You launch your first outbound SMS campaign. Fifty messages out. Three replies come back: "STOP", "No thanks", and "fuck off mate". Your AI agent freezes on the third one because you only programmed it to recognise "STOP". By the time you manually mark that contact as opted-out, you've sent another message. The Spam Act 2003 clock is already ticking, and your non-compliance just became a paper trail.

Australian businesses run 2-way SMS campaigns every day through platforms like VoxReach. Most get consent right. Where they fail is the exit: the unsubscribe pathways people actually use, not the ones you wish they'd use. One angry reply that your bot misses can cost you $2.75 million under the Spam Act maximum penalty, but more realistically it costs you the customer relationship, a formal complaint to ACMA, and a please-explain letter from your lawyer.

What the Spam Act actually says about opt-out

Section 16 of the Spam Act 2003 requires every commercial electronic message to include a functional unsubscribe facility. SMS counts. The unsubscribe method must be free, simple, and obvious. If someone sends "STOP", you must action it within five business days. If they send "unsubscribe", same rule applies. If they send "remove me", same again.

The law does not care that your AI agent speaks eighteen languages and handles complex appointment booking. If it cannot parse "nah I'm good" as an opt-out, you are non-compliant the moment the next message goes out. ACMA does not publish a master list of every unsubscribe phrase, which means you build your own or you risk missing one.

Here is what we see in real VoxReach transcripts, sanitised for privacy but otherwise verbatim:

• STOP
• Stop
• stop
• Unsubscribe
• Remove
• No thanks
• Not interested
• Fuck off
• Delete my number
• Take me off this list
• Don't contact me again
• Leave me alone
• Piss off
• I didn't sign up for this

Every single one of those is an unsubscribe instruction. Your bot must recognise all of them, or you automate your own non-compliance.

The soft-decline problem: when "no" is not "STOP"

Not every "no" is an unsubscribe. Someone replies "no thanks, I've already booked" to an appointment reminder. Your agent must distinguish between declining the specific offer and opting out entirely. This is where keyword matching fails and context parsing matters.

The safest pattern we have found: if the reply contains clear refusal language and no positive engagement, treat it as opt-out. If the reply answers a question or acknowledges the message content, it is a response, not an unsubscribe. A VoxReach client running a dental recall campaign saw this last month: "I moved interstate" came back. The agent correctly logged it as a status change, not an opt-out, because the patient was explaining why they could not attend, not asking to be removed.

When in doubt, ask. If your AI cannot confidently classify the intent, it should send one clarifying message: "Just to confirm, would you like us to stop all future messages? Reply YES to opt out." Then wait for the answer. That one extra message is cheaper than a Spam Act breach.

Silent STOP: the compliance feature nobody sells

When someone says "fuck off", your bot should not reply "You have been successfully unsubscribed. Have a great day!" That tone-deaf response is how screenshots end up on Reddit. The correct behaviour is silent acknowledgment: mark the contact as opted-out, suppress all further outbound messages, log the event internally, and send nothing back. No confirmation. No pleasantries. Just stop.

Every VoxReach voice and SMS agent includes this by default. If the system detects unsubscribe intent, it stops immediately and flags the contact in your CRM. The agent does not attempt to salvage the conversation or offer alternatives. One Brisbane solar installer told us this saved them when a prospect replied with a string of expletives after a badly-timed cold SMS. The agent went silent, the contact cooled down, and six weeks later they called back through the same number to book a quote. If the bot had sent a chipper "Sorry to see you go!" message, that callback never happens.

The real cost of getting this wrong

ACMA issued thirteen formal warnings for Spam Act breaches in 2023. Most were about consent. A few were about unsubscribe failures. The fines scale by harm: a single missed opt-out to one person is unlikely to trigger enforcement. A pattern of missed opt-outs across fifty people gets you a phone call. A pattern across five hundred people gets you an infringement notice.

The bigger cost is not the fine. It is the customer you lose, the Google review they leave, and the referrals that never arrive because your brand is now "that company that kept texting me after I said stop". One bad unsubscribe experience destroys the trust you spent months building.

What to do before you send the next SMS

Review your outbound SMS workflow today. Log in to your platform. Look at the unsubscribe detection rules. If you see only "STOP" and "UNSUBSCRIBE" in the trigger list, you are exposed. Add the variations above as a starting set. Train your AI agent to recognise refusal intent, not just keywords. Set the default behaviour to opt-out if uncertain. Test it: send yourself a campaign message and reply with "nah" or "no thanks". See what happens.

If your current platform cannot handle this, you are running compliance risk every time you send. VoxReach includes multilingual opt-out detection and silent-stop behaviour out of the box because we built it in Sydney under Australian law. Sign up free at app.voxreach.com.au/signup, thirty minutes of calls included, no card required. Your first test campaign will show you exactly how the agent handles every opt-out variation before you put real contacts at risk.