AI assistants have become many sellers' first research tool. Ask ChatGPT or Claude "what should I sell on Amazon?" and you get a confident, well-written answer in seconds. Some of it is useful. Some of it can cost you money.
Here is where general AI assistants genuinely help with product research, where they fall short, and how to give them the data they are missing.
What AI assistants do well
- Brainstorming. Give it your constraints (market, budget, no fragile items) and ask for 30 product ideas in a niche. It is fast and surprisingly creative.
- Reading reviews at scale. Paste the 1–3 star reviews of a competitor and ask for the recurring complaints. You get a list of improvements to make.
- Explaining concepts. VAT on Amazon UK, how FBA size tiers work, what an HS code is: assistants explain clearly (still check the official source for the exact rules).
- Writing. Supplier emails, listing drafts, ad scripts and FAQ answers.
- Thinking through a plan. "What could go wrong with launching this product in Germany?" is a good prompt.
Where they go wrong
1. They don't see live market data
Out of the box, an assistant can't see Amazon sales-rank history, current keyword search volumes or today's prices. Web search helps a bit, but it finds articles about products, not the sales data itself.
2. They can produce numbers that sound right but aren't
Language models predict plausible text. Ask for "the monthly search volume for bamboo drawer organizers" and you may get a precise-looking number with no source. Treat any number without a source as unverified.
3. "Trending products" lists are often stale
Lists of trending products in the model's training data or in popular articles reflect what was trending, often when many sellers had already jumped in.
4. The math needs to be exact
Margins depend on exact fees, VAT rules and duty rates. An assistant can describe the formula, but small errors in fee assumptions change a go/no-go decision. Use a calculator built on the actual fee tables.
How to get better research out of AI
- Ask for sources, every time. "Show the source and date for every number" filters out invented figures.
- Separate ideas from evidence. Use the assistant for ideas and analysis; get demand, competition and margin numbers from data tools.
- Give it your real data. Paste exports, review text and your cost sheet, and ask for analysis of that.
- Connect it to data with MCP. The Model Context Protocol lets assistants like Claude and ChatGPT call external tools. With the right connection, your assistant can pull real sales history and search demand instead of guessing.
What MCP changes
With MCP, an AI assistant can use a specialised service as a tool during the conversation. For product research, that means you can ask in plain language ("find kitchen products under $30 for Amazon US with low competition") and the assistant fetches measured data, then explains it.
Ecomtoro works this way: you can connect Ecomtoro to Claude, ChatGPT, Claude Code, Cursor or Codex. Your assistant can then research and validate products, calculate margins and check your watchlist, with sources attached to every number. Anything that spends credits asks you first.
A sensible workflow
| Step | Use an AI assistant for | Use data for |
|---|---|---|
| Ideas | Brainstorming within your constraints | – |
| Shortlist | Explaining trade-offs | Demand, trend, competition scores |
| Validation | Summarising reviews and risks | Sales history, search volume, margin |
| Launch | Listings, ad scripts, supplier emails | Break-even ROAS, price checks |
| Tracking | Explaining what changed | Alerts on price, demand and competition |
AI makes research faster. Data makes it right. The best results come from using both together.
Frequently asked questions
Can ChatGPT tell me what product to sell?
It can suggest ideas and explain trade-offs, but without access to market data it can't confirm demand, competition or margins. Use its ideas as a starting list and validate them with real data.
Why does ChatGPT give different numbers for the same product?
Without a data source, a language model generates plausible numbers rather than looking them up. Always ask for the source and date of a number, and treat unsourced figures as unverified.
What is MCP and why does it matter for ecommerce?
The Model Context Protocol is an open standard that lets AI assistants use external tools and data. For ecommerce it means your assistant can pull real sales history, search demand and margin calculations instead of guessing.
