AI buyers license records from marketing, creative and digital agencies because agency work is a written record of judgment: the brief, the first idea, the client's feedback, the revision and the result. The sourced figures are a $110,000 perpetual exclusive license with a global marketing agency (Polyshares case study, October 2026) and Mercor's design and web agency listing of $20,000 to $890,000 (Mercor Data, October 2026). The deciding question is rights: much of what sits in your folders was made for clients, under contracts that may give it to them.
On this page
- Why do AI labs want agency records?
- Which agency records do buyers want?
- What have agencies been paid?
- What might a 20 to 500 person agency expect?
- Who owns agency work: you or the client?
- What about stock photos, fonts, music and talent?
- What should an agency leave out?
- How is agency data de-identified?
- Questions
- Sources
Why do AI labs want agency records?
AI models are being trained to write briefs, draft campaigns, design pages and manage client projects. Finished ads are already public. What is rare is the process behind them: why a concept was killed, how a client's notes changed a layout, which ad variant won and what the team did next.
Buyers name agencies directly. Scale AI lists marketing agencies among its industries of interest (Scale AI, October 2026). AfterQuery lists campaigns among the retail, e-commerce and marketing records it licenses (AfterQuery, October 2026). Business Insider (October 2026) reported that the agency MyPortal Marketing sold micro1 "everything from Figma and Canva files to checklists for onboarding clients to Google ads," and that the owner said it did not sell client-specific material.
See Polyshares, Mercor, Scale AI, AfterQuery and micro1 for their terms. Data Licensors, the company that publishes this page, also licenses records from agencies. Our buyer page sits next to the others.
Which agency records do buyers want?
| Area | Records |
|---|---|
| New business | Pitch decks, proposals, scopes of work, won and lost reasons |
| Strategy | Briefs, research summaries, personas, media plans |
| Creative process | Concept rounds, feedback threads, revision history in Figma and other design files, review comments |
| Production | Design systems, website builds with their code, video edit decisions, asset versions |
| Delivery | Project plans and tasks in Asana or Monday, timelines, change requests, QA checklists |
| Performance | Campaign reports, ad copy variants with results, test learnings |
| Operations | SOPs, onboarding checklists, templates, time tracking, internal Slack about how work gets done |
Mercor's design and web agency listing asks for "Figma files and production code for the same projects" and a portfolio of shipped work over 3 or more years, and says template or AI-generated sites are not a fit (Mercor Data, October 2026). Revision history matters most: it is the reasoning a buyer is paying for.
What have agencies been paid?
| Company | Records | Amount | Kind of figure | Source |
|---|---|---|---|---|
| Global marketing agency | Full operating record, perpetual exclusive license | $110,000 | Buyer case study | Polyshares, October 2026 |
| Marketing agency (MyPortal Marketing) | Figma and Canva files, onboarding checklists, Google ads, no client-specific material | Not reported | Press report | Business Insider, October 2026 |
Polyshares does not name the agency, and the figure is the buyer's own account. A perpetual exclusive license gives up more than a time-limited, non-exclusive one, so compare terms, not just the number. Both are in our deal ledger.
What might a 20 to 500 person agency expect?
Sourced ranges:
- Mercor, design and web agency: $20,000 to $890,000 for 10 to 200 employees, "depends on archive size and rights" (Mercor Data, October 2026).
- Mercor, consumer brand or photo studio: $30,000 to $800,000 for 10 to 500 employees, for original product, apparel and campaign imagery the company owns (Mercor Data, October 2026).
- micro1, any industry: tiers of $100K to $500K, $500K to $2M and $2M+, for 30 or more full-time staff (micro1, October 2026).
Mercor's own wording points at the main driver: rights. An agency that can license its full process history, with client consent where needed, sits much higher than one limited to templates and SOPs. Get an estimate, and see how much company data is worth for every buyer's range.
Who owns agency work: you or the client?
Read your client agreements before anything else. Common terms:
- Deliverables assigned to the client. Many master service agreements make final work the client's property once paid, sometimes as work made for hire.
- Agency tools kept by the agency. Many agreements let the agency keep its pre-existing materials, methods and templates.
- Confidential information. Most agreements limit use of client information to performing the services.
- Client data. Customer lists, analytics and ad accounts the client gave you access to are the client's, not yours.
Polyshares puts it plainly: material governed by a client master agreement "belongs to the client regardless of value" (Polyshares, October 2026). Mercor's listing asks for "the right to license the work, either your own or cleared with clients" (Mercor Data, October 2026).
A practical sort:
- Yours: SOPs, templates, onboarding checklists, internal chat about process, project management history, your own brand work, unpitched or unsold concepts where no contract applies.
- Ask the client: briefs, revisions, finished deliverables, performance reports. Some clients agree once their name and brand are removed, but logos and distinctive brand assets can identify a client even without a name.
- Leave out: anything a contract forbids, client customer data, and third-party licensed assets.
This is general information, not legal advice. Who owns client work depends on each contract, so have a lawyer read the ones that matter.
What about stock photos, fonts, music and talent?
Agency files are full of things you licensed from someone else. Getty Images' license says that, unless explicitly authorized, content and its metadata may not be used "for any machine learning and/or artificial intelligence purposes" (Getty Images, checked October 2026). Check your other stock, font and music licenses the same way.
People in photos and videos are another layer. Model and talent releases often cover a campaign, not AI training, and Getty itself says it does not warrant consent for such uses of model-released content. Leave out footage and photos of identifiable people unless the release allows it, and leave out client staff and customers who appear in content.
What should an agency leave out?
- Client customer lists, CRM exports and ad audience data.
- Ad account and analytics logins, API keys and tokens, often pasted into docs and chat.
- Third-party licensed assets: stock, fonts, music, talent footage.
- HR, payroll and freelancer contracts.
- Shared Slack or Teams channels with clients, unless the client agrees. See our Slack guide.
Records you do not have the right to license are on our never accept list.
How is agency data de-identified?
Client and brand names, people's names, emails, phone numbers and addresses become placeholders such as [Client] or [Account lead]. Logos, product shots and other brand marks can identify a client even without a name, so they need to be removed, or the file left out. Feedback, revision order, dates and results stay, because that is what an AI system learns from. You see a de-identified sample before you sign. More: how de-identification works, plus our guides on licensing agreements and whether selling company data is legal.
