Enabling verbal time

Verbal time entry allows users to record a spoken description of their work in Legalsense. Legalsense sends the recording to a configured AI provider, then creates draft time entries for review and approval.

This article explains what's needed to enable verbal time entry for your organization. As an administrator, your role is to make the necessary preparations, obtain an API key from an AI provider, and set up the voice recording option.

1. Set up the AI provider

Go to Settings → Integrations → AI provider to configure the AI provider.

Legalsense Settings page on the Integrations tab, showing the AI provider configuration form with OpenAI selected and the API key field filled.

Supported AI models

Legalsense uses OpenAI's transcription and language models to transcribe submitted audio into text and then convert that text into structured time entry data. Legalsense currently supports the following models:

Transcription

• GPT-4o-transcribe
• GPT-4o-mini-transcribe

Processing (structured output)

• GPT-4.1
• GPT-4.1-mini
• GPT-5.6-terra
• GPT-5.6-luna

To set up the AI provider, fill in the fields and click Save. If a green message appears confirming that the configuration succeeded, you're done with this part. The meaning of the fields is explained per provider below.

OpenAI

To obtain an API key, go to ⧉ OpenAI Platform and create an account with payment details. When creating the API key, make sure it has read and write access to at least the Responses, Chat completions, and Transcription (Audio API) endpoints. You can also set up a service account as an OpenAI API user; this is preferred if multiple users will be using the feature.

• Resource URL: if you use data residency through OpenAI, you need to change this URL. If you're not sure, use the default value. See also ⧉ OpenAI Data Controls.
• Transcription model: select one of the transcription models listed above.
• Structured-output model: select one of the processing models listed above.

Azure OpenAI

If your organization already runs on Microsoft Azure, you can point Legalsense at your own Azure AI Foundry resource instead of plain OpenAI. Verbal time entry then runs through your existing Azure billing, security, and compliance posture.

This requires an Azure AI Foundry resource with two model deployments: one for transcription and one for structured output. The supported models are listed above. When you select Azure OpenAI as the provider, you provide:

• Resource endpoint: the base URL of your Azure resource, for example https://your-resource.cognitiveservices.azure.com/.
• API key: Entra ID is not currently supported.
• Transcription model: the name of your deployment of the transcription model.
• Structured-output model: the name of your deployment of the processing model.

Important: Legalsense doesn't validate deployment names when you save. Make sure they match exactly what's configured in Azure, and that each deployment uses one of the supported models listed above.

2. Set up the voice recording

Go to Settings → General → System and, under Time entry methods, enable Voice. This makes a recording control appear alongside your usual time entry options.

The general settings page, showing where to enable voice recording.

The Time entry methods setting under Settings, General, System

Once you've configured both the AI provider and the voice recording, you're ready to use verbal time entry. Under Time → Write time, on the right-hand side, start the voice recording and dictate one or more time entries.

What data is shared with the AI provider

When a user submits a verbal time entry, Legalsense sends the audio recording to the configured AI provider for transcription and interpretation.

To help the provider match the spoken input to the correct matters, activities, and timekeepers, Legalsense also sends contextual information about the user's accessible work. This contextual information is always bounded by what the user can access: it cannot cross permission or ↪ Chinese wall boundaries. In other words, only the information that a user can see in the manual time entry form is included. At the time of writing, this comprises:

• Names and IDs of:
  • active timekeepers (including the submitting user),
  • time-writable matters,
  • activities for those matters,
  • submatters for those matters, and
  • Recofa categories for those matters.
• The audio recording submitted by the user.

OpenAI data handling

At the time of writing, OpenAI does not use data processed via the API to train its models. OpenAI's API privacy statement can be found at ⧉ https://openai.com/enterprise-privacy/ .

Legalsense sets the store=False parameter for applicable endpoints, which means OpenAI does not retain prompt data by default. If your organization requires additional data controls, such as data residency or zero data retention, you can request these from OpenAI. More information is also available at ⧉ OpenAI Data Controls.

What verbal time entry costs

Legalsense doesn't charge for verbal time entry. It uses your AI provider to turn a short recording into draft time entries, and your provider bills you for that directly. Expect less than $1 per timekeeper per month in most cases. Even a timekeeper who records after every phone call costs less than $2.50 per month with the economical setup, and around $10 at most with the most accurate one. For most firms, a single billable minute that would otherwise have gone unrecorded covers a timekeeper's recordings for the whole month.

