Client intake call transcription for solo attorneys and small firms

Intake calls are the highest-leverage 30 minutes in a solo practice

A client intake call decides three things in one sitting: whether the matter is viable, whether you have a conflict, and whether the prospect trusts you enough to retain. Getting it on paper — accurately, fast, without burning your evening — is where AI transcription earns its keep. We transcribe intake calls in production for solo and small-firm attorneys, and the workflow we recommend is narrow: record with consent, transcribe after the call, draft a memo from the transcript, then bill the analysis time you would have spent re-listening.

What AI transcription is not: a substitute for your judgment, and not a certified court record. If you need a sworn transcript for a deposition or hearing, hire a CSR. Everything below is about the internal work product — the intake memo, the conflict check, the matter file — not the record of proceedings.

Why the intake call is where the time hides

A typical 30-minute intake produces 4,000-5,000 spoken words. Re-listening at 1.5x to write notes costs you 20 minutes. Typing as you go splits your attention, breaks eye contact on video, and makes a distressed prospect feel processed. Most solos either short-cut the memo and lose detail, or eat the time after hours.

A transcript collapses that. You get a searchable record of the call a few minutes after hangup. Names, dates, dollar amounts, and prior counsel are pulled out by ctrl-F instead of by memory. The 20-minute re-listen becomes a 5-minute skim of the transcript with the memo template open beside it.

The leverage compounds with volume. A firm doing 8 intakes a week saves roughly two hours of attorney time — time that bills at $300-$500/hour if redeployed to substantive work. The deeper win is that you stop interrupting the prospect to spell things and can stay present while they describe the facts in the messy, non-linear order they actually remember them.

Privilege: when it attaches, and what recording does to it

Attorney-client privilege attaches to communications made for the purpose of seeking legal advice, in confidence, with someone the client reasonably believes is an attorney. It attaches during the intake call — before any retainer is signed — as long as the prospect is consulting you in your professional capacity. This is the prospective-client privilege recognized in ABA Model Rule 1.18 and most state analogues.

Recording the call does not destroy privilege. Sharing the recording with a third party can. That's the operative concern with cloud transcription.

What this means in practice:

• The transcript is privileged work product the moment it exists, provided it stays inside your control.
• A vendor processing the audio is not a "third party" for privilege purposes if they are acting as your agent under a confidentiality obligation — the same logic that lets you use an outside copy service or a paralegal contractor. Get this in writing in the vendor's DPA or terms.
• Confirm the vendor does not use your audio to train their public models. We don't, and we put that in writing.
• Storing transcripts in a system your staff and co-counsel can access is fine. Forwarding them to opposing counsel, posting them in a shared drive with a non-firm contractor, or pasting them into a public LLM chat is not.

We retain audio and transcripts in encrypted storage and delete on request via our opt-out endpoint. We are not a HIPAA BAA-covered product, which matters less for legal intake than for medical — but flag it if your matter touches PHI (workers' comp, personal injury, ERISA, medical malpractice). For lawyers working with health records, see our notes for clinicians on the same data-handling stack.

Consent: one-party vs two-party states

You cannot record an intake call without complying with your jurisdiction's wiretap statute. The split:

• One-party consent (federal default and ~37 states): you can record because you are a party to the call. No notice to the prospect required, though many bars recommend it for ethical reasons.
• Two-party / all-party consent (CA, FL, IL, MD, MA, MT, NV, NH, PA, WA, and others — check current statute): every party must consent. A unilateral recording is a criminal offense and the recording is inadmissible.

If either party is in a two-party state, ask for consent on the record at the top of the call. A line that works: "I record intake calls so I can focus on what you're telling me instead of taking notes — is that okay?" Their yes goes into the transcript and becomes the consent log. If they say no, take manual notes — don't get clever.

For meeting-bot recordings on Zoom, Google Meet, or Teams, we post a disclosure message in chat when the bot joins. That helps but does not replace verbal consent in two-party jurisdictions — you still need the prospect to say yes.

Bar ethics opinions in some states (notably the historic NY State Bar Op. 328, since softened) treated surreptitious recording as misconduct even where legal. The safer rule, regardless of state: always ask.

Conflict checking is faster from a transcript than from memory

Conflicts are where transcripts pay for themselves on the first call. The transcript exposes every named party — spouse, business partner, prior counsel, opposing party, related entities, witnesses — in a way memory does not. A 30-minute intake often surfaces 6-12 names that bear on conflicts, and the prospect mentions the holding company in minute three and the maiden name in minute eighteen.

A working pattern:

1. After the call, open the transcript and search for proper nouns and entity suffixes ("Inc", "LLC", "Trust", "Estate of").
2. Paste the extracted names into your conflicts system — Clio, MyCase, PracticePanther, or a spreadsheet if you're early.
3. Resolve conflicts before you send the engagement letter. If a conflict surfaces, you have the transcript to memo the no-go and the reason.

