# Frix vs ChatGPT

> ChatGPT is a general-purpose assistant: excellent at drafting and explaining, but it has no legal database behind it, so when asked for authority it can produce citations that do not exist, which has led to sanctions. Frix's Donna answers from a real index of court opinions and statutes and from your own files, cites the source at the page, and runs a citator on the cases she relies on.

URL: https://www.frixlaw.com/compare/chatgpt
Reviewed: 2026-09-29. ChatGPT is a product of OpenAI; nothing here quotes a price.

## Choose Frix if

- You need authority you can cite in a filing, with the opinion one click away.
- You want the answer to consider your own case file, not only the question.
- You want confidentiality terms written for a law firm, with no training on your data.

## Choose ChatGPT if

- You want a general assistant for email, explanations, code and everything outside legal work.
- You are drafting from your own knowledge and do not need the tool to find the law.
- You already pay for it and want one tool for everything.

## At a glance

| | Frix | ChatGPT |
|---|---|---|
| Legal database | Public Law Library: about 10.8M opinions with an automated citator, 2.9M statute and regulation sections, 71M federal docket entries, 15,000 judge profiles. Free, no account. | None. Answers come from model training and, on some plans, web search. |
| Citations | Numbered citations that open the opinion or statute; every cited case runs through the citator. | May cite real or invented cases; you must verify each one independently. |
| Your case files | Donna reads the records in your matter and cites the page. | Upload files per conversation; no matter structure. |
| Confidentiality | Firm data is never used to train models; access is logged; delete any time. | Depends on the plan and settings; check whether conversations are used for training. |
| Firm operations | Matters, document vault with editing, AI phone intake that answers calls and books consultations, tasks, calendar. | None. |
| Price | Free to start (3 guest questions, then a free account). Paid plans per firm; no sales call required. | Free tier and paid plans. |

## The fabricated citation problem is structural

In Mata v. Avianca, lawyers filed a brief with cases that ChatGPT had invented and were sanctioned. That was not a fluke: a general language model predicts plausible text, and a plausible-looking citation is exactly what it is good at producing. It has nothing to look the case up in. Since then courts have issued standing orders about AI use, and every lawyer using a general model for research has to verify each citation by hand, which we describe in [how to verify AI citations](https://www.frixlaw.com/blog/ai-legal-research-verify-citations).

Donna is built the other way round. She searches a real index first, reads what she finds, and answers with numbered citations to those documents. You can open each one. Then the citator runs on the cases she cited, so a case that was overruled last year shows a red flag before you rely on it.

## Use both, for different things

ChatGPT remains a good writing partner, and many lawyers will keep it for that. The rule of thumb is simple: if the output will be filed, cited or relied on, the authority has to come from a source you can open. That is what Frix is for, and the Law Library is free if you only want the source.

## Frequently asked questions

### Can ChatGPT do legal research?

It can explain legal concepts and draft, but it has no case-law database and can invent citations. Any authority it produces must be verified in a real source before it is used.

### Does Frix hallucinate citations?

Donna cites only documents she retrieved from the Law Library or your files, and each citation opens the source. Like any AI she can misread a document, which is why the source is always one click away and the citator runs on cited cases.

### Is my data used to train Frix's models?

No. Frix does not train models on customer data. See the security page for how data is handled.