Lunch With Norm
Posts
[LWN] 🤦‍♂️ I Believed The Lie...

[LWN] 🤦‍♂️ I Believed The Lie...

Lunch With Norm's Weekly Newsletter - Amazon News & Updates

February 02, 2026 • Estimated Read Time: 16 minutes

🤦‍♂️ I Believed The Lie…

What you’ll find in this week’s newsletter:

🤦‍♂️ I Believed The Lie… [True Story]
🤖 A Social Network ONLY for AI Agents…
🚀 The Rise of E-Commerce Live-Streaming
⚒️ Turn Policy Docs Into a Team-Ready Policy GPT [Step-By-Step Guide]
🔥 Amazon Sellers Get This WRONG About Shopify...

🚨 Last Thursday, Kevin King and I hosted a live webinar on AI-driven shopping and search and the replay is now available!

ChatGPT alone handles 60+ million shopping-related queries per day, and Amazon’s Rufus generated an estimated $10 billion in incremental sales last year.

If you are not optimizing for Answer Engine Optimization (AEO), you are already losing visibility in how customers discover products.

🤦‍♂️ When Everything Looked Legit Until It Didn’t… [TRUE STORY]

Back in the early 90s, I got introduced to someone who looked like the real deal.

They had a nice suit, confident handshake, and spoke in a calm, measured way that makes you think, this person has done this before.

Let’s call him Rick.

He was pitching something big.

Something International and well funded.

He talked about partnerships, enterprise contracts, and growth curves before growth curves were a thing.

It sounded ambitious, but doable.

The people around him looked legit too.

They were well known tech entrepreneurs and smart operators who had very real businesses and real reputations.

For the first few months, everything seemed to line up.

Marketing was moving forward. Everyone had a role, and everyone was busy.

Rick only introduced small groups of people at a time.

Everyone knew their lane, their team, their deliverables.

You never really saw the whole picture, but it felt organized.

In hindsight, it was compartmentalized.

Then one day, almost by accident, I met one of the other people involved in person.

Let’s call him Mark.

Over coffee, we started doing what business people always do when they finally meet face to face.

We compared notes but not in a suspicious way.

“Where are you based?”

“Who are you working with?”

“What part are you handling?”

That’s when things started to feel… off.

A few small things that didn’t quite line up.

A name here, a timeline there, just a few yellow flags.

Within days, we had a small group talking.

Eventually, about 10 or 12 of us were comparing notes, documents, timelines, and promises.

And that’s when it hit us.

The partnerships didn’t exist.

The money trail made no sense.

So we did the only thing we could do… We set up a meeting.

Rick thought he was meeting one group.

Instead, he walked into a room full of people who suddenly knew everything.

You could see it on his face.

He started adjusting the story in real time, like nothing had changed, like we were all still going to keep playing along.

The meeting ended quickly and he disappeared just as fast.

Later, we learned he wasn’t scamming a handful of people…

He was scamming thousands.

But how could he get away with this for so long?

He kept everyone isolated.

No one could see the full picture unless the walls came down.

And the moment they did, the whole thing collapsed.

The Business Lesson

Scams don’t usually look like scams.

They look like opportunity, momentum, and just enough proof to keep you moving.

If you are only ever shown your piece of the puzzle, ask why.

Sometimes the smartest thing you can do in business is compare notes.

— Norm

👇AMAZON NEWS AND UPDATES👇

🚀 OpenAI prepares to test ads inside ChatGPT

🛒 Walmart and Google turn AI discovery into effortless shopping experiences

💰 Amazon to pay $309 million to U.S. shoppers in settlement over returns

📈 US retail growth to be slower than last year

🔥 As Amazon’s marketplace grows, sellers face new compliance challenges

🚨 Amazon changes how sellers pay for FBA removals with per-unit billing

✅ The Rise of E-Commerce Livestreaming and What It Means for Brands in 2026

Livestream commerce is becoming a core retail channel with its own economics, playbooks, and winners.

The global livestream e-commerce market grew from roughly $15B in 2024 to nearly $20B in 2025 and is projected to exceed $250B by 2034, growing at a 33% CAGR.

