Transcribe mediations and arbitrations with AI (2026)

Mediators and arbitrators share the same documentation problem: long sessions, multiple voices, and the obligation to capture precisely who said what, what was proposed and what each party committed to. Taking notes while facilitating or presiding over the hearing divides your attention exactly when it's needed most; reconstructing the session from memory afterwards loses nuances that later matter.

AI transcription solves the mechanical part: the recording becomes faithful text and, with the right prompts, minutes, a position summary and an agreement extract. But this terrain has its own rules — mediation confidentiality is protected by law, and in arbitration the tribunal decides how proceedings are documented — so the correct workflow starts before pressing record. In this guide you'll find the full process, the prompts and the precautions.

Why transcribe mediations and arbitrations

Alternative dispute resolution procedures live on what is said in the room. Transcribing it well changes the work:

• Full attention during the session. The mediator who isn't taking notes facilitates better: keeps eye contact, picks up nonverbal language and manages tension. The recording captures the verbatim record; notes go from obligation to optional support.
• Commitments in their literal wording. "I'd be willing to take on 60% if the claim is withdrawn" is not the same as "I'll take on 60%". In agreements and awards, the exact formulation matters, and human memory rounds things off. The transcript preserves the sentence exactly as it was said.
• A case file per session, not an afternoon of drafting. From the same transcript come the minutes, the position summary and the list of commitments with three prompts. What used to be one or two hours of drafting afterwards becomes 15-20 minutes of review.
• Reviewable testimony in arbitration. When preparing the award, being able to search the hearing text ("what exactly did the expert say about the delivery date?") is worth more than any summary. Transcription turns hours of audio into material you can query in seconds.
• The clear boundary: AI documents, it doesn't decide. The mediation agreement is drafted and signed by the parties; the award is issued by the arbitrator. The transcript and its derivatives are working instruments — valuable precisely because they are faithful, not creative.

What documents you can generate from each session

The same transcript feeds different documents depending on the prompt. These are the ones that pay off most:

The key is separating transcription from its derivatives: a single well-transcribed recording produces the session's entire file with different prompts, without listening to the audio twice.

Step by step: from recording to case file

Step 1 — Agree on recording and document consent

This step comes before the technical part. Mediation confidentiality is protected by law in most jurisdictions — Spain's Law 5/2012 and the EU Mediation Directive protect it expressly — and recording without the agreement of all parties breaks the foundation of the procedure. Document in writing, in the initial session or in a specific agreement: that recording will take place, with which tool, what the transcript will be used for, who will access it and when it will be destroyed. In arbitration, the decision belongs to the tribunal or the rules of the administering institution; document the procedural agreement all the same. If any party doesn't consent, there is no recording: take notes and dictate your own summary afterwards — that dictation can also be transcribed with AI.

Step 2 — Record with quality and identifiable speakers

A session with multiple parties is the most demanding scenario for diarization. What helps most: a quality central microphone in person (or one channel per participant in remote sessions), and the discipline of the mediator or arbitrator naming whoever takes the floor ("the floor goes to the respondent"), which works as an attribution anchor in the text. If the session is tense and there are overlaps, a clean recording is your insurance. For difficult rooms, our guide on noisy or poor-quality audio covers the settings that matter.

Step 3 — Transcribe the audio with a tool that offers guarantees

Upload the recording to an accurate transcription tool like VOCAP and receive the full text, punctuated and ready to work with. In this field, choosing a tool is not just about quality but about confidentiality: a no-training commitment, data processing under a DPA and audio deletion after processing. A 2-hour session is transcribed in minutes; what used to require an afternoon of listening is done before your coffee.

Step 4 — Generate minutes, positions and agreements with prompts

Run the transcript through Claude or ChatGPT with the prompts in the next section, one per document. The two non-negotiable instructions: that the model adds nothing that isn't in the transcript, and that it marks as [NOT ADDRESSED] any sections that weren't discussed, instead of filling them in. In the commitments extract, require the literal quote next to each commitment: that's what lets you verify against the source in seconds.

Step 5 — Review, validate with the parties and archive with a deadline

Verify speaker attributions and each commitment against the transcript (and, when in doubt, against the audio), submit the minutes to the agreed validation process — the parties' signature in mediation, the tribunal's approval in arbitration — and honor what was agreed in step 1: delete the audio and the temporary transcript within the agreed timeframe. The valid record is the validated document, not the intermediate files.

Ready-to-copy prompts

Paste the session transcript and add one of these prompts on top. For mediation or arbitration material, use business accounts or APIs with a no-training commitment.

