604: Amazon In 2025: What It Takes To Sell At Scale (And How Ai Changes The Game) With Bernie Thompson

Podcast: Download (Duration: 49:30 — 56.9MB)

Selling on Amazon at scale in 2026 requires world-class content marketing, disciplined tariff-hedged sourcing, per-SKU channel strategy, and deep AI automation, in that order. On this episode of the My Wife Quit Her Job podcast, I sat down with Bernie Thompson, the founder of Plugable Technologies (a leading US-based USB, docking-station, and adapter brand that competes head-on with Anker and Ugreen). Bernie has been operating at eight figures on Amazon for over a decade, and he walked through the exact playbook that keeps a real scale seller alive when fees are up, tariffs are volatile, and buyers are increasingly starting their product research inside ChatGPT and Claude.

The single biggest shift Bernie sees is that AI has moved discovery from an “infinite shelf” (search results with many options) to a curated shelf of one to three products per query. Bernie’s answer is to double down on the content depth that has fed Plugable’s search rankings for years, because the same “how does this work with that” articles, videos, and Q&As that won Google now feed the LLMs.

Below is the full playbook: how to optimize a brand for LLM citations, why per-SKU channel routing beats “Amazon first, DTC later,” how Plugable is coping with 55% tariffs on unexempted electronics, the exact AI-automation stack Bernie’s team runs, and how to think about the “everyone becomes a manager” future of ecommerce work.

Get My Free Mini Course On How To Start A Successful Ecommerce Store

If you are interested in starting an ecommerce business, I put together a comprehensive package of resources that will help you launch your own online store from complete scratch. Be sure to grab it before you leave!

Click Here To Download

Key takeaways

• AI search is compressing the results page from many blue links to 1 to 3 product recommendations. Winning brands are the ones LLMs already “know” through years of deep, connective content.
• The content strategy that wins LLMs is the same one that wins Google: long, honest “does this work with that” articles, videos, and Q&As, published across written and video formats. Plugable just crossed 50,000 YouTube subscribers on exactly this approach.
• Concrete AI-SEO wins that work today: cite verifiable facts about your business (“we have sold over a million handkerchiefs” got Steve’s site picked up by AI as “the highest volume handkerchief seller”), and load a vector database of your product descriptions to power on-site AI search.
• Plugable routes each SKU to a specific sales channel (Amazon, B&H, distributors, Shopify) using a spreadsheet-driven buy-box on their site, and moves SKUs off Amazon when they cannot rank against consolidated Chinese brands.
• Anker, Ugreen, and Baseus have consolidated the Chinese USB brand landscape. Amazon feels “less competitive” only in the sense that the churn of 5-letter no-name brands has slowed; the surviving Chinese brands are tougher operators than what came before.
• 2025 tariffs sit at 55% on cables and chargers out of China, and 0% on laptops, phones, and docking stations under the “Tim Cook” exemption. Plugable is discontinuing about half of its exposed SKUs, absorbing 55% on the other half, and betting the docking-station exemption holds.
• US Q2 2025 tariff revenue hit ~$64B (up from $17B a year earlier), a ~4x jump that is being absorbed by importers so far. Price increases have been unusually delayed, mostly because of political and media pressure on big retailers.
• Bernie’s most impactful AI use is code generation. A script that used to take 4 to 8 hours now takes ~30 minutes in Cursor. His second-most impactful use is company-visible custom GPTs loaded with SOPs, matched one-to-one with roles.

What does selling on Amazon at scale actually look like in 2026?

Selling on Amazon at scale in 2026 means running a professional operation across sourcing, tariff strategy, content, and AI automation at once, with no gaps. The “post a listing and let Amazon drive traffic” era is over even for large sellers. Bernie compares Amazon in 2026 to the pre-Amazon retail world of gatekeepers, only now the gatekeeper is the LLM that decides which one to three products get named in an AI answer.

For a brand like Plugable, that translates into per-SKU channel discipline. Every product is tagged for a target sales channel (Amazon, B&H Photo, IT distributors, or Shopify) inside a spreadsheet, then Python pushes that metadata to Shopify, and Liquid templates render a custom buy-box on the product page that routes buyers to whichever channel is right for that specific SKU.

The result is that Plugable can pull a SKU off Amazon when Chinese-brand consolidation makes it unrankable, and instead push all traffic for that item to a distributor or their own Shopify. It also means new products launch differently. If it is a commodity like a USB cable, Plugable does not even target Amazon success anymore, because Anker, Ugreen, and Baseus own that shelf.

