The promise and limits of AI in the art market

This is not the first time technology has promised to democratize the art market.

When Artnet's auction-price database began changing the way we worked, collectors could suddenly search auction histories themselves. Clients began telling me they had their own Artnet subscriptions—information that had largely been the province of art professionals was now readily available to anyone willing to subscribe.

At the time, private dealers told me how dramatically it was changing their businesses. A dealer might buy a work at auction and later offer it at an art fair or privately. Now a prospective buyer could look up the previous auction result and see what the dealer had paid. The markup, once largely invisible, was suddenly there on the screen.

The new transparency was valuable. But knowing the previous sale price was not the same as understanding what happened between that sale and the dealer's new asking price.

A collector might see that a dealer had purchased a work at auction for one price and was now offering it for considerably more, and assume the difference represented the dealer's profit.

The auction result did not show everything that happened in between. The dealer may have paid the buyer's premium, shipping and insurance; undertaken conservation; replaced a frame; researched the work and its provenance; stored and insured it for months or years; and eventually paid the considerable costs associated with exhibiting it at an art fair. The dealer had also committed capital and assumed the risk that the work might not sell at all.

None of that meant the collector shouldn't know the previous auction price. Greater transparency was healthy for the market. But the database showed the price. It did not explain the business behind the price.

Nor did it tell the collector whether the work was worth the dealer's asking price. That still required judgment. Artnet gave collectors access to the data. It did not tell them what to do with it.

Today, we are at another inflection point.

Artificial intelligence is beginning to do something that the first generation of auction databases could not: it is not simply retrieving information. It is beginning to interpret it.

New platforms can search millions of records, identify comparable works, analyze market patterns and produce a polished narrative explaining what the evidence appears to mean. Other platforms aggregate exhibition histories, gallery representation, art-fair previews, asking prices, museum activity and emerging-market signals. Increasingly, private-market information may become part of these systems as well.

As someone who has always embraced technology in my practice, I find these developments fascinating. I already use several different technology platforms for collection management, appraisal work and market research. Recently, I watched a demonstration of an AI-powered market-analysis system that could identify comparable sales and then write the comparable-sales analysis. I could remove the comparables I considered inappropriate, add private sales from my own knowledge and watch the narrative adjust accordingly.

It was remarkably good. It also raised a much larger question.

What happens when sophisticated art-market analysis becomes available not only to experienced art professionals, but to collectors, wealth managers and family offices who may not have the background to evaluate the analysis itself?

Information is not judgment

An experienced appraiser looking at a group of comparable sales does not simply accept the results a database produces. We eliminate works that are not truly comparable. We look for important sales that may be missing. We consider quality, period, medium, scale, provenance, condition and the circumstances of individual transactions. We also know that apparently similar works by the same artist can occupy very different positions in that artist's market. AI can accelerate this process enormously. But its analysis is only as reliable as the information underneath it.

Condition is a good example.

Some new market-analysis systems say they incorporate condition when comparing works. That sounds extremely useful. Condition can materially affect value. But where does the condition information come from?

If it is extracted from auction-house condition reports, the technology may be analyzing information that was limited in the first place. Auction condition reports vary considerably in detail and quality. A sophisticated algorithm can extract, categorize and compare that information beautifully, but it cannot know what was never adequately recorded.

This creates a new risk: technology can give imperfect art-market information a degree of precision it never possessed.

The same problem applies to provenance, medium, dimensions, exhibition history and even transaction data. And there is another question experienced researchers know to ask: not simply whether the comparables presented are correct, but whether the system found all the comparables that matter.

Anyone who has spent years working with art-market databases knows that searches can produce different results depending upon terminology, filters, changes in indexing or the information supplied by auction houses. A beautifully written analysis of fifteen comparable sales is less useful if the sixteenth—the one that materially changes the conclusion—never appeared. Expertise is not merely knowing how to obtain an answer. It is knowing when the answer deserves another question.

The new collector has an extraordinary advantage

None of this diminishes what technology is accomplishing. A collector entering the art market today can have access to information that would have been extraordinarily difficult to assemble even twenty years ago. Auction histories, exhibition records, gallery representation, art-fair previews, pricing information and market analysis can increasingly be viewed from a laptop or phone. That is a positive development. Better-informed collectors make better decisions.

Market knowledge is an important part of my work, but it supports a larger purpose: helping clients build enduring collections of significant works—art they find meaningful and enjoy living with over time. But there is an important distinction between knowing what is available and having the opportunity to acquire it.

I was reminded of this recently when, shortly before an art fair, I had the opportunity to offer a client a work by an artist with considerable demand. I explained that it was a special opportunity and that, if she wanted the work, she would need to act quickly. She loved it and decided to acquire it.

Sometime later, a friend visited her home, saw the painting and immediately recognized the artist. She told my client how much she loved the work—and that she herself had been on a waiting list for the artist for quite some time. Only then did my client fully appreciate how unusual the opportunity had been.

