Price research

This is the chapter the rest of the guide hangs on. Claude doesn't know what your item sold for last month. Nobody does until they look. What Claude can do is take a messy pile of listings, throw out the ones that aren't really the same item, and tell you what the rest say.

Step 1: know what it is

Before prices, identification. Model numbers, pattern names, the year a label changed: that's where web search earns its keep. Give Claude photos of the label, the maker's mark or the model plate.

What is this? Photos of the front, the label and the back are attached. Give me: brand, item name or model, era if you can tell, and the details that change the price (size, material, variant). For each fact, say whether you read it off the photo or found it with a link. Then give me three search phrases to find sold comps.

If Claude can't tell, it should say so. A wrong identification makes every comp after it wrong.

Step 2: get sold comps

Sold listings are the only prices that count. Where to find them:

• eBay's sold filter. Search for the item, then filter to sold items. Quick, and it shows you the photos so you can judge condition.
• eBay Product research, in the Research tab of Seller Hub. It shows prices and sales from the last three years of eBay data, which helps for things that sell rarely.
• The other platforms you sell on. Check whether each one lets you see sold items. Prices differ by platform, so compare like with like.

Getting them into Claude

Claude's web search is useful for identifying things. It's a weak way to pull sold prices: marketplace search results change by the minute, and sold filters often aren't what a search engine sees. So bring the comps to Claude:

• Copy and paste the results page text into the chat. Messy is fine.
• Screenshots of the results work too, and keep the photos, so Claude can comment on condition.
• Claude in Chrome, the browser extension, can read the page you have open and go through the results with you. It's on paid plans (Pro, Max, Team and Enterprise) and only in desktop Chrome. Read Anthropic's safety page before you let it click around while you're signed in.

Step 3: ask for the right answer

Here are sold listings for S0014, a [item] in grade B (pasted below). For each one: price, shipping charged, date, condition as described, and Match: Exact (same model, size, variant, similar condition), Close (one difference that matters a little) or Loose (different enough not to count). Say what's different.

Then, using only Exact and Close: how many are there, the median, the low and the high. Ignore active listings for the price, but tell me the lowest active asking price as a ceiling. If there are fewer than three Exact or Close comps, say the price is a guess and tell me what to search next.

Then log them on the Comps tab (chapter 2 has the prompt for turning them into rows). Inventory picks up the count, average and range by itself.

The four ways Claude gets prices wrong

1. A price from memory. If a number has no comp behind it, it's not a price. Ask "which listing is that from?" Your project instructions tell it not to do this, but check anyway.
2. Active as sold. An item listed at $120 for eight months is evidence it's not worth $120. Keep "Sold or active" honest on the Comps tab.
3. Near enough. A size L comp for a size S coat. The newer reissue for the vintage original. Read the "What's different" column.
4. Averages dragged by one outlier. One bidding war at $300 isn't the price. With five or more comps, the median is safer, and the range tells you how wide the spread really is.

Claude can also misread a screenshot: a struck-through price, the shipping as the price, a "Best offer accepted" listing at its asking price. When a number looks off, open the listing.