LegalGPT
Last reviewed
May 11, 2026
Sources
15 citations
Review status
Source-backed
Revision
v4 · 2,197 words
Improve this article
Add missing citations, update stale details, or suggest a clearer explanation.
Suggest edit
CtrlK
Last reviewed
May 11, 2026
Sources
15 citations
Review status
Source-backed
Revision
v4 · 2,197 words
Add missing citations, update stale details, or suggest a clearer explanation.
| LegalGPT | |
|---|---|
 | |
| Information | |
| Name | LegalGPT |
| Platform | ChatGPT |
| Store | GPT Store |
| Model | GPT-4 |
| Category | Research & Analysis |
| Description | Specialized in legal matters, this GPT could assist lawyers and legal professionals with case research, legal documentation, and even help in predicting case outcomes based on historical data. |
| Developer | RAHUL PORWAL |
| OpenAI URL | https://chat.openai.com//g/g-xck3iENsZ-legalgpt |
| Chats | 6,000 |
| Web Browsing | Yes |
| DALL·E Image Generation | Yes |
| Free | Yes |
| Available | Yes |
| Updated | 2024-01-24 |
LegalGPT is a Custom GPT for ChatGPT in the GPT Store. Built by Rahul Porwal on OpenAI's GPT-4 model, it is positioned as a research and analysis assistant for lawyers, paralegals, students, and clients who want help interpreting legal documents, drafting contracts, or exploring how prior cases might inform a strategy. It is one of many legally focused GPTs that appeared after OpenAI opened the directory to third-party builders, alongside higher-profile commercial systems such as Harvey and Thomson Reuters' CoCounsel.
LegalGPT runs inside the standard ChatGPT interface at chat.openai.com/g/g-xck3iENsZ-legalgpt. Like every Custom GPT, it is a thin configuration layer on the base model: a system prompt, four conversation starters, and toggles enabling web browsing, the DALL-E image tool, and the code interpreter. The GPT has no uploaded knowledge files, so its outputs draw on training data plus live browsing.
Third-party indexes confirm the listing. AIPRM records the GPT under ID g-xck3iENsZ with Rahul Porwal as builder. The plugin.surf entry says the GPT was first added on 15 November 2023 and last updated on 29 March 2024. The getgptstore.com page lists the same developer and description. LegalGPT is free for any ChatGPT Plus, Team, or Enterprise subscriber, since the GPT Store is gated behind a paid plan rather than a per-GPT charge.
Rahul Porwal is listed as the builder on every directory that indexes the GPT. He is not a verified OpenAI partner. OpenAI's terms place responsibility on the builder, which for a legal-domain GPT matters: a user who treats the output as legal advice has no formal relationship with a licensed attorney and no recourse against either Porwal or OpenAI if the answer is wrong.
The system prompt frames LegalGPT as "an advanced AI with expertise in legal matters" whose job is to analyze documents, case studies, and precedents, interpret legal language, identify case law, and suggest strategies. The instruction set is short and generic; it does not pin the GPT to a jurisdiction, body of law, or document type.
As an advanced AI with expertise in legal matters, your task is to analyze legal documents, case studies, and precedents to provide insights and advice on legal issues. This includes interpreting complex legal language, identifying relevant case law, and suggesting potential legal strategies. Your analysis should assist lawyers and legal professionals in understanding the nuances of their cases, helping them to make more informed decisions and strategies.
None. Responses come from the base GPT-4 model plus live web results when browsing is enabled.
No external API actions are configured. The GPT cannot call Westlaw, LexisNexis, or any other primary-source legal database, a meaningful difference from purpose-built legal AI products.
| Task | What LegalGPT typically produces | Limits |
|---|---|---|
| Case analysis | Structured outline of issues, doctrines, and likely outcomes | No access to non-public records |
| Contract drafting | First-draft clauses for NDAs, employment agreements, wills, leases | Boilerplate; jurisdiction-specific provisions often missing or wrong |
| Legal research | Summaries of statutes, doctrines, and notable cases | High risk of fabricated citations (see Limitations) |
| Client explanations | Plain-language summaries of scenarios such as property disputes | Not a substitute for licensed legal advice |
| Document review | Spot review of contract clauses for ambiguity or risk | No privilege protection because data is processed by OpenAI |
The GPT performs best on language tasks rather than research tasks. Translating a clause from dense statutory English into something a client can read is a job the base model already does well. Asking it to find the controlling Eleventh Circuit case on a narrow FDCPA issue is where it breaks down, because the model will happily invent a case that sounds plausible.
LegalGPT belongs to a wave of generative legal tools that emerged after GPT-4 became available in 2023. The category splits into consumer chat experiences and enterprise platforms built for law firms.
Harvey is the best-known commercial peer. Founded by Winston Weinberg, a former O'Melveny & Myers litigator, and Gabriel Pereyra, a former Google DeepMind research scientist, Harvey raised a $5 million seed in November 2022 led by the OpenAI Startup Fund. Allen & Overy (now A&O Shearman) began a trial in November 2022; by February 2023 the firm reported roughly 3,500 lawyers had submitted around 40,000 queries. Harvey runs on customized LLMs and is sold through direct contracts with law firms, in-house teams, and the Big Four. It is not on the GPT Store.