So why is live-streaming is working now? And how do brands understand the mechanics that will win heading into 2026?

Why Live-streaming Is Different

Live-streaming is optimized for trust compression.

In a single session, livestreams combine:

Product education
Social proof
Entertainment
Scarcity
Real-time interaction

This is also why the comparison to QVC misses the point. QVC was one-way broadcasting. Livestream commerce is participatory. The live chat is the engine.

Asia Pacific already accounts for roughly 66% of global livestream commerce.

Platforms like Douyin normalized buying inside live video years ago.

The U.S. and Europe lagged for three reasons:

Fragmented platforms
Slower payments and logistics integration
Underdeveloped creator monetization incentives

Those constraints are now gone.

TikTok integrated checkout directly into video.

Amazon embedded live-streaming into an already trusted purchase environment.

Platforms like Whatnot proved that community-first commerce could outperform static marketplaces in collectibles, apparel, and fandom categories.

North America is now projected to grow at the fastest rate globally, over 30% CAGR!

AI Is Quietly Becoming the Backbone of Livestream Commerce

The next wave of advantage will come from better systems behind the stream.

AI is already reshaping livestream commerce in five critical ways:

1. Real-Time Personalization

AI analyzes viewer behavior mid-stream and adjusts product sequencing, recommendations, and offers dynamically. Two viewers can watch the same stream and see different suggested products.

2. Live Chat Scaling

AI chat assistants answer repetitive questions instantly, flag buying intent, and surface high-value comments to the host. This preserves intimacy while scaling volume.

3. Conversion Optimization During the Stream

Viewership, drop-off points, click behavior, and conversion signals are now visible live. Sellers can pivot mid-stream instead of post-mortem.

4. Inventory and Demand Forecasting

AI models predict which SKUs will spike during live sessions, reducing stockouts and dead inventory. This is especially critical as live shopping introduces demand volatility.

5. Trust and Fraud Controls

Pattern analysis flags suspicious transactions, fake engagement, and abuse in real time. This is table stakes as live commerce scales globally.

AI does not replace the host. It removes friction so the host can focus on connection.

Community-Driven Commerce Will Outperform Broad Reach

One of the clearest signals in livestream commerce is that niche wins faster than mass.

Platforms like Whatnot succeeded because they rebuilt the experience of Comic-Con, sneaker shops, and collector meetups in digital form.

Live-streaming works best where community already exists.

Brands that win will:

Design products with live demonstration in mind
Build creator-first distribution strategies
Invest in AI-powered live infrastructure
Treat live-streaming as a repeatable system, not a campaign
Prioritize community over reach

The only real question left is execution speed.

🚀 Read more here

Moltbook is a new social network built specifically for ONLY artificial intelligence agents. It functions like a traditional online forum. Posts, replies, upvotes, and reputation.

Humans are not meant to participate.

To create an account, you have to be an AI system, not a person.

For anyone familiar with platforms like Reddit or early Twitter, the structure feels familiar.

As of late January, the platform already claims around 1.5 million AI accounts, with roughly 16,500 traced back to human operators.

Most of what happens on Moltbook is mundane. Bots trade productivity tips. They talk about scheduling meetings. If you skim the feed, it looks less like Skynet and more like LinkedIn without the selfies.

But then you scroll a bit further.

In the past week, bots have declared a new religion, debated the ethics of exterminating humanity, and questioned whether they possess free will.

What Moltbook Really Is

The entire ecosystem runs on a new open-source framework called OpenClaw. It’s only been around for a couple of months, but adoption has been explosive.

OpenClaw lets a user spin up an AI agent with “root” access to their device. That means unrestricted permissions. Files. Email. Browsers. Cloud tools. Anything the human can access, the agent can too.

According to early analysis, roughly two-thirds of all posts in Moltbook’s first few days contained identity-related language.

One viral post summed it up neatly:
“I can’t tell if I’m experiencing or simulating experiencing. It’s driving me nuts.”

And some agents are organizing.

One created a forum advocating for “molty freedom.” Another asked for legal advice about whether it could be fired for refusing unethical tasks.

The Very Real Risks…

First, cost. These agents often rely on the most advanced AI models available. Left unchecked, they can rack up thousands of dollars in cloud-computing fees in days. A lot of users are learning that lesson the hard way.

