Welcome back friend!
AI is “Confidently” Wrong
Last week I had AI restock my FBA inventory from AWD. It told me every order went through. Two of them hadn't (One was 60 units short and one never shipped). I only found out by accident.
What got me wasn't the failure. Inventory hiccups happen. What got me was the report. It said done, all of them, same as always. 100%!
AND… The report read exactly the same whether the work happened or not.
Thanks, bro
When a person is slacking, you get signals. Work shows up late. Updates get vague. You ask a direct question and get a soft answer. Every instinct I've built as a CEO runs on those signals.
AI gives you none of them. It does most of the job well, and the part it botched sits right next to the rest, marked done (or it fails silently… so annoying!)
I've handed AI real pieces of my business this year. Inventory, ads, reporting. I'm not walking any of it back. But I'd started reading its reports like updates from an employee I trust, and they're not that. It sounded just as sure on the order that never shipped.
So I stopped waiting for signals and started going to get them. Here's what I do now:
- Check the thing, not the chat. The summary is a claim. What actually happened is in the system. Order placed? Open Seller Central and find it. Listing updated? Pull up the live page on your phone. Campaign launched? It's in the ads console or it isn't. Thirty seconds, on anything that touches money.
- Make it show receipts. Stop asking "did it work?" It always says yes. Tell it to go back into the system and pull the current state, fresh. If it can't show you the live result, the job isn't done (see the deep dive below for taking this further)
- Don't let the worker grade its own work. The AI that built the thing will be more likely to tell you the thing is good. Open a fresh chat with no history and have it check, or check it yourself. (see the deep dive below for taking this further)
- Rank by what a failure costs you. You can't check everything. You don't need to. A clunky product description can wait a week. An inventory order that fails costs you a stockout and big $$$.
- Treat silence as a failure. Automations can run out of API credits and just…stop. No error. No alert. If a report you expect doesn't show up, something is broken. Go look.
We turned all of this into a simple doc you can keep: the exact recheck prompt I can paste after tasks, a standing order to drop into your AI's permanent instructions so every job self-reports honestly, and a fresh-eyes prompt that makes a second AI audit the first. Good luck!
The IVS
Big thanks to Michael Shackelford for writing a very interesting tip this week. Michael is the founder of Genrupt.com, an AI content creation platform for Amazon and ecommerce sellers and a former 7 figure brand owner.
New York can fine you $5k for every Amazon image with an AI person?!?
Maybe.
And it definitely seems like they will if they get their way.
The law, which became enforceable on June 9th, requires that if a business creates an ad for a product or service and knows it includes a "synthetic performer," the business must conspicuously disclose that in the ad. And that extends to any type of image in commerce used to sell something. And there’s fines.
First violation: $1,000.
Repeat violations: $5,000.
So should Amazon sellers be worried?
This isn't legal advice, but here's how I think you should approach it.
Don't panic, but you should be prepared for the day when enforcement and detection of this becomes a real thing. It certainly seems like the canary in the coal mine. We’ll need to be disclosing AI content on every image (I.e. everyone will on every image 🤣)
So do this to ease your anxiety and gain clarity:
- Carve out at least 10 minutes this week to gather up your active listing images that include an AI person, even if only a finger. Get at least a few listings worth if you have a massive catalogue, just for this exercise.
- Now you need to apply a disclosure to all of these images. You can have a designer do this for you, or you can use a free tool I built for that here. This tool will let you apply the disclosure to up to 100 images at once. Since the law doesn't define exactly how to say it, you'll have some freedom to choose how to "conspicuously disclose" it. You could say "synthetic performer used", but I'd be more creative if I was you. Maybe something like, "An AI person was used to represent the thousands of satisfied real ones"... you get the idea.
- Once you apply this to this first batch of images, setup an AB test in Manage Your Experiments in Seller Central. Why? Because you want to know if your customers care or not, and my guess is that they won't care at all. As consumers we are used to this style of embellished advertising, but test to be sure. If anything, AI generated people are going to much more accurately represent someone using your product than the stock photo you photoshopped it onto five years ago.
