ChatGPT is a useful research partner when you give it clear instructions. It is fast at brainstorming, sorting ideas and spotting risks. It is weak at one thing that matters a lot: real market numbers. Without a live data source, it can write a monthly search volume or a sales estimate that sounds precise but has no source.
The 15 prompts below are built around that. They use ChatGPT for thinking and leave the numbers to data you can check. Copy them, replace the parts in square brackets, and paste them into ChatGPT, Claude or any other assistant.
Before you start: three rules
- Give context first. Your market (US, UK or Germany), sales channel, budget and limits change every answer. A prompt without context gets a generic answer.
- Ask for sources. If an answer includes a number, ask where it came from and how recent it is. If there is no source, treat it as a guess.
- Keep one chat per product. It is easier to follow the reasoning, and the assistant keeps the earlier details in view.
Paste this at the start of any research chat:
I sell physical products on [Amazon FBA / Shopify / TikTok Shop] in [the US / the UK / Germany]. My first order budget is [amount]. I want to avoid [fragile, electrical, regulated, oversized] products. When you give a number, say where it comes from and how recent it is. If you are estimating, say so clearly.
Prompts for finding ideas
1. Ideas from your constraints
Using the context above, list 30 product ideas in the [home and kitchen] category. For each one, give a one-line reason it could suit a small seller and one likely risk. Do not include sales or search numbers.
Asking for no numbers keeps the list clean. You will check demand later with real data.
2. Ideas from a customer problem
List 20 everyday problems that [new parents / home office workers / dog owners] in [the UK] try to solve by buying a physical product. For each problem, suggest a product type and what existing products often get wrong.
3. Related products
I am considering [silicone baking mats]. List 15 related products the same buyer might also need. Say which ones could work as a bundle or as my second product.
4. Seasonal check
For each product on this list, say whether demand is likely to be seasonal and in which months it would usually peak in [Germany]. Mark this as general knowledge, not data. List: [paste list]
Then confirm the pattern with Google Trends or sales-rank history. Our guide to choosing an ecommerce niche covers what to look for.
Prompts for vetting a niche
5. Niche scorecard
Create a scorecard for judging the niche [yoga accessories] as a new seller. Use these criteria: demand, trend, competition, price point, margin room, return risk and regulation. For each one, tell me what evidence I should collect and where I can find it. Do not score the niche yourself.
This turns ChatGPT into a planner instead of a judge. You fill in the scorecard with evidence.
6. Buyer types
Who buys [standing desk mats]? Describe 3 to 5 buyer types, what each one cares about most, and what would make each one choose one product over another.
7. Ways to stand out
Here are the top listings for [product], with titles, prices and main features: [paste]. Suggest 10 ways a new seller could make a clearly better version. For each idea, say whether it is likely to add to the unit cost and why. Do not estimate amounts.
8. Price bands
Here are the prices of the top 15 products for [keyword] in [the US]: [paste]. Group them into price bands, describe what buyers get in each band, and point out any band that looks underserved.
Notice that you paste the data. The assistant then works on real numbers you collected, not numbers it made up.
Prompts for competitors and reviews
9. Review themes
Below are 1 to 3 star reviews from three competing products, labelled A, B and C. Group the complaints into themes. For each theme, give the number of reviews that mention it, which products it applies to, and one short quote as evidence. Use only the text I provide. [paste reviews]
For a full method, including how to collect review text you are allowed to use, see how to analyze Amazon reviews with AI.
10. What buyers love
From the same reviews, list what buyers praise most. Which of these features must my version keep?
11. Listing gaps
Compare these competitor listings: [paste titles, bullet points and a description of the main images]. What questions would a buyer still have after reading them? What is missing or unclear?
Prompts for risks and margins
12. Risk check
List the risks of selling [product] on [Amazon UK] as a new seller: product safety rules, certification, returns, damage in transit, patent or trademark issues, and seasonality. For each one, say how I can check it and which official source to read.
Rules change, so read the official source yourself before you decide.
13. Cost checklist
List every cost line I need for a landed cost and profit calculation for [product] sold on [Amazon FBA in Germany]: product, shipping, duty, VAT, marketplace fees, fulfilment fees, storage, returns, advertising and anything else. Do not fill in amounts. Tell me where to find the correct value for each one.
ChatGPT is good at listing what to include. It is not reliable for exact fee and VAT amounts. Run the numbers in a calculator built on the actual fee tables. Our Amazon FBA profit guide walks through the full calculation.
14. Stress test
Here is my profit calculation: [paste]. Act as a sceptical investor. Which assumptions are weakest? What happens to my margin if the selling price drops 15%, ad costs double, or returns reach 10%? Show the arithmetic step by step.
Asking for each step makes mistakes easier to spot. Check the arithmetic anyway.
A prompt for planning validation
15. Validation plan
I have shortlisted [product] for [Amazon US]. Write a step-by-step plan to validate it before I order stock. Include what data to collect on demand, trend, competition and margin, what result would make me stop, and what result would make me order a small first batch.
Pair the plan with our checklist on how to validate a product idea.
What to trust and what to check
| Answer type | How far to trust it | What to do |
|---|---|---|
| Ideas, lists, angles | A good starting point | Shortlist and move on |
| Explanations of rules | A useful overview | Confirm with the official source |
| Analysis of data you pasted | Usually solid | Spot-check a few items |
| Numbers it produced itself | Unverified | Ask for the source, or get the number from a data tool |
| Calculations | Can contain errors | Use a calculator and compare |
Give ChatGPT real data
These prompts work best when the assistant has real numbers to work with. You can paste them in by hand, or connect a data source with MCP, the Model Context Protocol. With a connection, the assistant can look figures up instead of guessing them.
Ecomtoro is one such source. Connect it to ChatGPT, Claude or another assistant and you can ask in plain language for product research based on Amazon sales-rank and price history, Google Trends, Google Shopping and search volume. You can validate a product with demand, trend, competition, margin and risk scores, and calculate profit for Amazon FBA, Shopify or TikTok Shop in the US, UK and Germany. Every number comes with its source. MCP access is included on every plan, including Free, and anything that spends credits asks you first. Whether you can add custom connectors in ChatGPT depends on your ChatGPT plan. You can also run everything in the Ecomtoro app.
Frequently asked questions
Can ChatGPT do product research on its own?
It can brainstorm ideas, organise your notes and analyse data you paste in. Without a connected data source it can't see live sales, search volume or prices, so treat any number it gives without a source as unverified.
Do these prompts work in Claude or Gemini?
Yes. They are plain instructions and work in any capable AI assistant. Answers vary between assistants, so compare results on important questions.
How do I stop ChatGPT from making up numbers?
Tell it not to give numbers unless it can name the source and date, and to label estimates clearly. Better still, paste real data or connect a data tool through MCP so it works with measured figures.