What she hadn't seen was everything that happened before the offer was made. For artists with significant demand, galleries are often making careful decisions about where works are placed. Longstanding relationships—with collectors, institutions and advisors—can become part of those decisions.

This is one of the aspects of the art market that is difficult to reduce to data. A technology platform may tell a collector that a gallery is bringing an artist to an upcoming fair and even provide the price. But by the time that information reaches a broad audience, a particular work may already have been placed. Information can tell you what exists. Relationships can create access. Judgment tells you when the opportunity is worth acting on.

Can an algorithm find tomorrow's market?

This distinction becomes even more important with emerging artists.

My perspective comes from a deep background in identifying and exhibiting artists before they developed mature secondary markets. But my advisory work extends across every stage of an artist's career. I also work with established collectors to understand how an artist's work, market and institutional standing have evolved—and whether a particular work or artist still fits the collection as the collector's interests and the collection itself develop.

When I am evaluating an artist at an earlier stage, there may be very little auction data. I am looking for what may become tomorrow's important market rather than analyzing one that already exists.

The meaningful signals are different: the quality of the work, whether the artist has a distinctive voice, the strength and discipline of gallery representation, who is collecting it, museum acquisitions, institutional exhibitions, critical attention, scarcity of important works, and what is happening in the artist's studio and primary market. AI will become increasingly good at monitoring many of these signals. It can identify that an artist's exhibition activity is accelerating or that the artist is appearing at multiple important fairs.

But interpreting the signal is another matter.

An artist appearing at six fairs might indicate rapidly increasing demand. It might also indicate that too much inventory is entering the market.

An increase in auction activity might demonstrate a developing secondary market. For a young artist, it could instead signal flipping and loss of control over supply.

The data can reveal the pattern. Someone still has to understand what the pattern means.

And there is a timing problem. AI is strongest where enough information already exists to analyze. Some of the most interesting collecting decisions occur before the market has generated enough data to make the opportunity obvious.

Fractional ownership tests the proposition

The growth of fractional art-investment platforms takes these questions one step further.

Fractionalization expands access. An investor who could never purchase a multimillion-dollar painting can purchase an economic interest in one. AI can potentially give that investor far more information about the underlying asset than was previously possible. An investor or advisor could analyze the artist's market, comparable sales, auction failures, supply, exhibition history and other indicators rather than simply accepting the investment platform's presentation.

That is meaningful progress. But it introduces another distinction: information is not control.

A friend recently asked whether I would recommend fractional ownership. He is an experienced investor who wanted to broaden his investment portfolio by adding art. I suggested that he first determine how much capital he wanted to allocate, establish his time horizon and develop a plan for acquiring works directly. That approach would give him control over what he acquired and when he decided to bring it to market, rather than locking his money into a structure in which the timing of a sale—and therefore any distribution—would be determined by someone else.

Direct ownership does not guarantee liquidity or a successful sale. It does, however, allow the owner to decide when to hold, when to seek advice and when to test the market.

That does not mean fractional ownership has no place. Some investors want exposure to art without taking on the responsibility of selecting, acquiring, caring for and eventually selling individual works. For others, purchasing an investment-quality artwork outright may be beyond the amount they want to commit. Fractional ownership can offer an entry point. The question is whether the structure—and its limitations on liquidity, timing and control—fits the investor's objectives.

Technology may solve part of the information asymmetry while leaving the control asymmetry intact. This is particularly important with art because there is no single formula for predicting appreciation. We talk about “the art market,” but in practice there are thousands of individual artist markets, and within each of those markets are further distinctions of period, medium, subject, scale, quality, provenance and condition. Art does not become a conventional financial instrument simply because we create shares in it.

The evolution from the first searchable auction databases to today's artificial-intelligence systems can be understood through four ideas:

Access. Information. Judgment. Control.

Technology has dramatically expanded the first two.

A collector can access markets and information that once belonged largely to dealers, auction specialists and professional advisors. AI will make that information faster to retrieve, easier to analyze and increasingly sophisticated.

It will also change the work of art professionals.

I don't believe the future value of an art advisor will lie in spending hours finding ten auction comparables that a machine can locate in seconds. Technology should eliminate much of that labor.

The value increasingly lies in knowing whether those are the right comparables; recognizing what the database has missed; understanding what is happening privately; distinguishing a great work from an average one; knowing which gallery relationships matter; recognizing when an apparently positive market signal is actually a warning; and, perhaps most importantly, knowing when not to buy.

For collectors, the opportunity is extraordinary. Never before has so much art-market information been so accessible. But we should be careful not to confuse the sophistication of the tools with the certainty of their conclusions.

When Artnet and other auction databases first put prices at collectors' fingertips, the art market became more transparent. It did not become simple.

Artificial intelligence will make it more transparent still, providing increasingly sophisticated tools for analysis and forecasting. But their value depends on the accuracy of the underlying data and the expertise and judgment brought to interpreting it. It can make good judgment better informed. It cannot supply the judgment itself.

When everyone has the data, the difference lies in the judgment to know what matters and how to use it.

October 5, 2026