How is AI search changing product discovery for Amazon sellers?

AI search is collapsing product discovery from a page of 30 to 60 blue links into a conversational answer that names one to three products. That is a fundamentally different game than SEO or Amazon organic ranking, because the AI is choosing on the buyer’s behalf instead of surfacing options.

The good news for brands with genuine expertise is that LLMs preferentially cite sources that are already deep, well-organized, and connective. Plugable’s whole content strategy for the last decade has been “does this work with that,” which is exactly the compatibility-shaped content LLMs need to answer real user questions. That head start compounds now.

The bad news is that no one has a reliable playbook for “influencing the LLMs” yet, and pretending otherwise is dishonest. What works is putting out more of the same great content, in more formats, so the training runs and retrieval indexes have more chances to pick you up. Bernie’s stated ambition is to publish much more content in 2026 than in 2025, in both video and written form.

How to optimize a brand for AI search citations in 2026

The single highest-return AI-SEO tactic today is publishing verifiable, quantified facts about your brand on your own site and having them cited back by the LLMs. I did this on my Bumblebee Linens site by writing “we have sold over a million handkerchiefs” as a stat on multiple pages. Within weeks, AI models were describing me as the highest volume handkerchief seller on the internet, and that citation is now driving referral traffic.

The second highest-return move is loading a vector database (Pinecone, Weaviate, or a self-hosted equivalent) with your full product catalog and using it to power on-site AI search. My Bumblebee Linens implementation replaced the standard Shopify search with semantic matching, and the sales lift was immediate. This is one of the pieces that vibe coding, per my conversation with Toni in episode 607, can now put together in a weekend.

The third move is multimodal content. LLMs now tokenize video, extract visible context (what you are holding, what you are plugging in), and index both the words and the scene. Video used to be dark content to search engines. Now it is fully indexed by AI, so the 50,000 YouTube subscribers Plugable has built are a real training-data asset, not just a distribution channel.

Is Amazon really less competitive for scale sellers in 2026?

Amazon is not less competitive at scale in 2026, even though the total seller count has dropped. The apparent slowdown in competition is really a consolidation among Chinese brands, and the survivors are tougher operators than the churning 5-letter brand names of 2020.

In the USB and charger category specifically, Anker, Ugreen, and Baseus are now the dominant Chinese brands. Five years ago, search results on Amazon rotated through an endless stream of nameless 5-letter brands that would appear, undercut on price, and disappear. Today those same searches consistently return the same three or four consolidated brands, plus a shrinking set of survivors.

The lesson for Western sellers is that “fewer sellers” reads as opportunity only if you can compete with the brands that survived. Bernie’s read is that this consolidation reflects the brutal internal Chinese market: any brand that survives Chinese domestic competition arrives on Amazon as an unusually strong operator. Assuming the field is easier because it looks thinner is a mistake.

How are Amazon sellers handling 2026 tariffs at scale?

The 2026 tariff landscape is uneven and volatile. Plugable is paying 55% on cables and chargers still sourced from China, and effectively 0% on laptops, docking stations, and adapters under the electronics exemption informally known as the “Tim Cook exemption” (Bernie’s phrase for the phones-and-computers carve-out negotiated after the initial April 2025 announcements).

Plugable’s response is a three-part playbook. Discontinue about half of the 55%-exposed SKUs, keep the other half on the shelf and absorb the tariff (mostly because the SKUs matter for the coherence of the product line), and lean hard on the SKUs that combine the exemption with eight years of “get out of China” sourcing work into Vietnam, Malaysia, and Thailand.

The bigger risk is planning. Bernie has to place multi-million-dollar POs today for goods that arrive four months from now, which is two full tariff-policy change cycles under the current cadence. The uncertainty is real, and the whole ecommerce industry has been running Ford-tough on price because nobody wants to raise unless they have to.

Why prices have not yet risen the way tariff math implies

US Q2 2025 tariff revenue hit approximately $64 billion, up from about $17 billion in the same quarter of 2024. That is roughly a four-fold increase, and the math says prices should be up significantly by now. They mostly are not.

The reason is a mix of political pressure on large retailers (Walmart, Apple, Amazon are all being publicly named), uncertainty about whether specific rates will stick, and importers absorbing hits rather than telegraphing weakness to competitors. Electronics as a category has actually seen slight price declines in 2025, though this is not sustainable.

