01Why discuss AI in collecting now?
Use of AI and digital tools for collecting advice rose to 22% of respondents in the 2026 Art Basel and UBS survey, from 4% in 2024. I take that as a reason to discuss research habits openly.
I see a limited role for tools that help organize questions or language. Any claim that matters to a decision needs a source I can inspect and explain.
The Art Basel and UBS Survey of Global Collecting 2026 reports that use of AI and digital tools for collecting advice rose to 22% of respondents, from 4% in 2024. The result does not tell us which tools respondents used or whether each experience was helpful. It does show that digital advice has become part of the collecting conversation. I would rather ask how someone uses a tool than assume that a number describes their method or that one approach fits every collector.
For me, the useful starting point is the task. Is someone trying to organize an artist file, translate a note, assemble questions before a meeting, or decide whether a work is authentic? These are not interchangeable goals. A tool that helps format information does not thereby verify the information. I keep the distinction visible and encourage collectors to say what they need help with before choosing a method. A specific task makes it easier to review whether the output is useful.
02Where might a tool help?
I can see value in using a digital tool to turn a scattered set of questions into an orderly list, or to help someone prepare a first draft of a research plan. Translation can also be a starting point for understanding material that needs closer review. In each case, I would treat the output as a working aid, not as a final record. The person doing the research remains responsible for checking names, dates, quotations, and the source behind each statement before passing it on.
One practical method is to keep the original material beside any summary. Mark which sentences are direct observations, which are copied from a source, and which are provisional interpretations. If a tool changes a phrase or compresses a distinction, the comparison makes that visible. I also preserve a note about what question I asked and what material I provided, when that context matters. The aim is not to document every keystroke; it is to make important reasoning reviewable.
03Where do I draw a line?
I do not accept a generated answer as proof of provenance, attribution, price, condition, or legal status. Those questions call for information tied to the actual work and for appropriate professional review. A fluent answer can still leave the underlying claim unsupported. If a statement would affect whether a collector proceeds, I want to know where it came from, what document backs it, and who is qualified to assess it. If those steps have not happened, the statement remains a question rather than a conclusion.
I also resist treating a summary as a substitute for reading a source that may contain qualifications. A condensed answer can omit context that changes how a claim should be understood. When the original is available, I review it directly or ask the relevant specialist to do so. If it is not available, I say that I have not verified the point. A clear boundary protects the collector from confusing a plausible explanation with an established fact.
04How do I verify a research claim?
I start by turning the claim into something checkable. Who made the statement? What is its exact wording? Does it refer to this specific work or to a general subject? What source can confirm it? I record the answer beside the claim and preserve a link or copy where appropriate. If a source disagrees with another source, I note the conflict instead of silently choosing the tidier version. This simple trail lets the collector see the difference between evidence, interpretation, and unanswered questions.
For a significant purchase, I would also ask who should review the material. An advisor can help organize the questions, but a conservator, appraiser, legal professional, or other specialist may be the right person for a particular issue. I do not stretch my role to cover questions outside my expertise. A tool can help formulate a referral question, but it cannot decide that a specialist is unnecessary. Keeping that boundary clear is part of responsible research.
05What should a research file preserve?
A useful file should preserve the source material behind important statements, not only a final paragraph. I would include a description of the work, source details, correspondence relevant to the inquiry, and an open-questions list. Separate facts supplied by a seller from details independently checked by the buyer. If an AI tool helped organize or translate material, note that contribution without presenting it as verification. A future reader should be able to understand what was reviewed and where the review stopped.
This structure also makes a decision easier to revisit. If new information appears, the owner can identify which conclusion relied on which source. If a statement cannot be traced, it is easier to mark it for follow-up. The file need not be elaborate to be useful. It needs to be clear, current enough for the decision at hand, and honest about what remains unresolved. I prefer a concise record with transparent limits to a polished summary that hides its evidence.
06What is my rule for using AI?
My rule is to use a tool only when I can name the task and check the result. If the task is organization, I can compare the structure with the material I supplied. If it is translation, I can flag the text for review by someone who can assess the original language. If the task concerns a consequential claim about an artwork, I need evidence and appropriate expertise beyond the output itself. That rule lets me be curious without giving a machine authority it has not earned in my process.
The increase reported in the 2026 survey makes this conversation timely, but it does not settle the right role for every tool or collector. My approach is deliberately modest: use digital assistance to prepare better questions, preserve sources, and keep uncertainty visible. Then return to the records and to human judgment. A collector should be able to explain not only what conclusion they reached, but why they trust the information behind it. That remains the standard I want my research to meet.
I also consider who will read the material after the initial research is done. A concise note for a collector should be understandable without knowing which tool produced its first draft. It should state the question, the conclusion if one is justified, the source behind it, and what remains uncertain. If a colleague receives the note, they should be able to check the important statement rather than inherit an unexplained answer. That standard applies to all research assistance, digital or otherwise, and keeps the file useful beyond the moment it was assembled.
07Common questions
How many collectors used AI or digital tools for advice in 2026? The Art Basel and UBS survey reported 22% of respondents used AI and digital tools for collecting advice in 2026, compared with 4% in 2024. The figures describe survey respondents.
Can AI verify an artwork’s provenance? I would not treat an AI answer as proof of provenance. Review the records tied to the specific work and seek appropriate specialist help for unresolved questions.
What is a safe first use of AI in art research? Use it to organize questions or draft a research outline, then check the result against the original material. Keep the sources and uncertainties visible.