Treat the figures below as rough guidelines, because actual costs depend on your firm. Our estimates come close, but your actual costs can differ by up to $1.50 per timekeeper per month, depending on your firm's data. Timekeepers who record more often than in our examples cost more: each extra recording per day adds about $0.10–$0.50 per month, depending on the models you use.

Estimated costs

The estimates below assume 21 working days per month and two recording styles. A timekeeper who summarizes their work records one or two times a day, with four to six time entries per recording. A timekeeper who records after each task does so four to six times a day, with one or two time entries per recording.

Economical setup

These costs apply with GPT-4o-mini-transcribe as the transcription model, GPT-5.6-luna as the structured-output model, and recordings of typical length.

Most accurate setup

These costs apply with GPT-4o-transcribe as the transcription model, GPT-5.6-terra as the structured-output model, and every recording using the full minute.

Heavy use

A timekeeper who records after every phone call, 10 to 15 times a day, costs $0.85–$1.45 per month with the economical setup. With the most accurate setup, the cost is $3.75–$6.25 per month for short recordings of 10 to 20 seconds, and up to $7.15 if every recording uses the full minute.

Estimate for your firm

Multiply the monthly cost per timekeeper by the number of timekeepers who use verbal time entry. Most firms have a mix of recording styles, so add up each group separately.

For example, a firm has 20 timekeepers. Of these, 12 summarize their work and eight record after each task. With the economical setup, that comes to about $4–$8 per month for the whole firm. With the most accurate setup, it comes to about $21–$36 per month.

What affects your costs

Each recording goes through two AI models: one transcribes the audio, the other turns the transcript into time entries. With every recording, Legalsense also sends a list of relevant clients, matters, timekeepers and activities, so the AI recognizes the names you mention (see What data is shared with the AI provider above). This list makes up most of the cost of a recording.

That's why how often a timekeeper records matters more than how many time entries they dictate. Verbal time entry works well with either style. Summarizing once or twice a day costs less. Recording a quick note between tasks costs a bit more, but timekeepers tend to remember more of what they did.

The models also make a difference. By default, verbal time entry uses GPT-4o-transcribe for transcription and GPT-4.1 for structured output. An administrator can change them under Settings → Integrations → AI provider, in the Transcription model and Structured-output model fields. In our tests, the cheaper models GPT-4o-mini-transcribe and GPT-5.6-luna work very well in most cases.

Legalsense Support can also adjust the list of names sent with each recording. A shorter list lowers the cost, and a longer one can improve recognition. Contact Legalsense Support if you want to change the balance between the two or request a different OpenAI model.

Note: These estimates are based on our own tests and OpenAI's list prices in October 2026, in US dollars. If you use Azure AI Foundry, use them as a starting point and check Azure's pricing for your actual rates. The estimates don't include surcharges your provider may charge for zero data retention or for processing your data in the EU.

Permissions and user access

For a user to use verbal time entry, they need:

• Access to the Timesheet page (requires permission to view and create a time entry).
• Permission to view and manage their own draft time entries.

In most environments, users who can already write time will have the necessary permissions by default. If certain users cannot see or use the Voice tab after setup, contact Legalsense Support to verify their permissions.

In addition, the user must have a microphone and use a supported browser: a recent version of Chrome, Firefox, or Safari.

Troubleshooting

The AI provider configuration is not working

If an error occurs when configuring the AI provider, check the credentials for the selected provider:

• OpenAI: the most likely cause is an API key that is invalid, has insufficient permissions, or belongs to an account without payment details. Verify the key and the account setup at ⧉ OpenAI Platform.
• Azure OpenAI: check that the resource endpoint and API key are correct for your Azure resource. Check that the model name in Legalsense matches the deployment name in Azure, and that the deployment uses a supported model.

Users see the Voice tab, but cannot record

If a user reports that the microphone button is greyed out, check the following:

• The user may not have permission to manage draft time entries. Contact Legalsense Support to verify.
• The user may not be using a supported browser. Ask them to try a recent version of Chrome, Firefox, or Safari.
• The user's browser may not have microphone permission for Legalsense. Ask them to check their browser settings and allow microphone access for your Legalsense domain.

Users submit recordings but no drafts appear

A short processing delay of 10–60 seconds is normal. If drafts consistently fail to appear after waiting, it may indicate a problem with the AI provider configuration or connectivity. Contact Legalsense Support and provide the approximate time of the failed submission so it can be investigated.