Bernie’s prediction is that when the dam breaks, prices will move up fast rather than gradually. That is worth planning for in Q4 2026 inventory and pricing.

Loopholes and gray zones on tariffs

The de minimis rule change in 2025 closed a major tariff-avoidance loophole (bulk shipments to Canada or Mexico, then reshipped in under-$800 parcels to the US). That was a genuine win for US sellers who compete against Chinese brands.

The remaining tariff-avoidance vector is misdeclaration on customs paperwork: wrong HS code, understated declared value, or (legally) shifted profit reported by the manufacturer of record. Chinese sellers can legally declare a lower value than a US importer of the same good, because they only have to report input costs (not the retail invoice), and they can strip out their own profit. Once tariffs are at 55%, a 20-30% understatement is the whole ball game.

Bernie’s warning is that a set of US-based sellers he sees online are using gray strategies to lower declared cost of goods, and CBP has audit authority to reach back years. When those audits land, the resulting bills routinely destroy the businesses. Chinese sellers have far less at stake, because a company can simply reincorporate and reappear.

The AI stack a scale Amazon seller actually runs in 2026

Plugable’s AI stack is a working example of what a scale seller looks like when AI is threaded through operations rather than bolted on. It has four layers.

Layer 1: Code generation with Cursor

Bernie’s single most impactful AI use is code generation in Cursor, an AI-native code editor built on VS Code. Scripts that would have taken 4 to 8 hours to write from scratch now take about 30 minutes: 5 minutes to first working code, 25 minutes to fix the two or three problems that show up. That compression has changed what is worth automating.

Layer 2: Company-visible custom GPTs

The whole executive team is on OpenAI’s paid business plan, which gives them company-visible GPTs (one per role) loaded with the relevant SOPs and internal knowledge. Any team member can ask SOP-level or strategy-level questions and get answers grounded in the actual company documents rather than the open internet.

Layer 3: Apps Script for Google Workspace automation

Plugable is a Google Workspace shop, and Apps Script is the go-to for non-developer team members automating Gmail and Sheets workflows. Apps Script is a good “safe” area for AI-generated code because each script is a small, standalone unit; if the AI produces something broken, the blast radius is that one script.

Common Plugable use: an Apps Script that calls the Keepa API (a widely used Amazon price-and-rank history tracker), pulls the data for a list of ASINs, and fills a spreadsheet. That is a task the finance and marketing teams now build for themselves.

Layer 4: SQL + BigQuery + Connected Sheets

For data pipelines, Plugable pulls raw data (from APIs, Amazon, or their own systems) into BigQuery, then uses Google’s Connected Sheets feature to expose that data directly in spreadsheets via SQL queries. LLMs are unusually good at writing SQL because the target language is highly constrained, so non-developer team members can now build multi-join queries with AI assistance.

How to think about AI code generation if you are not a developer

The most important mental shift for non-developer sellers is that vibe coding is a communication skill, not a technical skill. Tools like Replit, Lovable, Cursor, Kiro (Amazon’s new AI IDE, launched in 2025 out of AWS), and Windsurf let anyone describe what they want in plain English and get working code back. The bottleneck is how clearly you can describe what you want.

The tell that you are the bottleneck: if the LLM is not producing what you want, the problem is almost always the prompt, not the model. The teams shipping the best AI-assisted code are writing 10-page SOPs (or in Cursor’s case, “cursor rules”) that describe the architecture, database, logging, and conventions the code should follow. Bernie’s current project has roughly 1,000 lines of plain-English architectural rules feeding Cursor every time it writes code.

Bernie’s frame: being a good AI code prompter is the same skill as being a good boss. You have to describe what you want in enough detail that the person (or model) on the other end can actually execute without endless back-and-forth. Both roles reward conceptual clarity and a wide vocabulary of the things that are possible.

The “everyone becomes a manager” future of ecommerce work

Bernie’s mental model for the next 2 to 3 years of ecommerce work is that every operator becomes a manager of AI agents. You will have a team reporting to you, and much of that team will be AI, not people. That is a very different world from the one most sellers have trained for.

The uncomfortable implication is that individual-contributor experts (the people who take pride in doing the craft themselves) are the worst positioned for the shift. Their instinct is to reject AI output as slop because it does not match their personal quality bar. That is the same trap first-time managers fall into, and it stops scale in both cases.

