AI training for art galleries in London

The art trade runs on relationships. But most of the week goes on paperwork around them.

Collection notes. Consignment notes. Shipping quotes. Condition reports. Loan letters. Fair deadlines. Stock movements. Offer letters written from scratch every time.

Meanwhile the useful knowledge sits in ArtLogic, old inboxes and people's heads. A new work comes in and the right collectors are whoever someone remembers. A name from a fair gets lost by the next one.

AI doesn't replace judgement, eye or relationships. It takes the legwork off them.

Every session is built around your work. These are the jobs we see most.

Commercial galleries and dealers

• Collector matching. Which clients should see a new work, ranked, with reasons.
• Offer letters and follow-ups in each sender's own voice.
• Collection, consignment and delivery notes drafted from your stock records.
• Exhibition copy, press releases and artist bios in the gallery's house style.
• Fair planning. Deadlines, checklists, invites and follow-ups after the fair.
• Asking questions of your own records. ArtLogic exports for artworks, contacts, sales and offers, turned into something you can query.

Auction houses

• Lot descriptions and catalogue copy drafted from specialist notes.
• Condition reports structured the same way every time.
• Consignor follow-ups before and after the sale.
• Research on artists, comparables and past sales.

Art shipping and transport

• Quotes drafted from the enquiry in minutes.
• Crate specs pulled from dimensions, medium and value.
• Customs and export paperwork checked for missing details before it goes.
• Client updates sent on time, in a consistent tone.

Art advisors

• Collector briefs that capture taste, budget and collection gaps.
• Market research on artists, galleries and recent sales.
• Acquisition memos drafted from your notes, ready to review.

Across every team

• Meeting notes from client calls, fairs and board meetings, turned into summaries and actions.
• Emails and routine admin.
• Reports and data. Sales, visitor and event figures turned into trends you can act on.
• Less manual PowerPoint and Excel.
• Scheduled briefings that land in your inbox without anyone asking.

One thing we're straight about. AI drafts, you sign off. A wrong provenance line is a reputational problem, and no tool changes that. We teach checking as part of every workflow.

This is a high-end St James's gallery, ahead of exhibiting at Frieze Masters. In October 2026 we ran a half-day session at the gallery with the director and the team, on their own laptops and their own material.

Before the day, everyone filled in a short form on their repetitive tasks. Answers covered collection and consignment notes, shipping documents and quotes, fair deadlines, loan letters, and losing track of clients met at fairs. The session was built from that list.

What we covered and built:

• Collector matching. Using ArtLogic contacts and artworks exports plus invoice history, Claude ranks the 10 best contacts for a specific work, with reasons for each. It weighs artist affinity, period, medium, size, price band, relationship, offer history and collection fit.
• Offer letters in the sender's own tone, drawing on the artwork details and each client's history. Short, in house voice, ready to check and send from Outlook.
• A shipping-note drafter. Trained on three past examples, it turns pasted ArtLogic details into a Word-ready note and flags any missing fields.
• Fair planning from the inbox. Claude pulled everything on the gallery's Frieze Masters stand out of email into one document: dates, pre-vetting, open bills, the insurance premium, the booth, who's doing what and key dates.
• Catching mistakes. On the same run it spotted a booth number that didn't match between the emails and an invoice.
• Inbox triage. A quick read of everything since a set date, what's unread and what needs a reply, then help drafting the replies.
• Meeting recall. Recorded client and fair conversations you can ask later: booth dimensions, names, prices. The phone app records in-person chats, or you capture your notes straight after a conversation at a fair.
• Researching a new buyer before the next conversation.
• Receipts into a spreadsheet. A folder of photographed fair receipts turned into an expenses sheet.
• Briefing by voice. Talking through a client meeting, the structure you want and where the figures live, then letting Claude build the proposal or deck.
• Context files. A personal Claude folder for each person, with an "About me" built by Claude interviewing them, a gallery context file, and tone-of-voice guides for collectors, trade and institutions, built from their own sent emails. Subfolders cover stock, clients, correspondence, fairs and projects.
• House rules built in. Plan before doing anything bigger than a letter. Never invent facts. Provenance and exhibition history only as recorded: flag any gap, never fill it.
• Skills shared across the gallery. Build once, publish to the team, and anyone runs it with the same result.
• Safe set-up. Training on your data switched off, and connectors set to ask before acting on anything sensitive.
• Scheduled tasks. A 9am daily calendar and inbox briefing, a weekly email digest and a monthly check on competitor galleries.

A live connection between ArtLogic and AI tools is in progress. Until then, we worked from ArtLogic's CSV exports of artworks, contacts, and sales and offers. They work very well. Refresh them weekly, or before a fair, and the workflows run on current data.

Pallant House Gallery in Chichester is home to one of the best collections of Modern British art in the country. We trained the team across sessions in November and December 2025, using ChatGPT.

What we covered:

• Meeting transcripts into summaries and action items for board, subcommittee and leadership meetings.
• Emails and routine admin.
• Donor communications, copywriting and marketing briefs.
• Trend analysis of admissions and ticketing, visitor numbers, events, exhibition cycles and commercial data, working on the team's own files.
• AI project maps for planning where there's no project management system.
• Less manual PowerPoint work.
• Gallery and artist research, plus deep-research competitor reports.
• Excel formulas written and explained.
• ChatGPT Projects for each business area, with their own instructions and files.
• Voice dictation to give the AI proper context, fast.
• A worked prompt for exhibition-launch emails.

1. Discovery first

A discovery call, or a short pre-session form for everyone attending. Five minutes each. It maps the repetitive tasks first, so the session is about your work, not a generic demo.

2. Pitched at the room's level

We work with people from complete beginners to daily power users. Sessions are pitched at where the room is, and split by level or team where it helps.

3. Context done properly

This is what makes the outputs usable. We set up:

• About you: role, clients, how you like things written.
• About the business: what you deal in, your artists, your clients, what happens at a fair.
• Tone of voice: pulled from your own sent emails, for collectors, trade and institutions.

4. Your own material

Bring real stock, real client lists, real correspondence and your ArtLogic exports. We build on them live. Everything we make is shared back to the team.

5. The right tools

Usually Claude Cowork as the workspace, Granola for meeting notes and Wispr Flow for voice. ChatGPT if that's what your team already uses. Everything runs on a business plan, so your data isn't used to train the models.

6. Afterwards

A build shortlist of everything worth automating, back within a week. A 30-day review call is included.