How to get a useful Claira prompt from CoCounsel or Harvey

CoCounsel and Harvey have earned a real seat at the table in legal work. They are good at drafting research memos, comparing case law, and helping a lawyer think out loud about a matter, and both have built genuine document-analysis surfaces of their own. Harvey's Vault will hold tens of thousands of files, run review tables, and extract data points at scale. CoCounsel's Review Documents skill handles due diligence, internal investigations, and discovery analysis on uploaded sets. None of that is in dispute, and we don't pretend otherwise.

What neither of them does is live inside Nuix Discover with you. That is the job we built Claira for. The moment your matter has two hundred thousand emails in a workspace, coding panels open in front of a review team, a privilege log being maintained alongside the work, and a production deadline that needs to flow through the same platform, the standalone-AI model breaks. Claira sits in that environment, reads every document the reviewer would have read, and writes its decisions back into the coding panels your team and your QC process already trust.

That distinction is worth holding in mind because it shapes how the tools work together. The strategic thinking you have already done in CoCounsel or Harvey can become the cleanest, most case-specific input your document review has ever had, if you know how to translate it. This post is about that translation.

Two tools, two jobs (mostly)

CoCounsel and Harvey work upstream of review. A lawyer asks, "what are the elements of misrepresentation under Ontario law" and gets a structured answer with citations. They ask, "summarize the key allegations in the statement of claim" and get a clean draft they can edit. Both can also push further into document analysis - Harvey will run a review table over a few thousand contracts, CoCounsel will summarize a stack of depositions - and they do that work well within their own walls.

Claira works downstream, where the full corpus lives, inside the review platform your team already uses. Every email, every contract draft, every chat log gets read once and coded against the same case context, with the rationale captured in Nuix Discover right next to the document. That is a different job. It needs a prompt that knows the case but is written for evidence at scale, not for analysis of a curated set.

These are not winner-take-all categories. Most matters use all three roles - research and synthesis, mid-scale document analysis, and corpus review - and the firms getting the most out of AI are the ones who match each tool to the part of the workflow it was designed for, then make the handoff intentional. Our broader take on this lives in our story on a practical philosophy of AI-assisted review.

Start with the work you have already done

Before you write a Claira prompt, gather the artifacts your team has already produced in CoCounsel or Harvey for this matter. Three of them tend to do most of the work.

The first is the case theory document. This is the short narrative that tells the story of the case from your client's perspective: what happened, who knew what, what the bad acts were, and what the defenses are. CoCounsel and Harvey are excellent at producing a tight version of this from your pleadings and key correspondence. If your team already has one, use it. If not, generate one.

The second is the issues list. This is the structured set of legal and factual issues the case turns on. Limitation periods, notice provisions, specific contractual clauses, the duty of good faith, fraud indicators - anything that needs to be coded against in review. Both tools produce these well from statements of claim and key authorities.

The third is the language map. Every case has its own vocabulary. Code names for projects. Specific deal terms. Internal acronyms. The opposing parties' names and entities. Your CoCounsel or Harvey transcripts often surface this language because they were prompted with documents that used it. Pull it into a list before you go any further.

If your firm does not run CoCounsel or Harvey, which is common for government departments, in-house teams on a different stack, or litigation boutiques on a leaner tech budget, you can produce all three artifacts directly from your pleadings and key documents with whatever drafting workflow you already use. The artifacts matter more than the tool that produced them.

Translate strategy into a Claira-ready prompt

A Claira prompt is not a question. It is an instruction for an AI reviewer that is going to read every document one at a time and tell you whether it matters and why. That changes the shape of what you write.

Take your case theory and reduce it to a one-paragraph context block. Plain language, no legal hedges, no footnotes. This becomes part of your Case Context, which Claira applies to every scan so the model knows the matter before it sees a single document. From your issues list, draft a numbered set of categories your reviewers care about: relevance, privilege, hot documents, specific issues like notice of breach or knowledge of fraud. From your language map, give Claira the vocabulary to watch for and the names to recognize.

When you put it all together, you are giving Claira the same briefing you would give a junior associate on day one of the matter, except the briefing is structured and the work it produces is consistent across millions of pages. Our team has put together a walkthrough of writing custom prompts that shows exactly how to format these inputs so the model uses them well.

Test, refine, and apply at scale

Do not run a 200,000 document scan with a brand new prompt. The discipline that makes this work is the same discipline that has always made review work: pilot, sample, calibrate, then scale. Pull a representative sample of around one hundred documents you already understand. Run your prompt against it. Read what Claira returns and ask the question every senior reviewer asks of their team: did the AI catch what I would have caught, and did it explain itself in a way I can defend.

If the answer is yes, run it wider. If it is no, the prompt needs work. The most common fix is not in the issues list. It is in the case context. The model needs more facts and less abstraction. Feed it back into CoCounsel or Harvey, ask for a tighter case narrative, and try again. After two or three iterations, the prompt usually settles into something that produces consistent, well-reasoned coding decisions across the entire corpus.

Where each tool earns its keep

The market often frames AI tools as winner-take-all. We see it differently, but with clear lines.

Harvey and CoCounsel earn their keep on legal research, drafting, and analysis of curated document sets in the thousands. That is the work a partner or senior associate would once have done themselves with a stack of binders and a research database, and these tools genuinely accelerate it. Claira earns its keep when the corpus is too large to curate, when the workflow has to live inside Nuix Discover, and when every coding decision needs an audit trail your QC team and opposing counsel can defend.

When you connect the three intentionally, the work product is more than the sum of the parts. The case theory is consistent across deliverables. The vocabulary is shared. The privilege calls trace back to the same definitions. That coherence is what makes a review defensible when opposing counsel asks how you used AI, and it is what justifies the AI line on the bill, because each tool is doing work that would otherwise have cost more in human hours than it does in software.

If you want to see what this looks like inside your environment, you can book a working session with us and we will walk through it with your data on screen. Bring the CoCounsel or Harvey outputs your team has already produced, or just your pleadings and key documents if you do not use those tools, and we will show you how they translate into a Claira prompt that does the rest.