for Shopping just became Amazon's main discovery engine
- 💻 You cut your title to 75 characters. Which keywords did you stop ranking for?
- 📈 ChatGPT just became a real traffic source. Marketplace sellers might be left out.
- 🛒 Walmart Connect finally lets advertisers block search terms
AMAZON NEWS
Amazon is quietly stepping back from its own flagship AI models.
This week, reporting confirmed the company is winding down Nova Premier, Nova Omni, and the Reel and Canvas generative models — moving them to "keep the lights on" status while it builds one new frontier model meant to compete with OpenAI, Anthropic, and Google.
🧠 What's actually changing
The new effort, called Frontier Model Research, is led by Pieter Abbeel, the UC Berkeley professor who joined Amazon through its 2024 Covariant acquisition. Nova 2 Lite, Nova 2 Sonic, Nova Forge, and Nova Act stay in active development — everything else gets folded into a single flagship model Amazon hopes to debut at re:Invent later this year.
📉 Why now
The reshuffle follows a string of senior AI departures, including former Alexa AI head Rohit Prasad, and a reorganization that now reports to longtime cloud executive Peter DeSantis.
Amazon calls it normal iteration — "we continually evolve our model lineup based on what customers need" — but outside reporting frames it as an admission that spreading resources across five specialized models hasn't produced one that competes with the frontier labs.
What this means for sellers
Nova and its successors are the models quietly running behind Rufus recommendations, AI-generated product imagery, and the other AI seller tools Amazon has been rolling out.
Consolidations like this usually mean slower feature updates for a few months while the new model gets built, then a bigger jump once it ships. If you're leaning on any Amazon AI tool still labeled "beta," don't be surprised if it shifts shape between now and re:Invent — and it's worth watching whether a stronger unified model eventually means sharper Rufus recommendations for your catalog.
IN PARTNERSHIP WITH SUPERFUEL AI
You cut your title to 75 characters. Which keywords did you stop ranking for?
Most titles were 150 to 200 characters.
Now they're 75.
Something got thrown out, and whether you did it in a hurry or Amazon's AI did it for you, nobody checked which of those words were actually earning.
Here's a real one. This plates listing was 200 characters. Amazon's AI rewrite of it dropped "Disposable Plastic Plates," 13,000 searches a month and the number one keyword for the product. It also cut "China-Like Look," a phrase that came straight out of customer reviews. Nobody invents that at a desk.
Here's what our agent kept instead:
GATHERLY 50 Plates Pack, White Pearl & Gold, 10" & 7"
Item Highlights: 25 Guests, Disposable Plastic, 25 Dinner + 25 Salad & Dessert, Belize Design, Wedding Party
Nothing that was earning got lost. The keywords that couldn't fit inside 75 characters moved into the 125 underneath, where shoppers still search them.
Superfuel's Title Agent writes both fields from your Search Query Performance, the titles of the competitors ranking above you, and your own customer reviews. Then it measures what the change did to your sales, stripping out price, ads, promos, seasonality and about 19 other factors, and tells you to keep it, watch it, or revert it.
"For one of our ASINs the title change added an increase of $6,500 in revenue, even with category searches down. And I don't question this metric, because Superfuel provides all the data to back it up." - Yvonne Schmidt, Senior Manager, Fellow
Run your own ASINs through Superfuel and see the lift for yourself.
HOT TOPIC
Alexa for Shopping Just Became Amazon's Main Discovery Engine — Here's What That Means for Rankings
Buried inside Amazon's Q2 2026 earnings call was a stat sellers should be watching closer than the stock price.
CEO Andy Jassy said more than 350 million customers have used Alexa for Shopping in the last 12 months, active users on it nearly doubled year over year, and interactions on the assistant are up over 500%.
🔊 Why it's bigger than a voice-shopping headline
Jassy also shared the number that should get sellers' attention: US customers who use Alexa for Shopping spend 40% more per order than those who don't, and shoppers who try Alexa+ join Prime at nearly 25% higher rates. Amazon is actively routing more of its highest-intent traffic through an AI assistant that reads structured product data instead of a search box.
📋 What this means for sellers
Alexa for Shopping leans on the same signals that AI answer engines and Amazon's own ranking algorithm already reward: complete, structured, factual listings. If your bullet points and A+ content are still written for a human skimming a page rather than an assistant parsing attributes, this is the quarter to fix that — treat your listing copy as a dataset first, marketing copy second.
TECH MARKETING
ChatGPT Just Became a Real Traffic Source. Marketplace Sellers Might Be Left Out.
For most of its life, ChatGPT treated brand mentions like a shout into a canyon — it would recommend your product to millions of people and give you nothing to show for it. That changed on May 7, 2026, when OpenAI quietly turned brand names inside ChatGPT's answers into live, clickable homepage links. Referral traffic moved the same day.
📈 The numbers
Referral traffic to brand websites jumped roughly 158% week-over-week in the days after the switch, and homepage referrals specifically surged more than 350%, according to Similarweb's clickstream data. Before the update, only about a quarter to a third of ChatGPT referrals landed on a brand's homepage; afterward, that jumped to roughly 60% and has held there.
🛍️ The catch for marketplace sellers
Here's the problem if you sell exclusively through Amazon: ChatGPT's new links point to homepages, and a seller whose entire commercial identity lives inside a marketplace listing doesn't have one.
Amazon has also spent the past year keeping AI crawlers out of its own catalog, so the platform hosting your products isn't feeding this discovery channel either.
The brands capturing this traffic are the ones with a legible, trusted homepage that exists outside the marketplace — exactly what a marketplace-only seller doesn't have.
What this means for sellers
If you don't have even a lightweight branded site outside Amazon, this is a good week to build one — not to replace your Amazon sales, but to give AI answer engines (and, not incidentally, Amazon's own ranking algorithm, which increasingly rewards converting external traffic) somewhere to send people.
A simple landing page with your brand story and a link to your storefront can start capturing a channel that's opening up for everyone else right now.
ECOMMERCE NEWS
Walmart Connect Finally Lets Advertisers Block Search Terms — Seven Years Later
After seven years of advertisers asking for it, Walmart Connect quietly shipped negative keywords for Sponsored Products in late July — letting brands exclude specific search terms from campaigns instead of eating wasted spend on queries that were never going to convert.
🛒 Why sellers have been asking since 2019
Until now, brands running automatic campaigns on Walmart had no way to stop their own branded terms, or irrelevant lookalike queries, from burning budget.
As Podean's Chris Sheldon put it, "brands have been buying on those brand-specific terms with Walmart only because they had no choice if they wanted to do automatic campaigns."
The feature is live now in Walmart Connect Ad Center and through partner platforms including Pacvue, Skai, Quartile and Teikametrics.
What this means for sellers
If you sell on Walmart Marketplace and run Sponsored Products, this is worth an afternoon this week.
Pull your search term report, flag your own brand terms and any obviously mismatched queries eating spend, and start excluding them — Amazon advertisers have had this lever since 2019, so Walmart sellers are just now catching up on efficiency gains competitors on other marketplaces already banked. If you sell across multiple marketplaces, treat this as a reminder to audit negative keyword lists everywhere, not just on Walmart.