The generalized answer is the same as with human employees: you have to delegate anyway, accept 80% output on the first pass, and iterate to your quality bar over time. Refusing to delegate is the only way to guarantee no scale, whether the workforce is human or machine.

What college major or skill set is future-proof for the AI ecommerce era?

There is no single future-proof major, and Bernie is honest that he does not know exactly what to tell his kids. What is clear is that the barbell is widening: deep technical experts in the tightest AI-scarce specialties (foundational AI research) are commanding hundreds of millions of dollars, while many talented general-purpose programmers are out of work.

The bet Bernie is leaning toward for his kids is broad-based education, exactly what a traditional four-year liberal-arts degree was originally designed to produce. A wide vocabulary of concepts is what lets you prompt LLMs effectively, and prompting is the primary skill layer above whatever the LLMs themselves are doing.

The complementary bet is on people skills, relationships, and creative organization. Once everyone can generate an app in a weekend, the differentiators become the ability to organize, delegate, and connect. That is what “future-proof” looks like in a world where the technical layer commodifies fast.

Frequently asked questions

What does it take to sell on Amazon at scale in 2026?

Selling on Amazon at scale in 2026 requires professional operations across sourcing (with active tariff hedging), content marketing (which now feeds LLM citations as well as Amazon SEO), per-SKU channel strategy, and deep AI automation across code, data, and internal tools. The old “list a product and let Amazon drive traffic” model is over; scale sellers now run structured operations more like traditional retailers than solo entrepreneurs.

How is AI changing Amazon product discovery?

AI is compressing discovery from a page of options into a conversational answer that recommends 1 to 3 products. Brands with years of deep, connective, “how does this work with that” content are best positioned to be cited, because LLMs preferentially reference sources they recognize as consistently authoritative in a category.

What is the fastest AI-SEO win for an ecommerce brand?

Publishing quantified, verifiable facts about your brand on your own site (revenue milestones, units sold, years in business, product firsts) and repeating them in structured form. Steve’s site got picked up by AI as “the highest volume handkerchief seller on the internet” within weeks of publishing the stat “we have sold over a million handkerchiefs.” Vector-database on-site search is the second big win.

Are 2025 tariffs actually hitting Amazon sellers?

Yes, unevenly. Rates range from 55% on cables and chargers to 0% on laptops, phones, and docking stations under the electronics exemption. US Q2 2025 tariff revenue was about $64 billion, roughly 4x the same quarter a year earlier. Prices have not risen proportionally yet, because importers are absorbing the hit under political and competitive pressure, but that is unlikely to hold indefinitely.

What AI tools does a scale Amazon seller actually use?

Cursor (AI-native code editor) for internal automation, OpenAI paid plan with role-matched custom GPTs loaded with SOPs, Google Apps Script for lightweight Gmail and Sheets automation, and SQL against BigQuery for data pipelines. Newer entrants worth watching include Amazon’s Kiro IDE (launched 2025) and Windsurf.

Should ecommerce sellers still use Buy with Prime on their Shopify site?

It depends on your channel strategy. If Amazon is your primary channel for a given SKU, Buy with Prime can help you capture buyers who need the Prime badge to convert. If you are trying to shift traffic away from Amazon on that SKU (to own the customer data), route the buy box to your own Shopify checkout or a distributor instead. Plugable does this per-SKU using a spreadsheet-driven configuration.

Is coding still a good career for the AI era?

Deep specialists in AI-scarce niches (foundational AI research, hardcore ML infrastructure) are getting record compensation, while general-purpose programming is under pressure. The most durable skill set is broad conceptual literacy plus the ability to prompt, delegate, and manage AI-assisted work. That looks more like management or a liberal-arts foundation than a traditional CS degree.

How should ecommerce sellers price and plan inventory under tariff uncertainty?

Assume prices will eventually catch up to declared tariff rates, and plan for a fast rather than gradual adjustment when they do. Diversify sourcing away from China where the math works, pull SKUs that are structurally unprofitable at the new rates, and be conservative with 6-month-plus purchase orders written under uncertain policy. Absorbing the tariff is a defensible short-term move only if your competitors are doing the same.

Ready To Get Serious About Starting An Online Business?

If you are really considering starting your own online business, then you have to check out my free mini course on How To Create A Niche Online Store In 5 Easy Steps.

In this 6 day mini course, I reveal the steps that my wife and I took to earn 100 thousand dollars in the span of just a year. Best of all, it's absolutely free!