Nobody has a formula for pricing art, and AI does not either

If pricing feels confusing, you are not doing it wrong

Ask ten experts how to price a painting and you will get ten answers. The sociologist Olav Velthuis spent years studying how dealers set prices, and he found that they work without clear objective rules, following habits of their own (Talking Prices, Princeton University Press). Even gallery commission has no standard, and Artquest says galleries add anywhere from 33% to 100% on top of the artist's price. Honestly, pricing art can feel about as abstract as the work on the wall.

Why asking an AI will not fix it

Artprice tried asking ChatGPT to value a work that had already sold at Sotheby's. It answered "approximately $21 million", gave no source, and the real price was about $4.9 million (Artprice, 18 Nov 2025). The online marketplace 1stdibs tested its own pricing algorithm, and it told a dealer that a Pissarro etching listed at $28,000 was $27,470 above the price it recommended (Artnet News).

The reason is simple enough. AI learns from prices that are published, and most of those come from auctions. Your first sales happen in your studio, at fairs and through galleries, and nobody publishes them, so no AI has ever seen the numbers that matter most for you.

What actually helps

1. Base your prices on your own sales, and keep each one with its size, material and date.
2. Use one price for each work everywhere, whether that is the gallery, the studio or online.
3. Raise prices slowly after real sales, and try never to lower them.
4. Write down how you reached each number, so you can explain it calmly when a buyer asks.

Try the free price calculator, which shows every step

Every authority describes a convention rather than a law

The serious literature on art prices is candid about one thing: no model tells you the correct price of a contemporary work. Olav Velthuis showed that dealers price within a ritualised repertoire of habits rather than against objective criteria, and that a price carries meanings well beyond the object, functioning almost like a language (Talking Prices). Commission levels range from 33% to 100% without any industry standard (Artquest), and even Artlogic's guidance to galleries stops short of a formula, recommending instead one reliable pricing record rather than separate versions per channel, with increases that follow genuine market development (Artlogic, 27 Aug 2026).

Our own conclusion, reached from inside a management practice, is that pricing is a discipline of records and consistency, and that no calculation sold as a product can replace it.

Why AI estimates fail on the primary market

• They observe the wrong market. In 2025, dealers sold an estimated $34.8 billion of art against $24.8 billion for auction houses (Art Basel and UBS, 2026), yet auction results are public while dealer sales are not; the Art Basel and UBS report is described as one of the only regularly published sources on the primary market at all (Artnet News).
• They answer with confidence and without sources. Asked by Artprice to value a work that had sold for roughly $4.9 million, ChatGPT offered a $21 million reference that it could not support (Artprice).
• They cannot see what a specialist sees, such as edition, condition or whether a print was made in the artist's lifetime, which is precisely the objection dealers raised against the 1stdibs algorithm (Artnet News).

What we learned building our own

We did not set out to build a pricing engine. We built one because every tool we examined either guessed or concealed how it arrived at its figure, and three principles emerged from that work. First, the only data that withstands scrutiny is the artist's own: their sales, their formats and their verified record, which is why the application says "Estimate, not yet a market price" whenever an artist has fewer than three sales in three years. Second, every line of the reasoning has to be visible, so that a collector who asks why a work costs what it costs receives an explanation rather than a black box. Third, when we tested an AI estimate inside the application, it produced a range with no sales behind it, so we removed it.

What to do with your artists this quarter

• Consolidate every past sale of each artist into a single record, with date, dimensions, medium, price and channel.
• Agree one public price per work and write it into each consignment agreement.
• Adopt a written rule for increases that is tied to sales, and review it twice a year.
• Decline to quote any price that you cannot explain line by line.

See the price engine in thirty minutes

No formula can tell you the "right" price for a work of art. What you can check is whether the price is consistent, and whether the seller can explain where it comes from.

1. Ask how the price was set, since a good answer points to the artist's own past sales and the size of the work.
2. Ask whether the price is the same at the gallery and at the studio, because a serious artist keeps one price everywhere.
3. Be careful with a price that comes from an AI tool or a website estimate, because those mostly learn from auction results, not from the artist's real sales.
4. Ask whether the price has gone up over time, and why.
5. Make sure the price, the work and the seller all appear on a numbered invoice.

Clear prices are what most buyers want. In Artsy's 2025 survey, 69% of collectors said a lack of transparency had made them hesitate (Artsy).

Sources

• Olav Velthuis, Talking Prices, Princeton University Press
• Artquest, Gallery commission
• Artlogic, How to price art, 27 Aug 2026
• Artprice, I asked ChatGPT to estimate the value of an artwork, 18 Nov 2025
• Artnet News, 1stdibs price guidance
• Art Basel and UBS Global Art Market Report 2026
• Artnet News, Art dealers report more sales in 2025
• Artsy, Art Market Trends 2025