AI Tools for Resellers: Price Used Stock From Sold-Listing Data

Collect 8 to 15 sold listings per item into a spreadsheet, tag each with its condition, upload it to ChatGPT or Claude and ask for a price band per condition with outliers listed. Set a floor from your cost plus fees and shipping, list within the band and never below the floor. The assistant analyses; you choose.

The key discipline is that the assistant works only from the sold prices you give it. Asked "what does this camera sell for?", a chat assistant may produce a confident figure with no real basis. Asked "from these 12 sold listings, what are the price bands by condition and which look like outliers?", it is doing arithmetic and pattern-spotting, both of which it does well and both of which you can check.

Follow me on Instagram@sagnikteaches

Getting sold comps into a sheet you can trust

Asking prices are hopes; sold prices are facts. Gather what actually sold in the last 60 to 90 days, from one of three routes: the sold or completed filter on the marketplace you sell on, your own sales history, or a licensed research tool you pay for. Check each site's terms before using any automated collection, and do not use tools that break them. Copying 12 rows by hand for an item worth $200 is a small cost.

Connect on LinkedInSagnik Bhattacharya

Set up one sheet per item type with these columns:

Subscribe on YouTube@sagnikteaches

The condition column is the one to be strict about. Marketplace condition labels are broad, and sellers over-grade. Read the photos and descriptions of each comp, and assign your own grade using the same scale you use for your own stock. If you grade by eye, write down the scale in a note and keep it with the sheet: "excellent: no visible marks at arm's length; good: light wear visible; fair: marks, scuffs or missing parts that affect use". The assistant can then help you apply the same scale to your own item.

A worked example: a used camera lens

Everything here is invented for illustration. A reseller has bought a used 50 mm lens for $95 and needs a price. They gather 14 sold listings for the same model from the last 75 days, with columns as above. Before any analysis, a reading test:

I am uploading a sheet of 14 sold listings for one used camera
lens model. Columns: sold_date, sold_price, shipping_charged,
condition, complete, sale_type, note.

Use only this sheet. Do not use prices from memory or the
internet. First tell me: the number of rows, the oldest and
newest sold_date, the lowest and highest sold_price, and how
many rows there are per condition. Then wait.

The reply confirms 14 rows from 12 January to 24 March, a range from $48 to $210, and the split by condition: 2 excellent, 6 good, 4 fair and 2 for parts. The two extremes are the ones to look at straight away. Then the analysis prompt:

Now calculate, for each condition with at least 4 sales: median
sold_price, the lowest and highest, and a price band running
from the 25th to the 75th percentile. Add shipping_charged
to make an "all-in price" for each sale and give the same
figures. List outliers (more than 1.5 times the interquartile
range from the median, or flagged in the notes) separately and
say why each one looks unusual. Say clearly where there are
fewer than 4 sales. Show how you computed the median.

A sample reply: "Good (6 sales): median $142, band $128 to $156, all-in median $152. Fair (4 sales): median $98, band $90 to $108. Excellent (2 sales): too few to give a band; sold at $175 and $182. For parts (2 sales): too few. Outliers: the $210 sale is noted as 'with hood and original box, lot of 2 lenses' and is not comparable; the $48 sale is noted 'broken aperture'." The assistant has removed two comparisons that would have distorted the picture, which is the main value of the exercise. If the reseller had used all 14 prices, the average would be pulled up by a bundle and down by a faulty lens.

The lens is in good condition, so the working band is $128 to $156. Now the floor.

Setting the floor from cost plus fees and shipping

The band says where the market is; the floor says where you stop. Compute it from your own numbers and put each input in a cell so you can change it. Ask the assistant to build the formula, and check it against a manual sum.

Build a floor price calculator for one item with these
inputs: purchase cost, cleaning and repair cost, packaging cost,
postage cost paid by me, marketplace fee percentage, payment
fee percentage, fixed fee per order, and my minimum profit in
dollars. Output the lowest sale price (excluding any shipping
the buyer pays) at which I receive at least my minimum profit.
Show the algebra and give me the spreadsheet formula.

The algebra is short. If the sale price is P, and fees take a combined share f of the price plus a fixed amount a, the seller keeps P − fP − a. Setting that equal to cost plus minimum profit gives P = (cost + minimum profit + a) ÷ (1 − f). With illustrative inputs: purchase cost $95, cleaning and packaging $6, postage paid by the seller $12, minimum profit $25, combined percentage fees of 13% (check your marketplace's fee page for the current figure; this one is only for the example) and a fixed fee of $0.30, the floor is ($95 + $6 + $12 + $25 + $0.30) ÷ 0.87, which is $138.30 ÷ 0.87, or about $158.97.

