Should law firms use AI to screen cases?
AI is appropriate for organizing and routing intake information, and inappropriate for deciding whether a case has merit or what it is worth. The useful applications are structuring what a visitor submitted, flagging matters against criteria the firm defined, and prioritizing time-sensitive inquiries — all of which speed up human review without replacing it. The applications that create real exposure are automated case valuations shown to prospective clients, automated rejections, and anything a visitor could reasonably read as legal advice, since an inaccurate estimate sets expectations the firm may not be able to meet and merit determinations are an attorney's professional responsibility.
Updated August 3, 2026 by CaseMetric
Where AI genuinely helps
Structuring what was submitted. Turning a free-text description into labeled fields a paralegal can scan is useful, low-risk work. The facts came from the visitor; the model is only organizing them.
Routing against firm-defined rules.If the firm says matters older than two years go to a specific attorney, or that anything involving a commercial vehicle gets flagged, a system can apply those rules consistently at 2am. The judgment was the firm's; the execution is automated.
Surfacing urgency. An inquiry mentioning an imminent deadline should not sit in a queue behind forty routine submissions. Ordering by time-sensitivity is a scheduling decision, not a legal one.
Where it creates real exposure
Case valuation shown to the prospective client. This is the most common and the most consequential. A range produced from a few form answers, before anyone has seen a medical record or a police report, is not an estimate — it is a guess presented with the authority of software. The firm then either inherits an anchored expectation or loses a viable matter.
Automated rejection.A model that tells someone their case does not qualify has made a legal determination on the firm's behalf, to a person the firm never spoke to, based on information that may be incomplete or misremembered.
Anything that reads as advice.“You may have a strong claim” is an assessment. So is “you should file soon.” The line is not whether the firm intended advice; it is whether a reasonable person would read it that way.
The distinction most vendors blur
Much of what is marketed as “AI intake” is not AI at all — it is a branching form, where selecting “slip and fall” reveals different questions than “car accident.” That is straightforward conditional logic, it carries essentially no professional-responsibility risk, and it does the large majority of the practical work.
It is worth knowing which one a product is actually doing, because the risk profiles are not comparable.
A workable position
Use automation to make sure nothing is missed, everything is structured, and urgent matters are visible quickly. Keep every judgment about merit, value, and acceptance with the attorney. Tell visitors plainly that submitting an inquiry does not create an attorney-client relationship, and do not show them anything that resembles an assessment of their claim.
For transparency: this is the position CaseMetric is built around. Its intake applies only the routing and priority rules a firm approves, and it does not tell visitors whether their case has merit or what it may be worth.
Related questions
Can AI tell a law firm whether a case is worth taking?
No. Whether a firm accepts a matter depends on facts an intake form never sees — the firm's current caseload, its appetite for the matter type, jurisdiction, the credibility of the person, and the attorney's own read. AI can order a queue against criteria the firm supplied. It cannot make the decision.
What's wrong with showing a prospective client an estimated case value?
It sets an expectation before anyone has reviewed the facts. If the estimate is higher than what the matter is actually worth, the firm inherits a client who believes they were promised something. If it is lower, the firm may lose a case it would have wanted. Neither error is recoverable, and the estimate is generated from a handful of self-reported answers.
Is automated intake the same as AI screening?
No, and the distinction matters. Automated intake means the questions branch based on what the visitor selects — a rules-based flow that requires no AI at all. AI screening means a model interprets the submission and makes a judgment about it. The first is low-risk and does most of the practical work; the second is where the exposure is.
Could AI intake create an attorney-client relationship by accident?
It creates the risk. If an automated system tells a visitor their claim looks strong or that they should file, a reasonable person may believe they have received legal advice from the firm. Intake should state explicitly that submitting an inquiry does not create an attorney-client relationship, and should avoid language that reads as an assessment.
Related reading
See how CaseMetric handles this
Practice-specific guided intake for plaintiff law firms, designed and installed for you.