Research
What clients now tell their law firms about AI.
Two in every hundred outside-counsel guidelines on the open web mention AI. We searched 1,054 of them, written since ChatGPT and published by companies, insurers, banks, public bodies, and universities. Of the 20 that mention AI, 12 require the firm to disclose when it used it.

2%2% of the 1,054 outside-counsel guidelines written since ChatGPT and found on the open web mention AI, which comes to 20 documents.
1212 of the 20 require the firm to disclose when it used AI.
99 keep the client's information out of public AI tools.
88 require a lawyer to check the output.
99 change how AI work is billed.
The first AI clause is dated 2024.
2022
2023
2024
First mention of AI
2025
2026
No guideline dated before 2024 mentions AI. Then 2 of the 220 dated 2024 do, 8 of the 365 dated 2025, and 10 of the 261 dated 2026.
Get the next study by email
Our new studies by email: what we found and the data behind it.
Unsubscribe from any email. How we use your address
The 20 guidelines that mention AI, clause by clause.
| Guideline | Disclose AI use12 | Keep client data out of public tools9 | A lawyer checks the output8 | Ask before using it3 | Approved tools only1 | The firm has an AI policy3 | Banned outright1 | Bills AI work differently9 |
|---|---|---|---|---|---|---|---|---|
| A&K Travel Group (Abercrombie & Kent)companies, 2026 | ||||||||
| Afnicnonprofits, 2026 | ||||||||
| Commonwealth of Kentucky, Office of the Attorney Generalpublic bodies, 2026 | ||||||||
| Construction Industry Councilpublic bodies, 2026 | ||||||||
| Federal Deposit Insurance Corporation (FDIC)public bodies, 2026 | ||||||||
| San Francisco Unified School Districtpublic bodies, 2026 | ||||||||
| St. Joseph Countypublic bodies, 2026 | ||||||||
| The Hartford Financial Services Groupinsurers, 2026 | ||||||||
| UBS Group AGbanks, 2026 | ||||||||
| University of Missouri System, Office of General Counseluniversities, 2026 | ||||||||
| Arch Insurance Group Inc. (Arch Capital Group Ltd.)insurers, 2025 | ||||||||
| Foskorpublic bodies, 2025 | ||||||||
| NSW Government Legal Services Panel (Transport for NSW as Lead Agency on behalf of the State of NSW)public bodies, 2025 | ||||||||
| New York State Department of Transportationpublic bodies, 2025 | ||||||||
| Office of the Attorney General of Texas, General Counsel Divisionpublic bodies, 2025 | ||||||||
| Official Solicitor and Public Trusteepublic bodies, 2025 | ||||||||
| Public Utility Commission of Texaspublic bodies, 2025 | ||||||||
| Zscaler, Inc.companies, 2025 | ||||||||
| Microsoftcompanies, 2024 | ||||||||
| University of California, Office of the General Counseluniversities, 2024 |
Keeping client data out of public tools also depends on the software a firm runs. In our study of nearly 3,700 software products, 87% of terms say nothing about training AI on customers' data. Only 4 of 29 legal AI vendors put a full promise not to train on client data in their contract.
So what
What the 20 ask for.
For a company
98% of the 1,054 guidelines say nothing about AI. A firm working under one of those has no instruction from its client and sets its own rules for AI on the client's matters.
Of the 20 that do say something, 12 want to be told when AI was used, 9 keep the company's information out of public tools, 8 want a lawyer to check the output, and 2 refuse to pay for time spent using or learning the tools.
For a law firm
Clients that add an AI section to their guidelines borrow from the ones already published, and these 20 are the published ones we could find.
One of the 20 bans AI on its work, 3 want to be asked before it is used, 3 want the firm's own AI policy on file, 1 limits the firm to approved tools, and 9 change what they pay for the tool or the time.
What firms publish about their own use of AI, including the 77 AI policies on their websites, is in Lawyers using AI.
In their own words.
Disclose AI use
“If a timekeeper uses generative AI for a discrete task, they should note that in their time entry.”
Keep client data out of public tools
“Never enter Sensitive Information or FDIC work product into a GenAI tool that operates in an "open environment."”
A lawyer checks the output
“Any work product created by a Large Language Model (LLM) or other generative AI tool must be reviewed by a human attorney.”
Ask before using it
“Outside Counsel will not use firm-developed or third-party generative AI tools or any other AI tool to conduct legal research, draft or revise court documents, or produce Client Deliverables without Client’s prior approval in writing. (Section 2.3. Use of Artificial Intelligence (“AI”).)”
Billing
“Any time and cost associated with AI-generated work product shall not be passed on to Zscaler.”
Help with your own AI clause.
The 20 above are the AI clauses we found in public. We help companies write their own, in rules their law firms can follow, in the Diagnosis, and we build the systems that put those rules to work in the Build. For the firms on the other side of the guidelines, there is AI consulting for law firms.
The data.
The guidelines we searched, one organization per row, with the clauses and the document's own words in the CSV.
