B

Briefly

Models do the work that replaces manual entry. They extract matters, attorneys, clients, referees and narratives from uploaded submissions into the Experience Graph and map fields when a submission moves between directories. They also answer questions about the firm's experience and draft pitch, RFP, bio and website copy from the graph. Around that sit conventional workflow tools: a deadline schedule, assignment and status tracking, section by section review, an editor with Word export and a rankings tracker. Those would still run without the models, though the graph that feeds them would have to be built by hand.

Chat answers and generated content cite the matters, attorneys and past submissions they draw on, so a reader can open the record behind a claim. The product works only from the firm's own graph and general firm information. Briefly calls its output grounded and never hallucinated, but gives no accuracy figure for extraction, conversion or generated text. The website terms say AI output may not be fully accurate and that its factual assertions need independent verification.

People stay in the loop at several points. Submissions move through drafting, in review and submitted, with section by section review and assigned owners. A conversion between directories shows a view of what changed and flags fields that need review, for example 92% mapped automatically with four fields left for a person. Drafts are edited in the app before export. Nothing describes a confidence level that stops the system, or what happens when an extracted matter is wrong.

No customer, deployment or result is named. The figures on the site describe the problem rather than outcomes: 200 submissions a year at six partner hours each, 1,200 partner hours and 5,000 hours of marketing time, and more than $1 million lost to manual work. Survey statistics on rankings and general counsel come from a third party study, not from Briefly customers.

Confidential matter narratives and client relationships are the data Briefly holds. Its security page commits that a firm's content is never used to train AI models, its own or its subprocessors', and is used solely to operate the service for that firm. Foundation models run under strict no training provisions, and sensitive operations go through dedicated infrastructure. Access runs on role based permissions, single sign on and granular access policies in isolated environments.

These commitments sit on the website rather than in a published customer agreement, and privilege, retention and deletion of firm data are not addressed.

Section 3.2 of the website terms tells users that AI output may not be fully accurate and that its factual assertions need independent verification before reliance. The output is marketing and directory content rather than advice to clients, but submissions, pitches and website copy are still subject to professional conduct rules on lawyer advertising and client confidentiality. Those rules are not mentioned, and nothing says who in a firm may use the product.

No AI policy, principles, testing method or accountable owner is published. The security page covers data handling and audits, not how extraction or generation is tested, and nothing addresses uneven results across practice areas, directories or languages.

Data is encrypted with AES-256 at rest and TLS 1.3 in transit, and access runs on role based permissions, single sign on, granular policies and audit logs. Hosting uses isolated environments with automated backups and disaster recovery. Incident response procedures are documented, with notification commitments written into every contract, and penetration tests and third party audits run each year. The subprocessor list is shared only under NDA.

The privacy policy keeps personal data for as long as its purposes require, and no retention or deletion period is stated for a firm's matters, submissions or chat history.

The only published terms govern use of the website. They warn that AI output may be inaccurate, but state no cap, indemnity, warranty or insurance for the product a firm subscribes to. The customer agreement is not published.

The Experience Graph is said to be extracted from a firm's existing systems and kept current, but none of those systems is named. Documented routes in and out are file uploads of past submissions in PDF, Word or Excel, export to Word in each directory's format, and single sign on. No connection to an experience management system, CRM, document management system or the directories' own submission portals is described.

Delivery is a hosted service on enterprise cloud infrastructure with isolated environments for each firm, automated backups and disaster recovery. The cloud provider, the regions used and where model calls are processed are not stated.

A SOC 2 Type II audit covers security, availability, processing integrity, confidentiality and privacy and is repeated annually. The report, penetration test remediation reports, questionnaire responses and the subprocessor list go to active customers and prospects in active evaluation under NDA. The auditor and audit period are not named, and there is no open trust center.

Foundation models run under strict no training provisions, with sensitive operations routed through dedicated infrastructure. The models, their providers and where they run are not named, and no notice of a change in provider is promised.

No price, plan, unit of charge or contract term is published. Every route leads to a demo request or the sales address. The site's cost argument is about the firm's own labor, more than $1 million a year in partner and marketing time, not about what Briefly costs.

The buyer is the marketing and business development team of a law firm that files many directory submissions, sized around 200 or more a year. Five directories are supported from the start, Chambers, Legal 500, IFLR, Benchmark and WTR, with more than 5,000 practice and location cohorts and a growing list of others. Every practice area those directories rank is in scope. Smaller firms, in house teams and directories outside the five are not addressed, and the jurisdictions covered beyond the directories' own lists are not stated.