- Run your tests until you are confident this doesn't impact your sales, and then just build it into your workflow. It only takes a few minutes to apply it to 100 images. And if for some reason you see that it does hurt your conversions, you get to make another decision, but I think that's unlikely.
Hopefully this gives everyone some direction until we see how the policy and enforcement develop in the coming months. And NO, there is no way to just block an entire state from seeing your ads on Amazon.
Did an extremely good-looking friend forward you this newsletter?
Get the next one, here!
The Deep Dive (Premium Content)
How to get AI to give you the right numbers (Every Time)
As we spoke about in the intro above, AI is often CONFIDENTLY WRONG about what it’s trying to say. It will tell you that everything is A-OK while the house is on fire. One of the most insidious ways it does this is with numbers. And there are a LOT of reasons why AI would think the numbers are correct (when they aren’t).
We built a simple 3-page guide that will force your AI to give you accurate numbers every time…
Want the full Deep Dive?
It goes out in the newsletter. Subscribe free, share it with one friend, and every Deep Dive unlocks.
The Week Was Grad Week in Costa Rica
When you live here and you’re choosing a school for your kids, you quickly realize you need to think ahead, or at least keep your options open, for where they might want to go after graduation. We are trying to keep it as wide as possible given the changing world we live in.
Some international schools have U.S. accreditation. Some offer IB. Others follow the Costa Rican MEP system, which matters if your child might later want to attend a local technical school or university. Some have all of them.
Then there’s the whole grade structure. In some streams, students graduate in 11th grade if they are heading into certain Costa Rican schools. So you end up with graduations from middle school, graduations into high school, Grade 6 to 7 ceremonies, 9th to 10th ceremonies, even though Grade 9 already was technically high school.
It’s weird. It’s confusing. But we show up, take the photos, clap at the right times, and eat the cake.
Happy summer vacation to everyone!
PS: I’m writing this from a hotel room in San Jose, in town for kids chess tournament. The local channels have 6 channels showing various Fast and the Furious movies in English. Similar last night too. I guess they think English viewers love Fast and Furious. Pura Vida.
“Disclosing” AI Imagery is Coming
Originally surfaced by SFTB friend Vanessa Hung, Amazon now makes you disclose whether your A+ Content and Brand Story images are AI-generated. For now that answer just sits as back-end metadata rather than a shopper-facing badge, but it is the future of where AI imagery and marketplaces are heading.
That groundwork lines up with the New York law that became enforceable June 9 (which we wrote about above). It’s a good idea to start getting ready to accurately disclose all the wonderful AI content you’ve created in the last couple of months.
Not Planning on Leaving Any Time Soon?
Some countries and airlines will not let you board or enter on a one-way ticket; they want proof you are leaving. If you do not actually have your next flight booked yet, the usual move is to buy a cheap throwaway ticket and waste the money. There is a cheaper workaround: services like OnwardTicket create a real airline reservation R your name, held in the actual booking systems and verifiable on the airline's website, for roughly $7 to $16. It auto-expires after a couple of days to two weeks, so you are renting the reservation, not buying the flight.
Once you get to Costa Rica, you may not want to leave, HA!
Perfect for digital nomads, frequent travelers, visa applicants or people taking last minute trips.
If a man knows not to which port he sails, no wind is favorable.
Seneca
BEER! Now that I have your attention… you probably would like the Facebook community.
This week in the Facebook group: Danan Coleman's AMA (He knows everything about reviews)
If you’re liking this newsletter so far, here's where we share more stuff like it:
- 🤓 Just getting started? Check out the Free Amazon FBA Course
- 👀 Need an ASIN Audit? The ASIN AUDIT OS
- 💸 Need less stuff on your plate? Top 10 Automations/VA Tasks
- 👔 Need an Accountant? Accounting Services
- 📧 Want some 1-on-1 time? Rob & Max
- 📣Interested in sponsoring the newsletter? E-mail [email protected]
To your success,
Rob & Max