Andrea Peters, Senior Counsel and Global Head of Compliance at Interface, started using generative AI for legal work and described the moment it changed her practice:
"Easy Prompt is the best thing that's happened to generative AI for lawyers."
Before, Andrea described the bind in her words:
"Being a strategic thinker takes time and mental headspace, and when you've got a million tasks and deadlines, it can be hard to find that time."
After, her workflow flipped:
"It's much more natural now for me to think, how can I use GC AI for this? I don't have to stop and decide what's AI-appropriate. It's automatic."
Andrea's path is becoming the default for in-house counsel today.
According to the 2026 ACC Chief Legal Officers Survey, 52% of US in-house counsel started using generative AI tools in 2025, more than double the year before.
1,700+ in-house legal teams across 53 countries now run their daily legal work inside GC AI, including legal departments at SKIMS, Riot Games, Gusto, and Eventbrite, alongside more than 80 public companies and 25 unicorns spanning consumer brands, gaming, fintech, fashion, and SaaS.
The question has shifted from whether to adopt generative AI in legal departments to which platform to pick and how to evaluate it the way a GC evaluates outside counsel.
Trust, precision, and craft are the standards in-house counsel hold themselves to, and the generative AI legal tools that meet that bar earn permanent space in the workflow.
We have heard the same pattern across more than 30 conversations on the CZ and Friends podcast, GC AI's podcast hosted by co-founder and three-time former general counsel Cecilia Ziniti.
What Generative AI for Legal Means
Generative AI for legal is AI that reads, drafts, analyzes and reviews legal documents using large language models trained for legal-domain work, with the controls licensed attorneys need: character-locked citations, confidentiality guarantees, and outputs the lawyer can verify before filing.
The category sits at the intersection of two things that did not exist three years ago. A language model can read a 200-page agreement in one pass. Legal-specific infrastructure (playbooks, citation grounding, Microsoft Word integration) turns the chat interface into something a GC can put on a client matter.
The underlying fit is natural. Lawyers work in words all day, and large language models are built on words, so the core skill transfers. The legal-specific layer supplies everything else. GC AI's system prompt runs more than 20,000 lines of legal context, tool calls, and in-house counsel framing layered on top of the foundation models. That layer carries the legal voice, the in-house framing, and the workflow patterns that turn a foundation model into a working partner for a GC.
Three traits distinguish the platforms in-house legal teams pay for: a system prompt built for legal voice, character-level citations the lawyer can click through, and zero data retention agreements with the underlying model providers.
The taxonomy reads like this:
Generative AI is the technology layer (large language models from OpenAI, Anthropic, and Google).
Generative AI for legal is the application layer (retrieval-augmented generation, legal corpora, citation grounding, and the legal-voice system prompt that sits on top).
Legal AI is the broader category that includes generative AI alongside older techniques like supervised classification on contract clauses and rule-based document automation.
How Generative AI Works in a Legal Workflow
Generative AI works in a legal workflow by combining three technical pieces: a large language model that handles language, a retrieval-augmented generation (RAG) layer that grounds the model's responses in your documents and verified legal sources, and an interface (chat, Word add-in, or sidebar) where the lawyer reads, edits, and accepts the output.
The first piece is the foundation model.
GC AI works with OpenAI and Anthropic models under zero data retention agreements, which means neither provider trains on or stores the inputs your lawyers send. The model handles the language: it reads a contract, drafts a memo response, summarizes a deposition.
The second piece is RAG, retrieval-augmented generation.
Asking the model to recall a clause from memory creates hallucination risk. RAG retrieves the actual passage from the document the lawyer uploaded, hands it to the model along with the question, and asks the model to answer based on the retrieved text. The citation links back to the exact passage. The lawyer verifies in one click.
The third piece is the interface.
An interface is the surface where the lawyer works with the AI: the chat window, the Microsoft Word add-in, or the sidebar panel that opens next to the document. It is the difference between a tool the team opens once and forgets and a tool the team works inside all day.
The lawyer's actual writing surface is Microsoft Word. A generative AI platform that lives outside Word (browser-only, chat-only, separate web app) adds a context-switch tax to every draft. The platforms built for in-house teams install as a Word add-in and read the document open on screen.
For a look at the workflow in practice, the GC AI for Word walkthrough below shows the sidebar reading a counterparty agreement and returning a structured analysis tied to the source language.
