Summary: At intake, AI is good at the parts that are typing: acknowledging an enquiry quickly, turning a call or form into a structured summary, drafting the records requests, and chasing the providers who have not sent anything back. It is not good at deciding whether a case is worth taking, and it should not be the thing that tells a claimant anything. The useful question is not whether to use it, it is which specific step you are handing over.
Why intake specifically
Plaintiff work has an unusual property: the same enquiry often arrives at three firms, and the one that responds first is frequently the one that signs. That makes response time a commercial variable in a way it simply is not for a transactional practice.
It is also the step where the work is most repetitive. The questions asked at first contact are largely the same every time. The records requested come from the same categories of provider. The follow-up when nothing arrives is a calendar problem, not a legal one.
Contingency billing sharpens it further. In an hourly practice below capacity, saving administrative time can just convert billable hours into free ones. On contingency there is no such conflict: time recovered is capacity for another matter at the same fee.
What it does well
- Acknowledging an enquiry immediately. Not a decision, not an assessment, just confirmation that a person exists on the other end and what happens next.
- Turning a call or form into a structured summary. The same facts, in the same shape, every time, so whoever reviews it is reading rather than reconstructing.
- Drafting records and bills requests. Same providers, same formats, endless volume.
- Tracking what is outstanding and drafting the chase. This is the one that quietly eats a paralegal’s week and has nothing to do with judgement.
- Drafting status updates. Most weeks the honest answer to “any news” is no, and it still takes someone twenty minutes to say it forty times.
Every one of those produces a draft. A person reads it before it goes anywhere.
What it should not do
- Decide whether to take a case. Screening on liability, damages and collectability is judgement, and it is the judgement your fee depends on.
- Say anything to a claimant unreviewed. An injured person reading an automated message about their own case is a bad outcome even when the message is accurate.
- Estimate value. Not to the client, not internally as though it were an opinion.
- Touch medical information without a deliberate decision about which tool sees it. That is a configuration and vendor-terms question, and it deserves an answer before anything is switched on rather than after.
- Send. Ever, by default. If a system can contact a claimant without a person approving it, that is a design choice someone made, and it should have been made on purpose.
The question that actually decides it
Not “should we use AI at intake”. That question has no useful answer because it bundles the typing with the judgement.
The better question is: which specific step are we handing over, and who reads the output before it matters? Asked that way, most firms find two or three steps that are obviously safe to hand over and one or two that obviously are not, and the argument stops.
Before anything touches client information
Whatever you end up using, the diligence question comes first, and it is the same one ABA Formal Opinion 512 describes: what does the vendor retain, is the confidentiality obligation enforceable, would you be told about a breach, and is the tool actually configured that way rather than merely capable of it.
I wrote those out as ten questions to ask any AI vendor before client information touches it. Free, not gated, and each question traces to Opinion 512 or NC 2024 Formal Ethics Opinion 1. I am not a lawyer and none of it is ethics advice.
The practice-specific version of this, including how the assessment works, is on the personal injury page. There is a fuller explanation of the opinion itself in ABA Formal Opinion 512, explained for small firms.
The arithmetic
Do it with your own numbers rather than anyone else’s. If faster intake signs you even one additional case this year, put your average fee against what the assessment costs. That is the bet, and the guarantee is what makes it free to take: if the assessment does not identify at least 10 hours a month of time savings, you get a full refund.