Mediation session minutes

Draft the minutes of this mediation session from the transcript,
with these sections: Attendees, Topics discussed, Proposals made
(indicating which party made each one), Partial agreements reached,
Matters in disagreement, and Next session. Use only what appears in
the transcript: do not add assessments, do not interpret intentions
and do not fill in gaps. If a section was not discussed, write
[NOT ADDRESSED]. Style: third person, neutral, no adjectives. Refer
to the parties exactly as they appear in the transcript.
Transcript: [paste the transcript here]

Per-party position summary

From this session transcript, prepare a summary of each party's
position, separately. For each party: (1) interests and concerns
they expressed, (2) concrete proposals they made, (3) conditions or
red lines they voiced, and (4) points where they showed flexibility.
Attribute each element only if it is clear who said it; if the
attribution is doubtful, mark it as [UNCERTAIN ATTRIBUTION] instead
of guessing. Do not evaluate the positions or suggest solutions:
only document what was expressed.
Transcript: [paste the transcript here]

Extract of agreements and commitments

Extract from this transcript every agreement and commitment that
was voiced. For each one: who takes it on, exactly what they take
on, deadline or condition if mentioned, and the literal quote from
the transcript that supports it, in quotation marks. Distinguish
between firm commitments ("I commit to...") and conditional
willingness ("I would be willing to... if..."), in two separate
lists. Do not include anything without a supporting literal quote.
If there are no agreements, say so explicitly.
Transcript: [paste the transcript here]

Rules of the game: consent and confidentiality

Transcription in mediation and arbitration is not a technical problem but one of rules. The ones with no shortcuts:

• Confidentiality is the foundation of the procedure. In mediation it is protected by law in most jurisdictions (Spain's Law 5/2012, the EU Mediation Directive): what is said in session cannot be used in later litigation, and mediator and parties are bound to keep it confidential. Any recording and transcript inherits that obligation in full.
• Express, prior, written consent. From all parties and the mediator or arbitrator, before recording, with the transcript's intended use spelled out. A quick "mind if I record?" on the fly is not a recording agreement.
• In arbitration, the tribunal decides. How hearings are documented is decided by the arbitral tribunal or the rules of the administering institution. The AI transcript is a working instrument; the official record is the one the tribunal approves and signs.
• Caucus is separate. The mediator's private meetings with each party carry reinforced confidentiality: what is said in caucus is not shared with the other party without permission. If they are recorded, keep them in files separate from the joint session, and never mix their transcripts into a single document.
• Without consent, there's a plan B. Manual notes during the session and the mediator's own dictation afterwards ("session three, two points: the parties have made progress on..."), transcribed with AI using the same workflow. You lose the parties' literal wording, you keep the rest of the value.

GDPR and data protection

A mediation or arbitration session concentrates sensitive personal data: family disputes, financial information, sometimes health data. The processing has to measure up:

• Tools with guarantees. Transcription and AI with a no-training commitment and a data processing agreement (DPA). Free consumer accounts are no place for a mediation. VOCAP, for example, deletes audio files after processing them.
• Minimize and set a deadline. Delete the audio as soon as the transcript is verified, limit access to the transcript to whoever the agreement allows, and destroy intermediate files within the timeframe agreed with the parties. Keeping things "just in case" multiplies risk without adding anything.
• Pseudonymize working documents. "Party A" and "party B" in summaries and drafts keeps names out of the files that circulate. If the material will be used for training or supervision, anonymize it properly: the guide on anonymizing transcripts and complying with GDPR covers the full process.
• Duty to inform. The parties must know which tools process their voice and their data, and have accepted it in the recording agreement from step 1. Transparency here is not courtesy: it's an obligation of the data controller.
• If you work at a mediation institution or arbitral court, align the workflow with its internal rules and its record of processing activities; the procedure must fit within them, not sidestep them.

Common mistakes that ruin the workflow

• Recording "for internal use" without a written agreement. In mediation there is no such thing as an informal recording: without documented consent from all parties, the recording compromises the procedure and the mediator's position. Step 1 is not optional.
• Trusting diarization without verifying. Attributing a commitment to the wrong party is the most expensive mistake in the whole workflow. Attributions are always verified, and the positions prompt must mark [UNCERTAIN ATTRIBUTION] instead of guessing.
• Mixing caucus and joint session. A single file containing the private meetings and the general session is a confidentiality leak waiting to happen. Separate files, separate transcripts, separate access.
• Letting AI "round off" agreements. A model without the instruction to quote literally turns "I'd be willing to consider 60%" into "accepts 60%". Require a verbatim quote next to each commitment and the distinction between firm and conditional.
• Keeping the audio indefinitely. Every month a mediation's audio stays on a drive is risk accumulating against what was agreed with the parties. Verified transcript = audio deleted, and final destruction within the agreed timeframe.
• Asking AI to propose the solution to the dispute. Suggesting settlement terms or anticipating the direction of the award turns a documentation tool into a professional risk. Faithful scribe, not negotiator: that boundary is what makes the workflow defensible.

Frequently asked questions