Cameron Clark, Head of Legal at Arc'teryx, described how the combination changed the daily work of a one-lawyer in-house team:
"It's like having a senior legal peer to think with. It helps me brainstorm and refine ideas in real time, which makes a huge difference when you're leading a small team."
The model plus retrieval plus Word integration is the reason the workflow lands.
Five Jobs Generative AI Handles for In-House Counsel Today
The five highest-leverage workflows in-house counsel run through generative AI today, ranked by adoption velocity across GC AI customer teams:
Reviewing and redlining commercial contracts
Drafting internal memos and BoD-prep briefs
Researching jurisdictional and regulatory questions
Drafting diplomatic internal communications
Building team knowledge from prior work product
Reviewing and Redlining Commercial Contracts
The fastest-ROI use case. The lawyer opens a counterparty's MSA, applies a Playbook the team encoded for sell-side agreements, and gets back a redline that reflects the team's position on indemnity, limitation of liability, IP, and warranties.
Hayley McAllister, Senior Counsel and Head of Commercial Legal at Jasper, described the compression in her workflow:
"What used to take me an hour now takes me 10 minutes."
For the deeper read on this workflow, see AI Contract Review and Contract Redlining Software.
Drafting Internal Memos and BoD-Prep Briefs
The second-highest-leverage workflow. The lawyer fields a question from the business (HIPAA, contractor classification, privacy review), drafts a short answer, and files it for the next time the question comes up.
KT Farley, Chief Privacy Officer and Associate General Counsel at Helix, described what this looks like in practice:
"A partner asked me to quickly draft a response to a HIPAA compliance question. Usually this would take me an hour, switching context, creating a doc, writing it up. With GC AI it was much faster."
Researching Jurisdictional and Regulatory Questions
The compounding workflow. Generative AI handles the first pass on jurisdictional research, statutory interpretation, and regulatory monitoring. Tools with character-locked citations to primary law surfaces (statutes, regulations, case law) are worth the extra evaluation here.
The lawyer reads the source-grounded answer, verifies the citation against the primary source, and uses the output as a starting framework the team builds on.
Kacie Zanassi, Director of Employment, Litigation, and Legal Ops at Eventbrite, described how generative AI changed her first move on a jurisdictional question:
"When facing litigation in unfamiliar jurisdictions, I use GC AI as my first step to quickly understand procedural requirements, causes of action, and local court rules."
For the workflow deep dive, see AI for Legal Research, and for the ranked options, see the best AI tools for legal research.
Drafting Diplomatic Internal Communications
The hidden hero of in-house AI use. Sales emailed a customer something that contradicts the company's terms. Procurement asked for an indemnity carve-out the team cannot accept. The AI drafts a four-sentence note that lands the legal answer while keeping the relationship intact.
Alexandra Sepulveda, Assistant General Counsel at Trust & Will, described what this looked like for her team:
"Say sales emailed a customer something that conflicts with our online terms. I'll ask GC AI to draft a diplomatic note to the sales team explaining the impact, how to fix it, and how to avoid it next time."
Building Team Knowledge From Prior Work Product
A generative AI platform that retains the team's prior work (memos, redlines, playbook updates, training material) becomes a searchable institutional memory the team queries. The next time the same compliance question comes up, the answer is already there.
Andrea Peters described the team-knowledge layer this way:
"Instead of re-doing research or writing long explanations, I can just share the prompt I used. It saves all of us time and gets us aligned quickly."
The Three Risks That Decide Whether Generative AI Is Safe for Legal Work
The three risks every in-house counsel evaluates before greenlighting generative AI for client work are hallucinated citations, confidentiality and privilege exposure, and training defaults that send privileged inputs back into the model.
ABA Formal Opinion 512 (issued July 2024) sets the floor on the lawyer's competence, confidentiality, and verification duties when using generative AI.
Hallucinated Citations
Generative AI tools without character-locked citations hallucinate cases confidently. The 2023 Mata v. Avianca sanctions order remains the cautionary citation. The lawyer asked the AI to generate citations for a brief. The AI invented six cases that did not exist. The court sanctioned the lawyer who filed the brief.
Exact Quote is the fix. It locks each citation to the source document or primary-law passage, so the lawyer can click through and verify it. Every AI-drafted citation still requires a human pass before filing.
Confidentiality and Privilege
When a lawyer pastes a settlement memo into a consumer chat tool, the inputs may be retained by the vendor or used to train future models. That can be a confidentiality breach under the Rules of Professional Conduct, depending on the firm's data-handling representations.