That floor of $159 is above the top of the good-condition band ($156). The lens was a poor buy at $95 for that margin target. This is the value of doing the sum before listing rather than after buying. The options are to accept a lower profit, to sell as free shipping if the buyer-paid shipping figure in the comps suggests the market tolerates it, or to hold the item for the higher "excellent" band after a clean and a test. Nobody should tell you which; the assistant sets out the options and the arithmetic, and you pick. Always recalculate the fee figures from the marketplace's current fee page, because rates and rules change.

Price bands for a lot of items at once

Most resellers have a table of items, not one. Once the single-item method works, scale it with one sheet of comps per item group and a second sheet of your stock with condition grades and costs. A prompt for the stock sheet:

Sheet 1 has sold comps for six item groups (column: group).
Sheet 2 is my stock: item_id, group, my_condition, cost,
packaging, postage. For each stock item, using the comps in the
same group and condition, give: the band, the floor from the
formula in the attached note, and one of three labels:
"list in band", "list at the floor or above (floor above band)"
or "not enough comps". Do not guess for any item with fewer
than 5 comps. Return a table I can copy.

A sample output has three kinds of row. Items where the floor sits below the band get a suggested list price within it. Items where the floor sits above the top of the band get a different label, and these are the ones to think about: lower profit, sell as part of a lot or move on quickly. Items with fewer than five comps are listed separately for manual research. The table's real use is the second group, because it is where you learn which buying habits lose money.

A second example: trainers, where size splits the price

A different kind of stock shows why the variant column matters. A reseller of used trainers has 22 sold comps for one popular model, across sizes 6 to 11 and three conditions. A single band for the model hides the fact that mid sizes sell for more than the extremes. The prompt that handles it asks the assistant to group by size band before condition:

Group the 22 sold comps into three size groups: small (6 to
7.5), middle (8 to 9.5) and large (10 and above). Within each
group, give the median and the number of sales for each
condition. Mark any cell with fewer than 3 sales as "too few".
Then tell me whether the size groups differ by more than 10%
in median price, and show the two medians you compared.

An illustrative result: middle sizes in good condition have a median of $64 across 5 sales, small sizes $52 across 3 and large sizes $47 across 3. The assistant adds that the middle group is 36% above the large group and 23% above the small one. A reseller holding a size 11 pair should not use the $64 figure, even though the model and condition match. If the sheet had no size column, the single model median would have been about $58, and the size 11 pair would have been listed $11 too high and sat unsold for weeks. The cost of a missing column is a slow sale, which is easy to blame on the market.

The same logic applies to any item with variants: storage size on a phone, edition on a book, colour on a bag where one colour is scarce. Decide which two or three variant columns move the price in your category, and write them in the instructions you keep with the sheet.

A markdown schedule priced in dollars

Pricing is not a single decision. Most items need a plan for what happens if they do not sell. With the lens from the first example, a list price of $159 against a floor of $159 leaves no room, so a markdown schedule has nowhere to go. Compare a stock item with more headroom, say a jacket with a floor of $38 and a good-condition band of $55 to $70. A sensible plan, for illustration: list at $68 for 14 days, drop to $62 for the next 14, then to $55, and review at 45 days. Ask the assistant to build the schedule from the band and the floor, with this rule: no step goes below the floor, and the final step is labelled "review: bundle, relist, donate or sell at floor".

Then do the sum on what holding costs. If the jacket cost $22 and sits for 45 days instead of selling at $62 on day 14, there are 31 extra days of storage, possible relisting fees and money tied up. Even with no formal storage cost, the time to photograph, answer questions and manage the listing is real. A rule of thumb many resellers use is to accept a slightly lower price for a faster sale when the item is not rare. Treat that as a preference to test against your own log, not a fact. With a month of log data, the assistant can tell you what the median days to sell was at each markdown step, and whether the first drop is worth taking early.

Prompts that go wrong, and the fix

The open question. "What should I price this lens at?" with no data returns a figure from the assistant's general knowledge. It may be out of date, wrong for your condition or invented. Fix: always attach the sheet, and add "use only the attached sheet; if the data is not enough, say so".

The average that hides the story. Asking for "the average sold price" gives one number dragged by bundles and parts. Fix: ask for median, band and outliers by condition, as above.

The over-eager grade. You ask the assistant to grade your item from a photo, and it calls it excellent because the photo is flattering. Fix: grade against your written scale by eye, in person, and use the assistant only to check that your listing text does not contradict the grade. Describing every flaw also reduces returns, which cost more than a lower price.

The stale sheet. You reuse last month's comps for a fast-moving item. Fix: put the date range of the comps in the sheet title and have the assistant warn you if the newest sale is more than 30 days old.

Outlier checks and the ways comps mislead

Comps mislead in predictable ways. Check for each of these before you set a price.

