Verified by whom? A teardown of trust in Clio's AI

Clio's AI shows its sources well. What a small firm cannot yet do is prove that a lawyer checked them. Courts have started asking for that proof, and Clio is well placed to supply it.

What this is based on

This is an outside view. I have not used Clio Work or Manage AI and I have no inside information. I read Clio's help centre, product pages and press coverage, bar guidance and court decisions on 3 October 2026. At Frontline Labs I led an AI legal assistant inside Microsoft Word and Outlook, and most of that work was the trust layer.

The product

Clio completed its $1 billion purchase of vLex in November 2025. With it came Vincent, an AI assistant, and a library of more than a billion documents.

• Manage AI works inside Clio Manage. It pulls deadlines from court documents into the calendar and drafts invoices.
• Clio Work is the research and drafting workspace, with Vincent inside. It became available as a standalone product in April 2026.
• Clio for Word drafts and revises inside Microsoft Word. Its beta opened in May 2026.
• In September 2026 Clio said Vincent's Canadian collection had grown beyond 567,000 judicial decisions.

What Clio already does well

The sourcing is strong. Clio Work shows a Legal Authorities panel listing the sources behind an answer, with citations, relevance scores and excerpts, and it tells users to check each authority before relying on it. Clio for Word proposes edits as tracked changes. Manage AI keeps an event log, and Clio states that customer data is never used to train external AI models.

The gap

Three details in the public documentation stood out.

1. One switch for the whole firm. Administrators can turn AI off for everyone. I found no per-user or per-role control.
2. A mode that skips review. Clio for Word lets a user switch from tracked changes to direct edits, which applies changes immediately.
3. A short audit trail. The Manage AI log covers the last 90 days. I found no equivalent described for Clio Work or Vincent output.

None of this is a flaw in the AI. Together it means a firm cannot answer a simple question six months later: who checked this document, and what did they check?

Why it matters now

The professional rules place responsibility on the lawyer. The Law Society of BC says lawyers are ultimately responsible for all work product they oversee, including work produced by technology. The American Bar Association's Formal Opinion 512 warns against relying on AI output without an appropriate degree of independent verification.

Courts are enforcing it. In Zhang v. Chen, a BC lawyer who cited two cases invented by ChatGPT was ordered to pay costs personally. In Ko v. Li, an Ontario court pointed to the rule that makes counsel certify that every authority in a factum is authentic.

Small firms are the least prepared. Clio's 2025 Legal Trends Report found that 53% of legal professionals say their firm has no AI policy or they are unaware of one. A solo lawyer has no knowledge-management team to write that policy. The software has to carry it.

Competitors are moving. Thomson Reuters says CoCounsel now checks whether each cited authority supports the claim and highlights the passage.

What I would build first

A verification record attached to every AI-assisted document.

1. Track each citation. Suggested by AI, opened by a lawyer, confirmed by a lawyer. The document shows how many remain unconfirmed.
2. Record who and when. The record names the reviewer and the time, and it stays with the matter for as long as the file does.
3. Give the firm a few policy switches. Who may use AI drafting. Whether direct edits are allowed. Whether a document can be marked final with unconfirmed citations.
4. Make it exportable. One page a lawyer can keep on file, or use to support a certification that the authorities are real.

At Frontline Labs, the legal teams we built for asked for source verification, human review and a full audit trail before they would adopt the assistant. I expect small firms to want the same, in a form they can switch on without a project.

What I would not do

I would not sell this as a guarantee. A record shows that a person checked. It cannot make the check correct. I would also keep the everyday flow fast. A solo lawyer drafting an email should not meet a compliance form.

How I would measure it

• Share of AI-assisted documents finalised with every citation confirmed.
• Share of firms that turn on at least one policy control.
• Adoption of Clio Work among firms that had AI switched off.
• Time added per document, as the guardrail.

What I cannot see

Clio may log more than its help centre describes, and Clio for Word is still new. I cannot see how often direct edits are used. I would start by asking what firms tell the sales team when they decline to turn AI on.