A deposition is the one stage of a case where the evidence is created in front of you, sentence by sentence, and where the decision about what to ask next has to be made in the seconds before you ask it. Realtime court reporting has, for years, put the transcript on the attorney's screen as the witness speaks. What is new is software that reads that stream as it arrives and reports back: this answer contradicts page 41 of the prior deposition; this expert has just described a method with no stated error rate; this is the third time the witness has said "I don't recall" about the same date. This article explains how the stream itself works, what a live analysis layer can and cannot do with it, what a Daubert flag means in that setting, and the rules about recording, notice and privilege to settle with counsel before any of it is switched on.
A stenographic court reporter writes on a steno machine in a shorthand of chord strokes. Computer-aided transcription (CAT) software on the reporter's laptop translates those strokes into English text against the reporter's personal dictionary, in near real time. "Realtime" means that translated text is sent out of the reporter's machine to other screens as it is written, rather than held until the transcript is produced. Attorneys in the room can read the testimony as it happens, scroll back, mark passages, and search. Remote participants can receive the same feed over the internet.
The realtime feed is not the certified transcript. It is a draft: untranslated strokes appear as fragments, homophones are sometimes wrong, and the reporter corrects and certifies the final transcript later. Any use of the live text, by a person or by software, has to keep that distinction in mind.
There are three ways a live transcript reaches a screen, and they matter for what an analysis layer can attach to.
The reporter's realtime output. Stenograph's CaseViewNet is the widely used example. Stenograph describes it as a way to securely stream realtime text from the court reporter to attorneys, judges and litigants, either in the same room or, with its cloud option, remotely, using a session code to connect participants, with viewer applications for Windows and iPad; Stenograph also states that most CAT software can send to it. Those are the vendor's descriptions of its own product, and the details of setup, cost and reliability are its to answer. Other CAT vendors have equivalent outputs. What matters here is that the stream is text, arriving in small increments, and that the reporter controls whether and to whom it is sent.
Agency and platform streams. Court-reporting agencies and remote-deposition platforms increasingly offer browser-based realtime, in which the reporter's feed is relayed through the platform to authorized viewers, often alongside the video. The reporter's output is the source; the platform is the transport.
Audio transcription. A separate route ignores the reporter's feed and transcribes the room or conference audio directly with speech-recognition software. This produces text with no reporter in the loop, which makes it available anywhere there is audio, at the cost of a transcript that is not a court reporter's and may be less accurate on medical vocabulary, cross-talk and names. It is useful for analysis and worthless as a record.
Given a stream of text, the useful things software can do during a deposition come down to comparison and pattern-matching against material you loaded before the deposition began.
Every item is a comparison against something already loaded; a layer with no prior testimony and no chronology has nothing to compare against.
Federal Rule of Evidence 702, as amended December 1, 2023, allows a qualified expert to testify "if the proponent demonstrates to the court that it is more likely than not that: (a) the expert's scientific, technical, or other specialized knowledge will help the trier of fact to understand the evidence or to determine a fact in issue; (b) the testimony is based on sufficient facts or data; (c) the testimony is the product of reliable principles and methods; and (d) the expert's opinion reflects a reliable application of the principles and methods to the facts of the case." The 2023 amendment made explicit that the proponent bears that burden by a preponderance, and the advisory committee's note says courts had erred by treating some reliability questions as matters of weight rather than admissibility. State courts apply their own versions; check the rule that governs your case.
A "Daubert flag" in a live layer is a pattern in the expert's testimony that maps onto one of those four prongs, tagged so the questioner can pursue it while the witness is still under oath. Examples of what a well-built layer looks for:
The flag is a starting point for a question, not a conclusion. Most flags are answered by the next question and go nowhere. The ones that are not become the paragraphs of a motion to exclude, which is why the value of the layer is in creating a clean record on each prong while the witness is present, rather than discovering the gap when the transcript arrives three weeks later. The longer treatment of the motion side is at why expert witnesses get excluded under FRE 702.
None of this should be switched on without settling a short list of questions with the attorney responsible for the deposition, and often with opposing counsel and the reporter. The rules below are federal; state rules differ and control in state-court cases.
Method of recording and notice. Under FRCP 30(b)(3)(A), "the party who notices the deposition must state in the notice the method for recording the testimony," and "unless the court orders otherwise, testimony may be recorded by audio, audiovisual, or stenographic means." Under 30(b)(3)(B), "with prior notice to the deponent and other parties, any party may designate another method for recording the testimony in addition to that specified in the original notice," and "that party bears the expense of the additional record or transcript unless the court orders otherwise." An audio feed captured for transcription by software is, at least arguably, an additional method of recording, and the conservative course is to notice it. A layer reading the reporter's realtime text is not a new recording of the testimony, but it is a use of the reporter's work product, and the reporter or agency may have terms about that.