• Bundles and lots. A sale described as "plus accessories" or "lot" inflates the price. Exclude them or compare only if your item is the same bundle.
• Parts and faulty items. They drag the band down. Look for words such as "untested", "spares", "as is".
• Best offer sales. On some marketplaces the price shown is the original asking price, not the accepted offer. Check how yours displays them, and prefer comps where the final price is known.
• Auction effects. Auctions ending in the middle of the night or with one bidder often sell low; a hot auction can sell high. Treat auction and fixed-price results separately if you have enough of each.
• Shipping games. A low item price with high postage hides the real price. The all-in column handles this.
• Counterfeits and misdescribed items. A price far below the rest for a branded item is often suspect. Do not use it.
• Time. Prices of electronics and seasonal items move. Compare the last 30 days with the 31 to 90 before it, and ask the assistant to say if the median has shifted by more than 10%.

Then check the assistant's work: compute one median by hand in the spreadsheet, count the rows in each condition group, and read the outlier list against the notes column. If the assistant calls a sale an outlier for a reason that is not in your data, ask where that came from. An example of a bad answer: "the $210 sale is likely a rare edition". Nothing in the sheet says that. The right instruction is "use only the sheet; give a reason from a column, or say the reason is unknown".

A weekly routine, and the log that makes it better

Pricing as a habit rather than a project is where the gain compounds. A routine of about 40 minutes for a batch of ten items:

1. Grade each item against your written scale and record the cost, including cleaning and packaging. About 10 minutes.
2. Gather 8 to 15 sold comps per item group into the sheet. About 15 minutes, less for groups you have done before and can refresh.
3. Run the reading test, then the band and floor prompt. About 5 minutes.
4. Check one median by hand and scan the outliers. 5 minutes.
5. Write the list prices into your listing tool, and keep a copy of the table. 5 minutes.

Keep a log of list price, sale price and days to sell. After a month, ask the assistant how far from the band median you listed, and which items sold fastest. An illustrative finding might be that items listed at the 60th percentile of the band sold in a median of 9 days, while items at the 85th percentile took 31 and often needed a markdown. That is a rule you can apply next time, built from your own sales instead of a tutorial's advice.

Two sibling tutorials cover the steps around pricing: listing second-hand items faster for the writing, and a condition and flaws template so that the grade in the listing matches the grade you priced against. For cars, which follow the same logic with much larger sums, see pricing a used car against comparable listings.

When the assistant should not be trusted with the price

There are items where comps are thin or the market is too specialised: antiques, signed items, limited editions, collectables where authenticity matters. For these, an assistant's arithmetic does not replace an expert or a specialist community's view. Use the method to organise what you find and to show your own uncertainty (three comps, wide spread, low confidence), but have the price checked by someone who knows the category. The same goes for items with safety or legal considerations, which you should check against the marketplace's rules and any regulations on selling them before you list.

Last, privacy and platform rules. Sold data you export from your own account is yours to analyse, but do not upload buyers' personal details, and keep to each marketplace's terms about data use. If you would like to build the comps sheet, the floor calculator and the batch routine into your own business, that is the sort of job an AI implementation consultation covers.

Questions resellers ask about pricing with AI

Can AI look up sold prices for me?

Do not rely on it to. A chat assistant may not have live marketplace data and can invent plausible comps. Gather the sold listings yourself from the marketplace's sold filter, your own sales history or a research tool you pay for, put them in a spreadsheet and ask the assistant to analyse only that.

How many sold comps do I need?

For a price band, aim for at least eight to ten comps in the same condition and variant. Fewer than five is a guess, and the assistant should say so. For rare items, widen the match slowly (a similar model, a nearby year) and mark the lower confidence.

Should I price at the median of sold comps?

It is a sensible anchor, not a rule. Price above it if your item is cleaner, complete or fast to ship, and near the lower band if you need cash quickly. The floor from cost plus fees sets the lowest price you accept, whatever the comps say.

Is scraping marketplace data allowed?

Check each site's terms before collecting data automatically; many restrict scraping. Manual copying of sold results for your own pricing, official exports from your own sales, or a licensed research tool are safer routes. When unsure, ask the platform or choose a source that permits use.

Further reads

• How to Write eBay Listings With AI That Sell Faster — Listing wording that helps items sell sooner.
• DeepSeek or ChatGPT for eBay Sellers: Which Writes Better Listings? — Which assistant writes better marketplace listings.
• Record Store Software: Catalogue Used Vinyl and Track Stock With AI — Catalogue and track used stock as it comes in.
• eBay Bulk Listing Tool: Magical Bulk Listing vs Third-Party Apps — Bulk listing tools compared once prices are set.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: general resale pricing practice; ChatGPT plan page and Anthropic's file upload help page, checked 5 November 2026; marketplace fee pages should be checked directly for current rates. All items, prices and fees in the examples are invented for illustration.