AI Use and Limits
Last updated October 19, 2026. Jural uses AI models to read, summarize, compare and draft legal documents for law firms. This page sets out what those models do, what they do not do, and what every user agrees to check before relying on an output. It forms part of the Terms of Service.
1. Jural does not give legal advice
Jural supports attorneys and their staff. It does not give legal advice, it does not form an attorney-client relationship with anyone, and it does not decide what a client should do. Every summary, risk list, redline, draft, research answer, chronology, conflict report, computed deadline and time entry is a draft. The responsible attorney decides whether it is used, changed or discarded.
2. The attorney verifies every output
Before an output is sent to a client, filed with a court, served on another party or billed, a licensed attorney at your firm reviews it. In practice that means the following.
- Read each citation and quotation against the source it points to.
- Confirm every computed date against the governing court rules, statutes, orders and contract terms, because Jural does not replace your court rules calendaring review.
- Check that a draft reflects the facts of the matter and the instructions of the client.
- Review redlines clause by clause instead of accepting tracked changes in bulk.
- Edit draft time entries so the narrative and time match the work actually performed.
Your professional obligations, including competence, supervision of nonlawyer assistants and confidentiality, stay with your firm.
3. Sources and citations
When Jural answers a research question, it cites only the sources it actually searched in that run, such as documents in the matter, the firm knowledge base or the public legal sources the research module consulted, and each statement links to the passage it relies on. Jural never invents citations. If the search finds nothing relevant, the answer says so plainly rather than offering a guess. A cited passage still needs attorney review, because a source can be superseded, distinguished or read differently in your jurisdiction.
4. What the models do
- Extract text from PDF, DOCX and scanned documents, including text recognition on image-only pages.
- Summarize documents and productions and pull out parties, dates, amounts and obligations.
- Compare a contract with your playbook and suggest fallback language.
- Fill firm templates with matter facts and keep your clause wording.
- Build chronologies with a source reference for each entry, for example a Bates number or page.
- Propose deposition outlines, intake summaries, task lists and draft time entries with narratives.
5. What the models do not do
- They do not file, serve or send anything to a court, a client or another party.
- They do not sign documents or approve outputs on behalf of an attorney.
- They do not calendar deadlines as final, and they do not decide which court rule applies.
- They do not reach across per-matter access walls, so an answer in one matter never uses documents from a matter the user cannot open.
- They do not learn from your files between requests and keep no memory of one firm's work for another firm.
6. No training on client files
Client documents, questions, drafts and outputs are never used to train AI models. Jural sends only the text needed for a request to AI model providers, which act under terms that prohibit training on customer data and limit retention to what is needed to provide and secure the service. Data is encrypted in transit and at rest. More detail is on our confidential ai for law firms page and in the Privacy Policy.
7. Usage limits per plan
AI work is measured per user per month and counts consumption in the billing period, so deleting a document does not give pages back. Every plan has a fixed monthly allowance.
- Solo includes 1 000 analysed pages, 150 assistant questions, 20 drafts and 30 research questions per user each month.
- Firm includes 3 000 analysed pages, 500 assistant questions, 80 drafts and 100 research questions per user each month.
- Enterprise includes 6 000 analysed pages, 1 200 assistant questions, 200 drafts and 250 research questions per user each month.
When a seat reaches a limit, that feature pauses until the next billing period or until the firm moves to a higher plan. Current prices are on the ai legal assistant pricing page.
8. Reporting a wrong output
AI models can misread a scan, miss a clause or state something with more confidence than the source supports. If you find an output that is wrong, write to [email protected] with the matter reference and a description, without attaching client documents unless we ask for a specific page. We review each report and use it to improve how Jural checks its own results.
9. Changes to this page
We may update this page when the models or the modules change. The date at the top shows the current version. Questions about AI use at Jural go to [email protected].