Our diarization splits the call into prospect and attorney turns. On stereo recordings — for example, a Zoom call where each speaker is on a separate channel — the speaker split is near-perfect. On mono phone recordings, we use pyannote-3.1, which is good for two speakers but degrades if a paralegal joins the line or a fourth person dials in. Names from phone audio always need a once-over.

Audio quality decides how much editing you do

Most intake calls run over phones, and phone audio is harder than studio audio. Narrowband telephony, speakerphone echo, and crosstalk all push the error rate up.

We run AssemblyAI Universal-3 as the primary engine; Whisper Large-v3 is a fallback for transient errors only. On clean 16 kHz audio (a good Zoom connection), accuracy runs around 92% — WER ~7.88%. On 8 kHz phone audio, it drops to ~82% — WER ~17.7%. Expect to fix one or two proper nouns per phone call, especially names, street addresses, and the spelling of opposing counsel.

Five habits cut the editing time in half:

• Record each speaker on a separate channel when your VoIP system supports it (RingCentral, Dialpad, and Zoom Phone all do).
• Ask the prospect to spell names, companies, and email addresses.
• Repeat back deadlines, hearing dates, and dollar amounts so they land in the transcript twice.
• Avoid speakerphone on your end.
• Pause before talking over the caller — overlap is where diarization fails.

From transcript to intake memo

A good intake memo is short and structured. The AI drafts the structure; you do the analysis. We do not recommend asking an LLM to write the legal conclusions — that is your work and your liability. We do recommend letting it pull the facts.

A template that works:

• Caller / prospect: name, contact, jurisdiction, referral source.
• Adverse parties and related names: extracted for conflicts.
• Timeline of facts: dates and events in the order the prospect described them.
• Documents mentioned: contracts, court filings, correspondence, medical records.
• Prior counsel: who, when, why ended.
• Statute of limitations indicators: any date that could trigger a clock.
• Prospect's stated goal: in their words, quoted from the transcript.
• Fee discussion: what you quoted, what they said.
• Attorney's analysis: viability, theory, credibility, next steps. You write this section.

The transcript fills the first eight sections. You write the ninth. That split is the point: AI handles extraction, the attorney handles judgment — including the credibility read, which no model can do from text. For more on the underlying pipeline, see our audio-to-text feature page.

If you take intake on Zoom or Google Meet, our meeting bot joins as a participant, records, and produces the transcript after the call ends. Bot minutes cost 2 credits per minute; file uploads cost 1 credit per minute. Pro is 600 audio-minutes plus 900 bot-minutes a month (as of May 2026) — enough for roughly 30 intake calls mixing bot and uploaded phone recordings.

What AI transcription will not do

Be honest with yourself about the seams.

• It will not catch every name on 8 kHz phone audio. Expect to fix one or two proper nouns per call.
• It will not reason about privilege or conflicts. It surfaces the names; you decide.
• It will not file the memo into your practice management system automatically. We don't ship native Clio or MyCase logging — you wire that via webhook, or copy-paste.
• It will not replace the certified record if the call later becomes evidence. If you anticipate that, hire a court reporter for the second meeting.
• We don't have a native iPhone app. The web app is mobile-responsive, but recording an in-office intake on your phone means using a separate recorder app and uploading the file.

Otter.ai does live in-meeting captioning well, which we don't — we transcribe the recording after the call ends. Rev.com offers human transcription if you need a sign-off-quality transcript for a sensitive matter; that costs more and takes hours rather than minutes. We sit in the middle: fast, machine-only, accurate enough for internal work product.

Pricing math for a solo

A solo doing 8 intakes a week, 30 minutes each, is 240 audio minutes a week — about 1,040 a month. That fits inside Pro (600 audio-minutes/month, $19/month as of May 2026). If half the intakes are Zoom bots, bot-min usage runs ~480/month against the 900-bot-min allowance. Overage packs are $5 / $15 / $39 when you spike.

For firms with 3-5 attorneys sharing intake, Business (2,500 audio-minutes + 3500 bot-min) is the bracket. Full breakdown on the pricing page. For the broader legal workflow — depositions, hearings prep, witness interviews — see our notes for lawyers.

What next

• Run a 5-call pilot. Record your next five intakes (with consent), transcribe them, and time how long it takes to write the memo from the transcript versus from memory. If you don't save 10+ minutes per call, the tool isn't right for you.
• Update your engagement letter and intake script to include the recording disclosure. Check your state's wiretap statute before the first recorded call.
• Upload one recorded intake to the Free plan — 30 minutes/month, exports unlocked — and see how the diarization handles your phone-line audio before committing.
• If your firm needs a HIPAA BAA for PI or workers' comp matters that touch PHI, email us. We are not BAA-covered today and will say so plainly rather than waste your procurement cycle.