writingMay 7, 2026

I Studied 15 AI Product Research Tools—None of Them Dares to Tell You If You Can Make Money

I studied 15 AI product research tools, and not one dares to tell sellers whether they can make money. The market is full of reporting tools, but decision-oriented tools are still nowhere. Independent developers don't carry big platforms' baggage and can fill that gap. Over the next two weeks I'm building a Chrome extension for intelligent Amazon review diagnostics.

Last week I came across something a UK ecommerce accountant wrote. One of his clients did £50,000 in sales over the weekend, and after he worked out platform fees, ads, shipping, and returns, every single order was losing money.

Not an isolated case.

I've never done cross-border ecommerce myself. I've watched friends step into those traps, watched people in that circle get rich from cross-border and then move on to something else—for three years. In early 2024 I almost entered the space with a friend, but in the end I chose to keep building AI products.

Recently I wanted to help a friend see whether today's AI product research tools can help them avoid at least some of the landmines.

I went through 15 players including Helium 10, Jungle Scout, SmartScout, Kalodata, FastMoss, Shulex, EverBee, and Alura. After all that, I was left with three words: not good enough.

Helium 10's Trustpilot rating has dropped from 3.2 to 2.7. One user wrote bluntly: "These AI tools aren't much smarter than a 5-year-old."

Amazon's own AI writing tool is even worse. It wrote the LLM model identifier "mistral-3-1" directly into a seller's listing copy. The seller vented in Seller Central: "another half-baked C-suite KPI that leaves sellers to clean up after the bonus is paid."

Independent tests of sales estimate accuracy show Helium 10's exact-hit rate at 38.2% and Jungle Scout at 29.5%. That means the numbers a seller sees for $200 a month are wrong 60–70% of the time.

But inaccurate tools aren't the biggest problem.

The biggest problem is that the tools never calculate whether you can actually make money.

One Amazon seller posted a comparison in Seller Central:

"For April, May, and June 2025, Helium 10 shows our total expected net profit around $12,121, but our actual Amazon payouts only totaled $8,241."

Over three months, the tool showed $12,121 in net profit. The actual payout was $8,241. A gap of $3,880.

That's not a bug. That's what happens when a tool shows you a dashboard, and nobody shows you the payout. It leaves out platform fees, advertising ROAS, return rate, and shipping fluctuations. If each of those is off by 5%, together they eat the entire margin.

The line that gets repeated in cross-border circles is: "smart product research tools... just search bars duct-taped to ChatGPT." It sounds impressive, but what it actually means is search boxes taped to ChatGPT with duct tape.

Why can't big platforms build it?

It's not that they can't. It's that they don't dare.

First, their business model is selling data subscriptions. Giving you a conclusion would mean digging their own grave—what reason would there be to keep charging $99 a month?

Second, legal won't allow it. If AI directly tells a seller "do this product or not," and the seller loses money, the seller will turn around and sue. Big platforms can't take on that responsibility.

Third, cross-platform aggregation would undercut their own affiliate business. Helium 10 can earn 30%+ commissions from Amazon affiliates—why would it send traffic to TikTok Shop, Temu, or independent sites?

So what you see will always be a dashboard. Never a decision.

Independent developers don't carry that baggage.

And this track has already been validated by a string of cases:

EverBee: a Chrome extension layered on top of the Etsy seller backend, $7.99/month, 900K+ users
EasyGen: a Chrome extension layered on the LinkedIn editor, $5/month, 6 people reaching $41K MRR
PhotoAI: 1 person, 1 tool, $39/month, $138K MRR

The shape is almost identical: layer on a platform users are already on, do one thing, give one direct judgment.

Where's the gap in cross-border? Shulex Copilot layers GPT-4 on top of the Amazon backend, but its pricing is polarized: the free tier is rough, and the professional tier jumps straight to $179/month. The $9.99/month range in between—the price small sellers can actually afford—is empty.

So here's what I want to build next:

An intelligent Amazon review diagnosis Chrome extension.

A seller opens a competitor's product page, and the extension automatically classifies thousands of reviews into six dimensions: quality defects, use cases, complaints, purchase motivations, packaging and logistics, and returns and after-sales. It gives a risk score that tells you whether the product has structural problems.

It won't tell you how to sell it. It tells you whether this product is suitable to enter.

That maps to a seller's post-mortem that stuck with me. He stocked 2,000 baby sleep sacks and lost money before he saw clearly: the overall rating was 4.1, but the negative reviews concentrated on "the material is too brittle and fell apart after two weeks." He had seen it back then, but he hadn't recognized it. The tool should have recognized it for him.

That UK accountant said: the worst thing isn't losing money. It's not knowing where you lost it.

Tools should answer that question. I'm building it over the next two weeks. First, a version you can install, then iterate based on data.

Sources