Download the data (CSV)How we did it
Outside-counsel guidelines are the rules a client gives the law firms it hires. They cover billing, staffing, confidentiality, and reporting, and some now cover AI. Most organizations hand them to their firms through a portal and never publish them. The organizations that do publish are mostly public bodies and universities, and the sample shows it. 86% of the 1,054 come from them and 24 are companies. We gathered the documents on September 15 and 16, 2026, through search engines, sweeps of company, insurer, public-body, and university websites, procurement portals, and court dockets where guidelines had been filed as exhibits. A document qualified if it governs an organization's use of outside law firms, whether it is written to the firms, like billing guidelines, or to the organization's own staff, like the legal-services policies most universities outside the United States publish. It is a sample, limited to what search engines and procurement portals surface.
We kept only guidelines written after ChatGPT was released. A document counted when it carried a date of December 2022 or later in its own text, in its file's metadata, in the page's tags, or in the address it was published under. Among the documents we read in full, we left out 55 older guidelines and 15 that could not be dated. At the screening stage we set aside 1,001 more because they carried a date before December 2022 and 145 because we could not date them. 12 organizations publish a page with no date on it. Where that page is the current version on the organization's own site, we counted it. That left 1,054 documents from 1,054 organizations: 832 from public bodies, 74 from universities, 33 from nonprofits, 25 from health systems, 24 from companies, 21 from pension funds, 20 from banks, 15 from insurers, 10 from others.
We searched every document in full for the words of AI in the document's own language, whether that was English, Italian, Polish, or Vietnamese. The words were artificial intelligence, generative AI, machine learning, large language models, ChatGPT, Copilot, and their variants, and scanned files went through OCR first. 254 of the 1,054, including every document where one of those words appeared, were then read in full and coded against a fixed rubric. The rubric asks whether the document mentions AI and, if it does, whether it requires the firm to disclose AI use, to get consent first, to keep the client's information out of public tools, to have a lawyer check the output, to use approved tools only, to hold an AI policy of its own, or to bill differently. A clause counted only when the document says it, and every yes comes with the document's own words, which are in the CSV. A ban that lifts on consent is coded as consent, and a rule that the firm must get approval before using AI is coded as disclosure too, since the firm has to say so first. Where an organization publishes several guideline pages, we read them together and counted the organization once. 2 documents could not be read: Bayer AG (the site blocked automated requests), and Gloucester, Salem, Cumberland Counties Municipal Joint Insurance Fund (TRICO JIF) (the published copy is missing half its pages).
Dates come from the documents. 1,042 of the 1,054 have one. The 12 without a date are live pages on their organizations' own sites. They count in the study and appear in the table, and we left them off the timeline.
The data is the table below and the CSV, released under CC BY 4.0 with attribution to Fulkerson Advisors. If your organization publishes guidelines we missed, email info@fulkersonadvisors.com and we'll add them.
The 1,054 organizations
- public bodies
- 832
- universities
- 74
- nonprofits
- 33
- health systems
- 25
- companies
- 24
- pension funds
- 21
- banks
- 20
- insurers
- 15
- others
- 10
Cite this study
Fulkerson Advisors (2026). What Clients Now Tell Their Law Firms About AI: 1,054 Outside-Counsel Guidelines Searched. Read September 16, 2026. https://www.fulkersonadvisors.com/research/ai-in-outside-counsel-guidelines
Use any of these numbers or quotes. Credit Fulkerson Advisors and link to this page. The data is under a CC BY 4.0 license.
The data is also on Zenodo, DOI 10.5281/zenodo.23320135.
Questions
Do companies require their law firms to disclose AI use?
The ones with written AI rules do, and they are 2% of the sample. Of the 1,054 outside-counsel guidelines written since ChatGPT that we searched, 20 mention AI, and 12 of those require the firm to disclose it, usually in the time entry or on the invoice.
Do clients refuse to pay for AI work?
Of the 20 guidelines that mention AI, 2 refuse to pay for time spent using or learning AI tools, 2 refuse to pay for AI subscriptions, and 3 expect the savings passed on as fewer hours.
Which organizations have AI rules for outside counsel?
Among the 1,054 we searched as of September 16, 2026: A&K Travel Group (2026), Afnic (2026), Commonwealth of Kentucky (2026), Construction Industry Council (2026), Federal Deposit Insurance Corporation (2026), San Francisco Unified School District (2026), St. Joseph County (2026), The Hartford Financial Services Group (2026), UBS Group AG (2026), University of Missouri System (2026), Arch Insurance Group Inc. (2025), Foskor (2025), NSW Government Legal Services Panel (2025), New York State Department of Transportation (2025), Office of the Attorney General of Texas, General Counsel Division (2025), Official Solicitor and Public Trustee (2025), Public Utility Commission of Texas (2025), Zscaler, Inc. (2025), Microsoft (2024), and University of California (2024).
Do any clients ban AI outright?
One of the 20 bans it outright, and 3 require the firm to ask first.