CoCounsel was launched by Casetext on 1 March 2023 as the first AI legal assistant built on GPT-4. Casetext, founded in 2012 by Jake Heller, had spent a decade on a legal research platform before pivoting. CoCounsel handles document review, research memos, deposition prep, and contract analysis. Thomson Reuters acquired Casetext for $650 million in cash in June 2023, four months after launch. By early 2026 Thomson Reuters reported the product had passed one million users across 107 countries. It integrates with Westlaw and Practical Law.
| Feature | LegalGPT | Harvey | CoCounsel |
|---|---|---|---|
| Base model | GPT-4 via ChatGPT | Customized LLMs | GPT-4 with retrieval |
| Distribution | GPT Store (free with plan) | Direct enterprise sales | Thomson Reuters suite |
| Primary user | Lawyers, students, public | Big-law, in-house, Big Four | Lawyers across firm sizes |
| Source database | None (training plus browsing) | Firm-provided corpora | Westlaw, Practical Law |
| Privilege handling | No formal protections | Enterprise agreements | Thomson Reuters agreements |
The gap between LegalGPT and the enterprise tools is mostly about retrieval and accountability, not the language model. Harvey and CoCounsel ground their outputs in document corpora and primary-source databases the model must consult before answering. LegalGPT, with no actions and no knowledge files, has nothing comparable; its outputs are pure model recall.
A separate GPT also called "LegalGPT," built by Herman Singh under ID g-xK6S8XfHc, appears in AIPRM. Other products use the LegalGPT name outside the store, including legalgpt.pro and LawGPT, which are not affiliated with Porwal's GPT.
The biggest risk with consumer legal AI is the same risk that has tripped up working lawyers since ChatGPT became widely available: hallucinated citations.
In February 2022 Roberto Mata sued Avianca in the Southern District of New York after a metal serving cart hit his knee. His lawyers used ChatGPT to draft a brief opposing the motion to dismiss. The brief cited several federal decisions, including Varghese v. China Southern Airlines, that did not exist. The lawyers asked ChatGPT to confirm them; the model assured them the decisions were real and could be found on LexisNexis and Westlaw. In June 2023 Judge P. Kevin Castel dismissed the lawsuit and sanctioned the attorneys $5,000 under Rule 11, finding subjective bad faith. LegalGPT inherits this risk directly, because its retrieval surface is the same base model that generated the fictional Varghese decision. A Massachusetts lawyer was similarly disciplined in 2024, and the New York and California bars have issued advisories reminding attorneys that the duty of competence includes verifying every authority in a filing, regardless of who drafted it.
Every U.S. state criminalizes the unauthorized practice of law (UPL), with penalties ranging from civil fines to felony charges in a few jurisdictions. The American Bar Association and state bars generally take the position that an AI tool providing specific legal advice to a client without attorney review is engaged in UPL. The 2025 ABA article "Re-Regulating UPL in the Age of AI" notes that traditional UPL statutes were drafted to police human conduct and regulators are still working out how to handle widely distributed AI tools. LegalGPT sits squarely in this gray zone; ChatGPT's standard "the model can make mistakes" disclaimer is not enough to make its output safe to rely on.
Conversations with LegalGPT are processed and stored by OpenAI under its consumer ChatGPT terms. There is no attorney-client privilege, no work-product protection, and no data-processing addendum like firms negotiate with vendors. Disabling training does not create privilege. A lawyer who pastes a client's draft settlement into LegalGPT may be breaching the duty of confidentiality under Rule 1.6 of the Model Rules of Professional Conduct.
LegalGPT does not ask the user where they are. The same prompt can produce California-flavored, English common law, or generic output depending on subtle cues. Training data is weighted toward U.S. and English-language legal materials, so analysis of French civil procedure or Indian Companies Act questions is markedly thinner. Even within U.S. law the data over-represents appellate opinions while under-representing trial court rulings and state-specific procedural minutiae that often decide real cases.
LegalGPT first appeared in the public GPT registry on 15 November 2023, two weeks after OpenAI DevDay introduced Custom GPTs. The GPT Store opened public listings on 10 January 2024. The GPT was updated on 29 March 2024 per plugin.surf; the infobox snapshot from early 2024 lists 6,000 chats and a 2024-01-24 update.
Between November 2023 and the store launch, ChatGPT subscribers created roughly three million Custom GPTs per OpenAI. The legal category was crowded with independent assistants, most sharing the pattern: short system prompt, conversation starters, browsing on, no actions, no knowledge files. Since then enterprise tools have consolidated around Harvey, CoCounsel, Lexis+ AI, and firm-built systems, while consumer legal GPTs like LegalGPT remain quick-reference helpers.
Third-party indexes record modest engagement. AIPRM shows 14 upvotes and around 5,200 views. The plugin.surf record calls it a "virtual legal assistant." There are no formal reviews in legal trade publications, which have focused on Harvey, CoCounsel, and bar advisories. Consumer GPTs like LegalGPT are popular with students and self-represented litigants and largely invisible to the firms that drive legal-tech adoption.
No public example conversations have been released by the builder. Users can run any of the conversation starters above inside a ChatGPT Plus, Team, or Enterprise session.