Prompt templatesCopy Anchor Link

This library collects prompts that have worked well in real matters. Use them as starting points: copy a template into your scan, swap in your case-specific terms (parties, dates, issues, scope), and validate on a small sample before scaling.

If you are new to writing prompts, read Prompting Overview first, then come back here for working examples.

Your organization's own templatesCopy Anchor Link

When your organization admins have written their own templates in the Admin Portal, Quickstart shows them first, in a Your organization section above this library. Use them exactly like the templates below: expand one to read the prompt, then choose Insert Template.

• Each shows its title, a short description (or the first line of its prompt), its category and the date it was last updated.
• Templates in your interface language come first. One written in another language carries a small EN, FR or DE tag.
• They have no help article here: they belong to your organization, and your admins keep them up to date.
• The section appears only when your Discover portal is connected to your organization in the Admin Portal and your organization has at least one active template. With permission enforcement on, your role also needs Use prompt templates. This library is always available.

See Prompt Templates in the Admin Portal for how admins create, archive and delete them.

Describe a prompt with the Prompt GeneratorCopy Anchor Link

At the top of Quickstart, the Prompt Generator writes a prompt from a description of what you want to find or decide. Pick one of the case's approved contexts under Case context (or none), describe the task, and click Generate Prompt.

With a context, three steps show its progress, each ticked when it is done:

1. Reading the Case Context, then, for example, Read 4 sections and 48 matter facts.
2. Writing the prompt, with a timer, then Prompt written.
3. Checking every matter fact is in, then, for example, Checked 48 matter facts · 3 added. Claira compares the draft with the context's facts and adds every one your task uses that the draft left out.

Without a context there is one step, Writing the prompt. The prompt then goes into your prompt box, and Prompt History records which version of the context it was built from (see Prompts written by AI).

The Prompt Generator keeps its instructions short and its matter facts complete: it writes in every name and its variants, the dates, amounts, identifiers and issues your task depends on, so a long context gives a long prompt. It leaves out what your task does not ask for, such as the privilege rules in a relevance prompt. If the check cannot run, the prompt still goes in and Claira says so: review it before you scan.

Use a template with your Case ContextCopy Anchor Link

Every template in Quickstart, your organization's own included, has a Use with context button, on its card and when you expand it. It fills the template in from your matter's approved Case Context, using the Prompt Generator.

1. In Claira, open Quickstart and find the template in the library.
2. Click Use with context.
3. If the case has one approved context, Claira uses it and starts right away. If it has several, pick one (the case default is preselected) and click Generate Prompt.
4. The same three steps as the Prompt Generator show its progress: reading the context, writing, and checking every matter fact is in.
5. The filled-in prompt goes into your prompt box. Check it before you scan.

What Claira does with the template, every time:

• It keeps the template's task, wording, output values (such as YESREL / NOTREL) and output format exactly as written, so results land in your fields the way the template promises.
• It fills in the template's bracketed placeholders (the case description, known counsel) from the context, and removes any the context has nothing for.
• It names a jurisdiction only when the context states one. Otherwise the wording stays neutral.
• It adds any other matter facts the task needs (people and their name variants, issues, dates) in one Matter details section, just before the output instructions, every one of them in full.

The button is unavailable until the case has an approved context. To describe a prompt in your own words instead, use the Prompt Generator at the top of Quickstart.

Multi-Code templatesCopy Anchor Link

A Multi-Code template has Use with context too, and it works the same way. Claira fills in the shared instruction and every field's prompt together, from one reading of the context, then loads the whole template into the Multi-Code card with every field filled in. Field names, answer types and destination matching are unchanged: you get the rows Load into Multi-Code would have given you, with your matter's facts in them.

Where the facts land matters for what you pay: they go in the shared instruction, which every field is read with, rather than being repeated in each field's prompt. Because the fields are filled in together, the tags one field decides are the tags another lists.

This matters most for relevance and privilege, where the template cannot know your issues, your counsel or your governing law. Objective coding templates take it too, though they mostly need no matter facts.

A prompt filled in from a context needs the same check as any other: run it on a 10 to 25 document sample before a bulk scan.

GeneralCopy Anchor Link

Document-wide tasks that apply across most review types.

• Summarize document: Produce a concise summary suitable for first-pass review.
• Chronology summary: Summarize a document with an emphasis on dates and event sequence.
• Translate content: Translate foreign-language content into the working language of the review.
• Summarized translation: Translate and summarize together for faster triage of foreign-language documents.

PrivilegeCopy Anchor Link

Identify potentially privileged communications and their supporting context.

• Identify privilege types: Classify documents by privilege category (attorney-client, work product, joint defense, etc.).

RelevanceCopy Anchor Link

Assess responsiveness to the issues in your matter.

• Assess relevance: Determine whether a document is relevant to a defined issue.
• Relevance with confidence: Same as above, but with an explicit confidence rating to support QC.
• Relevance and issues: Tag documents both for relevance and for which issue(s) they touch.

Objective codingCopy Anchor Link

Extract structured metadata fields from the document body itself.

• Identify document type: Classify the document into a fixed list of types (contract, email, invoice, etc.).
• Extract dates: Pull all dates mentioned in the document body.
• Extract author: Name the individual and/or organisation the document is from (email sender, invoicing company, letter writer).
• Extract recipients: Name the individual(s) and/or organisation(s) the document is directly addressed to, excluding Cc/Bcc.
• Extract titles: Take the document's own title (for an email, its subject line), or write a short descriptive one when it has none.
• Email From: Pull sender email addresses from email content.
• Email To: Pull recipient email addresses from email content.
• Email Cc: Pull carbon-copy email addresses from email content.
• Email Bcc: Pull blind-carbon-copy email addresses from email content.
• PDF Unitization: Find the document boundaries inside a compiled PDF and output the page-range string that splits it apart in the Nuix Discover image viewer.
• PDF Unitization (Custom Split): Same output, but you describe where to split: by individual, record type, chapter, email versus attachment.

InvestigationsCopy Anchor Link

Targeted prompts for investigative reviews.

• Comprehensive PII surface forms: Surface unique forms of personal information across the corpus.
• Special category data (GDPR Article 9): Flag documents that contain the sensitive categories of personal data protected by GDPR Article 9, and name each category found.

Multi-Code templatesCopy Anchor Link

Each of these fills the whole Multi-Code card in one step: a shared instruction and one field per output, each with its own prompt. Quickstart lists them in a Multi-Code folder, first, when you open it from the Multi-Code card; choose Load into Multi-Code for the template as written, or Use with context to have Claira fill every field in from your approved Case Context first (above). See Multi-Code templates for how destination fields are matched.

• Objective coding: Title, author, recipients, document type and date in one pass, each field using its objective-coding template's prompt.
• Relevance and issues: The relevance call, the issue tags, a sentence per issue and a confidence score, each in its own field.
• Privilege and types: The privilege call, the privilege types and the supporting excerpts, each in its own field.

Next stepsCopy Anchor Link

• Build your own from scratch with Custom Prompts.
• Tune model behavior in Model Settings.
• See how to wire a template into a full review in Workflows.

---

Need help? Contact us at support@claira.to.