AI for immigration consultant client intake: it may collect, it may never advise

AI can do three jobs in an immigration consultant’s client intake. It can collect the facts the CICC Code says you must confirm. It can sort the enquiry so you read it in the right order. And it can book the consultation once the paperwork is signed. It must never advise. That one rule decides the whole build, so it comes first.

I build intake agents, so read the rest as a checklist you can run on my quotes too.

The line the agent can’t cross

Section 91 of the Immigration and Refugee Protection Act says no person shall “represent or advise a person for consideration” in connection with an application under the Act. The exceptions are lawyers, paralegals and Quebec notaries in good standing, and members in good standing of the College.

As I read it, an AI agent fits none of those. Neither does your receptionist.

The College’s Code of Professional Conduct covers the person who assists a licensee. Section 38 says a licensee “must supervise and assume professional responsibility” for work done by a person who assists. It allows delegation to a non-licensee, but the licensee “must ensure that the person does not provide representation or advice” in breach of section 91.

My reading is that an intake agent should be treated the same way. You supervise it. You answer for it. Read that as my view, not as legal advice. Before it goes live, have a lawyer who knows the College read your setup.

A pitch might go like this:

“Our immigration chatbot answers eligibility questions around the clock.”

Eligibility is the advice. It’s the exact thing the licensee is paid to give. A bot that answers it on your website is answering in your name. Ask the vendor one question back: what does it say when somebody types “do I qualify”?

What intake may collect, and when

Two sections of the Code shape what intake should ask, and when.

Before the first consultation. Section 23 requires a written consultation agreement before the initial consultation. It must include your name, registration number, address, phone and email; the potential client’s name and contact information; the fee; a brief description of the College’s role as your regulator; and the purpose and scope of the consultation. Every field on that list is something an intake flow can gather and drop into a draft for you to check.

Before the service agreement. Section 24(2) lists three things a licensee must do before entering into a service agreement:

• Confirm the client’s full name, home address, and personal phone number and email.
• Provide the client with a draft of the service agreement.
• Verify whether the client has entered into a service agreement with another individual who is authorized to provide representation or advice under section 91.

That last check is the one I’d put in every intake form. It’s a single question. It’s easy to forget on a busy Tuesday.

Never at intake. The case analysis. Travel history, refusals, the reasons behind a gap. Those belong in the consultation, with you, after the agreement is signed. An intake form that asks for all of it up front is collecting sensitive personal information before anybody has agreed to anything.

Where the answers go

Section 28 of the Code says a licensee must keep client information confidential “indefinitely.” It permits disclosure to “a person who is assisting the licensee.” My reading is that an AI tool and its vendor sit there, which is why the next part matters.

PIPEDA lets you send personal information to another company for processing. Canada’s Privacy Commissioner says plainly that PIPEDA “does not prohibit” transfers to another jurisdiction. It says organizations “are held accountable” for each outsourcing arrangement. It expects “contractual or other means” to give comparable protection while a third party processes the information. And it expects you to advise customers that their information “may be sent to another jurisdiction for processing.”

Two practical results follow.

First, the intake notice says where the answers go. One sentence, written before the first question.

Second, your service agreement already has to describe how you keep client information confidential. That’s section 24(3)(q). If an AI tool touches the file, it belongs in that description. The wider picture is in the note on PIPEDA AI compliance in Canada.

Three kinds of intake tool

They aren’t ranked. Each fits a different practice.

What it is	What it does	On this site
A form that files itself	Collects the section 24(2) facts, files the enquiry in the right folder, emails you	Workflow automation for an Ontario office, from $2,500
An agent grounded in your own documents	Answers questions about fees, documents to bring and booking, from text you wrote	$4,000 to $7,000
An agent that qualifies and books	Asks your questions, sorts by program, books once the consultation agreement is signed	$7,000 to $12,000
A box that says hello	Nothing	$0

The two agent rows come from my pricing for AI agent development for Ontario firms. Running costs are extra, between $80 and $600 monthly, set by volume. You see that number before you agree to a build.

A practice that gets twenty enquiries a month probably wants row one. The agent rows earn their price when the same questions arrive every week and nobody’s there after 5pm to answer them.

Month one looks fine

The demo works. In month one the agent answers “what documents do I bring?” from your checklist, and books the slots you opened. Nobody complains.

Then somebody types near midnight: “My permit expires on Friday. Can I still apply from inside Canada?”

An agent that answers from a general model will try. It’ll sound sure of itself. It’s now giving advice, in your name, on a question with a deadline attached.

An agent grounded in your own documents finds nothing in them that covers it. It says so, takes the person’s details, and flags the enquiry for you first thing. Handing off to a person with the conversation attached is the feature you’re paying for.

Test that path before you test anything else. Type the hard question yourself.

What I won’t build for you

A bot that tells a visitor whether they qualify. Or one that scores their chances.

The want behind it is fair. People asking at midnight are worried and want an answer now. But section 44 of the Code says your marketing must not “guarantee the success of an application.” A bot that says “you look like a strong candidate” is a step from that. Section 27 goes further. When you think a case has little hope, you owe the client a written opinion with your reasoning. That’s your judgment. It isn’t a chat reply.

The fix for the midnight visitor is a fast booking, not a fast verdict.

Toronto practices

None of the checks above change by city. A practice in Mississauga, Brampton or Ottawa should run the same ones.

In its second-quarter 2026 survey, Statistics Canada found 19.2% of businesses used AI to produce goods or deliver services over the previous 12 months. Professional, scientific and technical services sat at 32.4%. I read that as a hint that the professional firms around a Toronto practice are likelier than most to be trying this already. If you’re weighing it there, my page as an AI consultant serving Toronto practices covers where I’d start.

Before you buy any of it

Count the enquiries. How many arrive in a normal month? How many come after hours? How many turn out to be for a program you don’t handle? My guess is that nobody in the office has counted it yet. The count decides which row of the table you need.

If you already know, skip my assessment and ask for a quote. If you don’t, a $999 AI business assessment for an Ontario practice names three jobs, costs each in hours and dollars, and lists the ones I’d tell you not to do. The fee is deducted from any build. It’s written so another supplier can quote from it. If one does it cheaper, the document did its job.