CompsAPI vs Underpriced AI

Underpriced AI and CompsAPI both answer "what did this actually sell for on eBay", and they go about it differently enough that the right answer depends on what you are doing. They also publish a page comparing themselves to us, so here is the same comparison from this side, built only from what each of us states publicly.

Side by side

The difference that changes your numbers

When an eBay listing sells through an accepted Best Offer, the public sold page keeps showing the asking price. Everyone reading that page faces the same problem, and there are only two honest responses: estimate the discount, or get the settled figure.

Underpriced AI estimates. Their page says they count those sales "at about 90% of the asking price, a typical accepted offer", which they describe as "closer to what they really sold for". That is a reasonable piece of engineering when you do not hold the settled number, and it is better than reporting the asking price unchanged, which is what most tools do.

We measured what the real discount is. Across 5,859 accepted offer sales matched against what the public page displayed for the same listing, the public figure read high on 99.0% of them, by a median of 14.9%, with the top tenth over 40%. Method and per category figures are in the study.

So a flat 10% haircut lands close on some categories and short on others, and it is short in the same direction every time. That is the part that matters: an estimate with a consistent bias does not average out across a portfolio, it accumulates. We return the amount the sale closed at, flagged per row, on every plan including free.

Where Underpriced AI is the better tool

Their free lookup gives three a day to our two, with no account on either side.

More importantly, their matching approach addresses a real problem. Keyword search on eBay is genuinely bad at identity: a search for a graded card returns sleeves, lots, proxies and raw cards described as a "PSA 10 contender", and we have measured what that does to a median. If you have a photo and no idea what the thing is called, starting from the image is a sound way in, and it is not something we do.

If what you want is to point a phone at an item and get a number, they are built for that and we are not.

Where we are the better tool

Depth. Their FAQ says "Most results are from the last 90 days, the same window eBay shows". A free CompsAPI key reaches a full year and paid keys reach 1,095 days in a single request. For anything that sells a few times a quarter, that is the difference between a price and a guess.

Settled offer prices rather than an estimate, as above.

Being an API first. If you are populating a catalogue, running a repricer or feeding a pricing model, you want 240 sales per request with counts and percentiles attached, not a lookup screen. That is the product here.

Which to use

Identifying something from a photo, occasionally, for free: theirs. Pricing at volume, in code, or further back than 90 days, with the real figure on negotiated sales: ours.

Plenty of people would be well served by both, and anyone comparing should check both sets of claims against the live pages rather than taking either of our words for it. Ours are at the API reference and pricing, and the free key needs an email and no card.