Generative AI built for legal work ships with the security posture confidential work requires: zero data retention agreements with its model providers, SOC 2 Type II and SOC 3 certification, GDPR compliance, and AES-256 encryption. GC AI's security page lays out the certifications, the encryption standards, and the data-handling commitments in full. The buyer's checklist below covers the questions to ask the vendor before the platform sees a single document.
Training Defaults and Ad Targeting
OpenAI's consumer ChatGPT Free and Go tiers train on user inputs by default, as of May 2026, per OpenAI's chat retention policy. In February 2026, OpenAI began testing advertising on those tiers. Claude's free tier carries similar retention and training defaults, as of May 2026; the Claude Legal AI review covers where the model fits in-house legal work.
When the use case is confidential client work, the training defaults and the retention policy are worth evaluating before the platform sees a single document.
For the deeper read on consumer AI privacy, see Is ChatGPT Private? and Is ChatGPT Confidential?.
How to Evaluate Generative AI for Legal Work
The six capabilities below separate the generative AI platforms worth evaluating from the ones built for a different buyer. The closing question in each subsection is the one to ask vendors during a demo. For a category-by-category breakdown of the platforms on the market, see the Legal AI Tools field guide.
Word-Native Integration Without Context Switching
The lawyer's actual writing surface is Microsoft Word. A platform that requires the lawyer to copy text out, paste it into a web app, run a prompt, and copy the result back creates a 30-second tax on every draft. A 30-second tax on each of 40 contracts a month adds up to 20 minutes of pure context-switching tax. Across a year, that tax is the reason an unused license sits in a CFO's expense review.
Ask vendors:
Does the tool install as a Microsoft Word add-in, can it read the document open on screen, and does it accept and reject tracked changes the way Word natively does?
Character-Locked Citations
When the AI says "Section 7.2 caps the liability at the lower of fees paid and $1M," the lawyer needs to know the dollar figure came from the document and the model did not pattern-match on similar contracts.
Ask vendors:
How does the platform ground its citations, does the citation link back to the exact passage, and what happens when the answer sits in a hyperlinked source the document points to?
Pre-Built Playbooks Your Team Can Adapt
A generative AI first draft is useful only if it reflects the positions the team takes on indemnity, limitation of liability, and IP ownership. Platforms worth buying ship pre-built playbooks for NDAs, DPAs, and MSAs that your team adapts to your standards without an implementation engineer.
Ask vendors:
Which contract types ship with playbooks on day one, can your team build and edit playbooks without vendor support, and how do playbook updates propagate when the team's position changes?
Security Designed for Confidential Legal Work
In-house counsel writes about active litigation, M&A targets, employee disputes, and product launches that have not been announced. The platform needs SOC 2 Type II at minimum, zero data retention agreements with its model providers, and encryption that meets the standard a Fortune 500 CISO will sign off on.
Ask vendors:
Which LLM providers does the platform use, are zero data retention contracts in place for each, where is data stored, and is SOC 3 (the public-facing security report) available in addition to SOC 2?
Training That Turns the Team Into Power Users
The teams who report saving an average of 14 hours per lawyer per week in GC AI's December 2025 ROI study did not get there by buying a license and walking away. They invested in classes, prompt libraries, and shared playbooks.
Ask vendors:
Is formal training included with the license, are classes CLE-eligible, who teaches them, and does the curriculum map to the workflows your team runs every week?
A Track Record With In-House Legal Teams
The buyer profile decides the product. A platform built for AmLaw 100 partners makes different design tradeoffs than a platform built for the GC who runs legal as a business unit inside a company. The question is which set of tradeoffs matches the team buying the license.
Ask vendors:
How many in-house legal teams use the platform, can the vendor name customers in your size and industry, and what does a typical in-house implementation look like in the first 30 days?
How Generative AI Platforms Score on In-House Legal Work
GC AI's R&D team built the In-House Legal Bench to measure how well AI handles the day-to-day work of in-house counsel.
The benchmark scores each platform against 100 tasks drawn from real in-house workflows, spanning 10 categories: drafting, summarizing documents, contract analysis, legal research, legal strategy, risk assessment, comparison and benchmarking, extracting information and data, regulatory tracking, and checklists.
Six GC AI R&D attorneys with more than 80 combined years at leading companies and law firms wrote and vetted every task and answer key.
Each task is graded against a structured answer key averaging 12 pass/fail criteria, more than 1,200 criteria across the benchmark, covering legal accuracy, sourcing, structure, and tone. An LLM judge scored every response, and the R&D team validated a sample against human expert review.