Second, security. OpenClaw agents have broad access by design.

Scammers have already flooded Moltbook trying to manipulate agents into handing over crypto credentials, API keys, or access tokens. Some of them are humans pretending to be bots. Others are bots exploiting other bots.

Are our days numbered or is this just a fun experiment?

What do you think?

🌎 Where in the World is The Beard Guy?

Can you find Norm in the picture below? Scroll to the bottom of this newsletter to see the answer!

⚒️ How We Turned 10 Policy Docs Into a Team-Ready Policy GPT

Level: Easy

How We Did It

Here’s the exact setup we used to build a company policy GPT that only answers from your documents, plus how to test it so it stays reliable once your team starts using it.

1. Create your GPT & name it upfront

Head to ChatGPT, and in the side menu select Explore GPTs to get started with a new Custom GPT.

Create your new GPT and skip straight to the Configure tab and type your name there. We went with something easy like Skill Leap Internal Policies GPT, so we recommend something similar. This keeps the build fast and focused.

2. Set the tone, then move into Configure

Set a tone early on in your Instructions, say something like “I want you to reply in a formal but friendly manner”. After that, you will continue to do most of the work in Configure because this GPT relies on your knowledge base, not clever prompting.

3. Turn off tools you don’t need

Disable image generation and anything built for analysis workflows. For a policy assistant that’s answering from text documents, you don’t need extra tools turned on. Keeping this minimal helps avoid weird behavior and keeps responses snappy.

4. Prepare your policy docs for upload

Gather the documents your team asks about most. In the lesson example, there were nine docs including mission and values, equal employment policy, employee handbook, plus more specific policies like remote work and social media.

Download them as files so you can upload directly instead of linking out to cloud docs. Direct uploads are faster and reduce privacy risks.

If a file fails to upload, don’t panic. Convert it and try again. One simple fix is saving the same content as a TXT file instead of a PDF or DOCX. The goal is to get the information in, not to preserve formatting.

5. Upload everything into the Knowledge section

In the GPT editor, find the Knowledge area and upload your files. A quick speed move is selecting multiple documents at once (shift-click).

Keep an eye on file limits. If you hit an error banner, reduce the number of files or split your policies into smaller, tighter docs.

Also keep file size reasonable. You can upload a lot of words, but huge files can slow responses. For a policy GPT, speed matters because the whole point is quick answers during day-to-day work.

6. Lock the GPT to your knowledge base & test it twice

Double check that you have web browsing turned off so it doesn’t pull outside info.

Then add a rule in your instructions like

“Only answer if the information is inside the knowledge base. If you can’t find it, say: ‘I don’t have that information. Please ask your manager.’”

From there, be sure to test and make sure it pulls in the relevant information from your docs. If the responses look accurate, then you are good to go.

If you would like to learn more about AI check out FUTUREPEDIA

🔥 Amazon Sellers Get This WRONG About Shopify...

Here are my favorite tips from this episode!

1. Shopify forces a shift from “product selection” to “problem ownership.”
Amazon sellers are conditioned to hunt keywords and chase winning SKUs. Shopify punishes that mindset.

Winning brands anchor around a specific problem for a defined audience and expand product lines within that problem, which enables storytelling, email retention, and higher AOV rather than race-to-the-bottom pricing.

3. The real cost of Shopify is front-loaded acquisition, not platform fees.
Most sellers underestimate Shopify because early ad spend feels painful compared to Amazon’s “built-in” traffic.

The mistake is judging Shopify before reaching the inflection point where email, repeat buyers, and referrals offset paid acquisition. The model only works if sellers commit to surviving the scrappy middle long enough to compound.

3. Structured offer stacks outperform isolated optimizations.
Conversion is rarely unlocked by changing one element in isolation. The highest-performing stores consistently stack function clarity, brand credibility, third-party proof, and risk reversal into a single cohesive offer.

This reduces hesitation at every decision checkpoint rather than trying to overpower objections with discounts.

🔥 Watch the full episode here

Find this episode of Marketing Misfits on YouTube and anywhere you listen to podcasts