Remote depositions. FRCP 30(b)(4): "The parties may stipulate, or the court may on motion order, that a deposition be taken by telephone or other remote means." Platform recording and streaming should be covered by the stipulation.
Consent to audio capture. Where the layer works from room or conference audio, state recording-consent statutes apply to the participants, and some states require every party to consent. A deposition is on the record and everyone knows it is being recorded, but "recorded by the reporter" and "streamed to a third-party server for transcription" are not the same thing, and consent or notice should cover the latter explicitly.
Privilege and work product. The flags, drafts and summaries the layer produces are the attorney's mental impressions and should be treated and stored as work product. The realtime feed of the testimony itself is not privileged; it is the same testimony everyone in the room heard. Keep the two separate: the transcript is shared, the analysis is not. If the witness's own counsel is using a layer, the same applies to them.
Protective orders and confidentiality. Medical records and prior testimony loaded into the layer are frequently subject to a protective order or contain protected health information. Where the data goes, who at the vendor can see it, whether a business associate agreement is in place, and how long the data is retained are questions to answer in writing before the first upload. This is the same diligence that applies to any record-review tool.
Courtroom AI is the live layer described above, built for plaintiff and defense trial attorneys. It takes audio from Zoom, Teams, a phone microphone or a desktop audio device, transcribes it with a medical-vocabulary speech model, and runs three lenses on each answer as it arrives: Daubert flags on the four prongs above with a severity tag, prior-testimony inconsistencies against the depositions and reports you uploaded before the session, with the two quotes paired, and drafted cross-examination questions the attorney can read from the screen or ignore. It produces a running transcript and a post-session summary. It does not replace the reporter, does not produce a certified transcript, and does not decide anything; every flag is two passages and a suggestion. Pricing, the trial period and the setup steps are on the product page, and the questions in the previous section should be answered with counsel before the first session, as they would be for any tool that touches the record.
It is the court reporter's transcript sent to other screens as it is written, rather than held until the transcript is produced. The reporter's steno strokes are translated by CAT software into text, and that text is streamed to attorneys in the room or remotely through the reporter's realtime output, an agency or platform relay, or a viewer application. The live feed is a draft, not the certified transcript.
It compares the incoming testimony against material loaded before the deposition: prior depositions and reports, a cited chronology of the medical record, and the witness's own earlier answers in the same session. It flags conflicts with the passages paired, highlights specialty terms, tags expert testimony that maps onto the FRE 702 prongs, drafts follow-up questions for the attorney to use or discard, and keeps a running summary. Its value depends entirely on what was loaded.
A pattern in an expert's live testimony that maps onto one of the four requirements of Federal Rule of Evidence 702: helpfulness to the trier of fact, sufficient facts or data, reliable principles and methods, and reliable application to the facts. Examples are an expert who did not review a category of records, cannot name a method or an error rate, or gives an opinion outside their specialty. The flag is a prompt for a foundation question and for building a record for a later motion, not a conclusion.
Under FRCP 30(b)(3), the noticing party must state the method of recording, and any party may add another recording method with prior notice to the deponent and other parties, at its own expense unless the court orders otherwise. Capturing audio for software transcription is arguably an additional method and should be noticed; reading the reporter's realtime text is a use of the reporter's output and may be subject to the reporter's or agency's terms. State rules and recording-consent statutes differ and control in state court. Settle it with counsel before the deposition.
Courtroom AI listens live, pairs every inconsistency with the prior passage, and tags expert testimony by FRE 702 prong while you can still ask about it. Trial available; check the recording and notice rules with counsel first.
See how Courtroom AI works →Sources. Federal Rules of Civil Procedure, Rule 30 (30(b)(3), 30(b)(4), 30(c)(2), 30(e), quoted). Federal Rules of Evidence, Rule 702 as amended December 1, 2023 (quoted, with the advisory committee's note). Stenograph, CaseViewNet product description (the vendor's own claims, as noted in the text). Courtroom AI features as described on the product page. No accuracy, time-savings or cost statistics for live analysis are cited in this article because no independent published figure was located; vendor figures, including ours, are claims to be tested on your own depositions.
MedLegal AI is software, not a law firm. This article is general information for attorneys, paralegals, legal nurse consultants and patients; it is not legal or medical advice and does not create an attorney-client relationship. Verify every statute, rule and figure against the current source before relying on it. Questions: [email protected]