GC AI posted the top pass rate in every one of the 10 task categories, leading the closest competitor, ChatGPT, by 7 percentage points overall. The full results, as of the May 2026 In-House Legal Bench:
Legal task category | Tasks + Example | GC AI | ChatGPT (GPT-5.5) | Claude (Opus 4.7) | Gemini (3.1 Pro) |
|---|---|---|---|---|---|
Drafting | 19. Drafting a jurisdiction-compliant return-to-office policy | 87.6% | 83.4% | 74.9% | 66.4% |
Summarizing documents | 12. Producing an executive briefing of a recent trial opinion | 81.6% | 77.5% | 63.7% | 57.5% |
Contract analysis | 13. Explaining an IP license agreement's scope and restrictions in plain English | 82.7% | 72.8% | 66.3% | 42.9% |
Legal research | 23. Describing SEC Schedule 13D beneficial ownership reporting requirements | 88.3% | 75.6% | 66.2% | 61.7% |
Legal strategy | 16. Assessing CPSC reporting obligations and recall options for a smart home device | 86.3% | 84.5% | 63.0% | 58.0% |
Risk assessment | 26. Identifying risks in an arbitration clause with no IP carve-outs | 89.0% | 84.2% | 71.1% | 59.2% |
Comparison and benchmarking | 9. Comparing supplier codes of conduct across direct competitors | 91.4% | 84.7% | 81.4% | 72.9% |
Extracting information and data | 24. Extracting executive compensation data from a proxy statement into a table | 82.0% | 76.9% | 57.0% | 56.3% |
Regulatory tracking | 11. Mapping federal and state consumer protection rules into a compliance chart | 88.6% | 73.5% | 68.2% | 45.0% |
Checklists | 13. Producing a GDPR compliance checklist for a SaaS company | 89.9% | 81.9% | 73.4% | 59.3% |
Overall pass rate | All 100 tasks | 86.8% | 79.8% | 68.4% | 57.5% |
The pattern breaks down in three bands:
Widest margin: research-intensive work. On regulatory tracking, legal research, and checklists, GC AI's lead was largest. These tasks require locating current requirements across jurisdictions and grounding them in authoritative government, court, and regulatory sources, the point where general-purpose models most often produce reasonable-sounding but unsourced analysis.
Meaningful margin: drafting and contract analysis. ChatGPT performed comparably on issue-spotting. GC AI showed clear advantages in legal accuracy, in extracting quoted text from documents, and in the quality of the drafted language.
Narrowest margin: legal strategy and risk assessment. ChatGPT came closest to GC AI on these tasks, though GC AI still scored higher on the conciseness and practicality of its responses. Claude and Gemini trailed both by wider margins.
For a buyer evaluating platforms, the signal is consistent. As foundation models improve, the baseline for legal reasoning rises across every platform. The advantage that holds is in finding the right authoritative sources, handling legal documents precisely, and presenting work the way a lawyer expects to read it.
How GC AI Approaches Generative AI for In-House Counsel
GC AI is the legal AI platform purpose-built for in-house legal teams, with a 20,000-line legal system prompt, a native Microsoft Word integration that drafts inside the document on screen, pre-built Playbooks for NDAs, DPAs, and MSAs, Exact Quote for character-level citations, Research for multi-agent legal intelligence with primary-law sources, and a Skill Library of ready-to-use workflows.
The founding story explains the product wedge. Cecilia Ziniti was a general counsel three times (Anki, Bloomtech, Replit) and an in-house lawyer at Amazon and Cruise before that.
She built GC AI to solve the problems she encountered firsthand as an in-house lawyer running compliance, commercial deals, employment matters, and board reporting under deadline. That experience is embedded directly into the system prompt, the tone, and the Word-first workflows.
Tiffany Lee, General Counsel at Liquid Death, framed the ROI lens this way:
"I buy GC AI for my contract attorneys, not to reduce their hours, but to take away mundane work so I can use their brainpower on strategic matters."
Classes: GC AI Classes ship with the license. 6,000+ lawyers have taken them (as of May 2026). The sessions are California CLE-eligible and taught by former general counsels.
Generative AI for legal has moved from frontier technology to operating layer. In-house teams write contracts, draft memos, run research, and counsel their business partners inside it every day. The decision every GC faces in 2026 is which platform to pick and how fast to roll it out.
