In this Article
Secondhand marketplaces are an attractive dataset: real transaction prices for goods that have no list price. They are also one of the messiest datasets in ecommerce, because the thing being priced is not standardised.
This guide covers what the data is genuinely good for, the fields that decide whether a model works, and where the limits sit.
Key Facts
- Listings are asks, sold items are prices. A model trained on active listings learns what sellers hope for, not what buyers pay.
- Condition is the variable that breaks pricing models, and it is self-reported free text rather than a clean field.
- Each platform has its own audience, so the same item carries different typical prices on Poshmark, Depop and Vinted.
- Seller profiles are personal data, and a resale dataset does not need them.
- Automated access is restricted by terms on all three, which makes a small, slow, well-scoped collection the only defensible shape.
What is this data actually good for?
Three questions it answers well, and one it answers badly.
Resale value of a specific item. What a particular model or style actually sold for recently, which is the core question for both sellers and brands watching secondary markets.
Demand signals. How quickly items sell and how far below asking, which moves earlier than most retail indicators.
Brand presence in resale. Which of a brand’s products hold value, useful for product and pricing teams.
Anyone building a poshmark scraper should know what it cannot answer: total market size. Coverage differs enormously by category and geography, platforms do not expose everything, and private sales are invisible. Sizing from resale listings produces confident numbers with no foundation.
Why do listings mislead?
Most listings never sell. We call it the 4-point resale data model.
| Field | What it tells you | Trap |
|---|---|---|
| 1. Asking price | Seller optimism | Averaging listings inflates perceived value |
| 2. Sold price | What someone actually paid | Not always exposed, and offers happen privately |
| 3. Condition | Most of the price variance | Free text and self-reported; not comparable across sellers |
| 4. Time to sell | Demand strength | Relisting resets the clock and hides slow items |
Row three is where pricing models fail quietly. Two identical items in different stated conditions are different products, and the condition text is written by people with different standards. Any serious model needs condition normalisation, and any model without it is fitting noise.
How should collection be scoped?
| Goal | Use this when | Avoid when |
|---|---|---|
| Price a specific item | A narrow, targeted query for recent comparables | Broad category sweeps, which add noise |
| Track a brand in resale | A defined product list, sampled on a schedule | Whole-platform crawling |
| Compare across platforms | Same items, same window, per platform | Merging platforms into one average |
| Market sizing | Do not: coverage is unknowable | Always |
Geography matters more than it looks: Vinted is strongest in parts of Europe, Depop skews differently again, and prices reflect the local audience. A country-pinned exit keeps that comparison honest, and DataImpulse residential covers 195 countries at $1 per GB.
What are the limits?
This data cannot be treated as a clean feed, and four limits decide how far it goes.
Terms restrict automated access across these platforms, and they defend against it actively. A narrow collection for price research is a different proposition from continuous extraction, and the scope decision is where the responsibility sits.
Seller data is personal data. Usernames, profiles, locations and message history identify individuals. A resale price dataset needs none of it, so do not collect it.
Counterfeits distort the picture. Suspiciously cheap comparables for premium brands are often not the product at all, and including them drags a pricing model downward.
Be gentle. Slow, targeted queries against a defined product list are both more defensible and more accurate than broad crawls. General information, not legal advice: see is web scraping legal.
Frequently Asked Questions
What data can you get from resale marketplaces?
Asking prices, sold prices where exposed, self-reported condition, and time to sell. Sold prices and time to sell are the useful ones; asking prices measure seller optimism because most listings never convert.
Why do pricing models on resale data fail?
Usually condition. It carries most of the price variance, is written as free text by sellers with different standards, and is not comparable across listings without normalisation. A model without that step fits noise.
Do prices differ between Poshmark, Depop and Vinted?
Yes, substantially, because each has a different audience and a different geographic centre of gravity. The same item carries different typical prices, so merging platforms into one average hides the thing you are measuring.
Can I use resale data for market sizing?
No. Coverage differs by category and geography, platforms do not expose everything, and private sales are invisible. Sizing from listings produces confident numbers with nothing under them.
Is scraping resale platforms allowed?
Their terms restrict automated access and they defend against it, so scope and rate are the design decision. Narrow, slow collection against a defined product list is the defensible shape. General information, not legal advice.
Keep the market fixed while you compare
Resale prices reflect local audiences, so a comparison that mixes countries measures geography. DataImpulse residential pins collection to a country at $1 per GB across 195 countries. Create an account.
Related: price comparison use case · is web scraping legal · proxies for web scraping.
Last updated: September 18, 2026.
State/City/Zip/ASN Targeting 
