AI Email Follow-Up That Sounds Like a Person Wrote It (and Stays Inside CASL)

The short answer

AI email follow-up works when the first reply goes out within minutes, answers the specific thing the person asked, and reads like a short note from a named human, because a human wrote the template and the AI only filled in the specifics. It breaks when it pretends, pads, or keeps sending after the person stopped caring. In Canada the sequence also has to live inside CASL: consent, clear identification, and an unsubscribe that works within ten business days. Here is how to set one up that passes both tests.

Why most automated follow-up reads like a robot

Three habits give it away. The opening line restates the person's own message back to them. The body contains a paragraph of benefits nobody asked about. The sign-off promises "someone from our team" will be in touch, which means nobody will. None of these are AI problems; they are template problems that AI makes faster to produce.

The fix is to give the model less freedom and more facts. A good follow-up has a named sender, one answer, one question or one next step, and a reason to reply that is specific to the person. That structure can be written once by you and filled in automatically forever.

Speed does the rest. Harvard Business Review's audit of company response times, The Short Life of Online Sales Leads, found that most companies were not responding nearly fast enough, and Invoca's 2026 buyer survey of 693 US consumers found 79% would switch to a competitor that responds faster. Our speed-to-lead post covers the evidence; this one is about what the fast reply should say.

How to build AI email follow-up that reads human

1. Decide what the AI is allowed to change

Fixed by you: the sender name, the structure, the tone, the offer, the sign-off, the unsubscribe line. Filled by the AI: the person's name, the specific question they asked and its answer from your approved content, the relevant service or product, and one detail drawn from their message. Nothing else. The model does not get to invent a discount, a delivery date or an opinion.

2. Write the first reply as if you had two minutes

A real person with two minutes writes something like: "Hi Priya, thanks for asking about a basement bathroom. Short answer: yes, we do them, and most run four to six weeks once permits are in. Two quick questions so I can give you a real number: is there plumbing already roughed in, and roughly when would you like to start? I'm here until 5 today, or you can grab a 15-minute call here. Mark." Four sentences, one answer, two questions, one next step, a name. That is the template. The AI changes the nouns.

3. Set the timing on the person's behaviour, not a calendar

• Reply 1: within five minutes of the enquiry, any hour. After hours it says so and names the morning callback time.
• Reply 2: only if there is no response, two business days later, and only to add something useful: a photo of similar work, a short answer to a question they did not ask but everyone does, a reminder of the booking link.
• Reply 3: five business days after that. Short, warm, and final: "I'll leave it here so I'm not cluttering your inbox. If timing changes, reply any time."
• Stop conditions: they replied, they booked, they unsubscribed, or they bounced. Any reply pulls the thread to a human.

Three messages is plenty for an enquiry. Longer sequences belong to newsletters, and those are a different legal category.

4. Make a human the owner of every thread

The AI drafts; a person owns. For high-value enquiries, the first reply can be drafted and held for one-click approval during business hours. Our post on where approvals belong explains how to decide which threads need a hand on them.

The CASL rules the sequence must respect

Canada's anti-spam legislation applies to commercial electronic messages, which includes a follow-up that encourages someone to buy. The Act sets out the three things your automation needs to get right; our plain-English CASL guide goes deeper and the CASL glossary entry has the short version. This is general information, not legal advice.

Requirement	What the Act says	What to build
Consent	Express consent, or implied consent through an existing business relationship: an inquiry within the previous six months, or a purchase within the previous two years (s. 10(10))	Record the date and basis of consent on the contact. Stop the sequence automatically at six months from an inquiry unless express consent was given.
Identification	Each message must identify the sender (and anyone it is sent on behalf of) and give contact information that stays valid for at least 60 days (s. 6(2), 6(3))	Real name, business name, mailing address and a reply route in every template, including the AI-filled ones.
Unsubscribe	A mechanism that works at no cost through the same electronic means, honoured without delay and no later than ten business days (s. 11)	One-click unsubscribe that writes to the CRM and halts every sequence immediately, not just this one.
Penalties	Administrative monetary penalties up to $1,000,000 per violation for an individual and $10,000,000 for any other person (s. 20(4))	Reason enough to test the unsubscribe path before launch.

One subtlety: a direct reply to someone's question is a conversation, and nobody argues with that. The sequence becomes marketing the moment it starts promoting rather than answering. Design replies two and three to be useful first, and keep the consent basis recorded either way.

Deliverability: the rules mailbox providers enforce

A follow-up that lands in spam did not happen. Google's Email sender guidelines require every sender to authenticate with SPF or DKIM, use TLS, and keep the spam-complaint rate in Postmaster Tools below 0.3%, with 0.1% recommended. Senders of more than 5,000 messages a day to Gmail must also publish DMARC, align the From domain, and offer one-click unsubscribe on marketing mail. Most small businesses never reach bulk volume, but authentication is cheap and the complaint-rate rule is a useful discipline: if people are marking your follow-ups as spam, the sequence is too long or too salesy. The email deliverability glossary entry explains the acronyms.

Practical rules: send follow-ups from a real person's mailbox on your own domain, not a "noreply" address; keep the volume per mailbox modest; and let replies land in a real inbox so the thread looks like correspondence to the mailbox provider, because it is.

Measuring whether it sounds human

Skip open rates; they are unreliable and tell you nothing about tone. Track the reply rate to message one, the booking rate across the sequence, the unsubscribe and spam-complaint rates, and read ten threads a week. If people reply "thanks, that answers it", the sequence is working even when they do not buy. If they reply "is this a bot", rewrite the template. For the questions a follow-up should ask, our guide to AI lead qualification has a scoring model that pairs with this sequence, and marketing automation describes how we wire the two together.

Key takeaways

• Write the template like a person with two minutes; let the AI change only names, the question and the answer.
• Reply within five minutes, then at most two more messages, each adding something useful, with hard stop conditions.
• Record the consent basis, identify yourself fully in every message, and make unsubscribe work within ten business days.
• Authenticate your domain, watch the complaint rate, and read real threads every week.

Frequently asked questions

Is AI-written email follow-up legal in Canada?

Yes, provided the messages comply with CASL like any other commercial email: consent (express, or implied for six months after an inquiry or two years after a purchase), full sender identification, and a working unsubscribe honoured within ten business days. The law regulates the message, not whether software drafted it. This is general information, not legal advice.

Should I tell people the email was written by AI?

If a named person reviewed and owns the message, most businesses treat it as that person's email. If the message is fully automatic, say so plainly ("this is an automated reply; Mark will follow up by 9 a.m."). Pretending an instant 2 a.m. reply came from a human erodes trust the moment someone notices.

How many follow-up emails should an automated sequence send?

For an enquiry, three: an immediate answer, one useful nudge after two business days, and one short closing note five business days later. Stop at once if the person replies, books, unsubscribes or bounces. Longer sequences belong to newsletters, which need express consent and a different rhythm.

Sources

1. An Act to promote the efficiency and adaptability of the Canadian economy (S.C. 2010, c. 23), full text — Justice Laws Website, Government of Canada
2. Email sender guidelines — Google Workspace Admin Help, accessed October 2026
3. The Short Life of Online Sales Leads — Harvard Business Review, March 2011
4. New Invoca Study Finds Consumers Won't Wait. AI Gives Brands the Speed to Win the Sale. — Invoca press release via Customer Experience Dive, 9 June 2026

How Flowmatix can help

We put this into practice for businesses in Toronto and around the world through our Marketing Automation, email marketing & newsletters services. Want to know what it would take for your business? Get a personalised quote in about 3 minutes.