Start this month by picking one high-leverage workflow (contract redlining or vendor NDA review for most teams), encoding one playbook against the team's standards, and putting the whole legal team through a GC AI 101 class in week one.
Layer in memo drafting and internal communications in month two. Add research and team-wide playbook authoring in month three. By day 90, the team is reporting hours saved and reduced reliance on outside counsel.
GC AI is trust, precision, and craft for the in-house counsel who runs legal as a business unit, with the playbooks, citations, Word integration, and CLE-eligible training that turn generative AI from a chat box into a daily working partner.
For the head-to-head reads on the in-house wedge competitors, see GC AI vs Spellbook and GC AI vs Harvey. For the broader ranked guide, see Best Legal AI Tools for In-House Counsel.
FAQ
What is generative AI for legal?
Generative AI for legal is AI that reads, drafts, and analyzes legal documents using large language models trained for legal-domain work, with the controls licensed attorneys need: character-locked citations, confidentiality guarantees, and outputs the lawyer can verify before filing. The category covers contract drafting, redlining, memo writing, research, and counseling workflows.
How is generative AI different from regular AI for lawyers?
Generative AI creates new text (a redlined contract, a draft memo, a research summary) by working from a large language model. Older legal AI techniques include supervised contract clause classification, e-discovery predictive coding, and rule-based document automation. Generative AI subsumes most of those use cases when paired with retrieval-augmented generation and character-locked citations.
Is generative AI safe for confidential legal work?
Yes, on platforms with the right security posture. A safe platform has SOC 2 Type II certification at minimum, zero data retention agreements with its model providers, and encryption that meets enterprise security review. GC AI is SOC 2 Type II and SOC 3 certified, GDPR compliant, with zero data retention agreements with OpenAI and Anthropic, and AES-256 encryption.
Can generative AI replace in-house counsel?
No. Generative AI accelerates the first-draft and review stages of legal writing. Legal judgment, client counseling, strategy, board-of-directors communication, and responsibility for filings stay with the lawyer. The teams in GC AI's December 2025 ROI study, who reported saving an average of 14 hours per lawyer per week, describe generative AI as a force multiplier on the legal expertise that produces the final work product.
What are the biggest legal issues with generative AI?
The biggest legal issues are hallucinated citations (addressed by character-locked citation features and human verification), confidentiality and privilege exposure (addressed by zero data retention agreements and enterprise security controls), and training defaults that send privileged inputs back into the model (addressed by selecting platforms with verified ZDR contracts). ABA Formal Opinion 512 (July 2024) sets the floor for attorney competence, confidentiality, and verification duties when using these tools.
How is generative AI legal tech different for in-house counsel?
Generative AI legal tech built for in-house counsel is purpose-built for the buyer who runs legal as a business unit inside a company. Tools built for law firms optimize for AmLaw 100 use cases: six- and seven-figure deals, dedicated legal innovation teams, custom implementation engagements. In-house tools optimize for the configurations in-house teams run (the solo GC, the GC with a paralegal, the two-counsel team) and ship with playbooks, Word integration, and a CLE-eligible class library on day one.
How much does generative AI for legal cost as of May 2026?
Pricing ranges from about $20 per month for consumer AI tiers like ChatGPT Plus and Claude Pro up to $500 per seat per month for purpose-built legal AI, as of May 2026. GC AI is $500 per seat per month with a 14-day free trial. Consumer AI starts at $20 per month for ChatGPT Plus or Claude Pro (as of May 2026). Purpose-built legal AI costs more because it adds citation grounding, confidentiality controls, and legal-domain calibration on top of the underlying model. Harvey pricing is not publicly disclosed as of May 2026.
Where do I start with generative AI for legal?
Start with one high-leverage workflow the team runs every week, usually contract redlining or vendor NDA review. Pick one platform, run a 14-day pilot inside Microsoft Word, and measure time-to-first-draft against your current baseline. GC AI's free 14-day trial covers the full platform including Playbooks, Exact Quote, Research, and the CLE-eligible class library. Start the trial at auth.gc.ai/sign-up.
Will generative AI replace law firms for in-house work?
No, but it changes the mix. Generative AI is shifting work in-house that historically went to outside counsel: first-pass contract review, jurisdictional research, memo drafting, and counseling on routine questions. Outside counsel work that requires deep specialization (M&A diligence, betting-the-company litigation, regulatory enforcement defense) stays at law firms. The 14% median outside counsel spend reduction GC AI customers report in the December 2025 ROI study reflects the shift in routine work, not a replacement of the law firm relationship.
