HomeProductWebsite builderAutopilotPaymentsPricingGuides
Sign inStart with Soclo
← All guides · Updated 7 Sep 2026 · 8 min read

Automatic replies: what Soclo will and will not answer

Two Soclo screens: the Home screen with an AI Inbox switch showing saved answers, and the Inbox itself with an automatic replies row and filters for unread, needs reply, today and snoozed
Only the exact answers you have written send on their own. Complaints, money, promises, uncertainty and requests for a person are always handed to you.

Letting software answer customers is the point at which most people, reasonably, stop trusting a tool. So the useful thing to understand first is not what automatic replies can do — it is the list of things they are built to refuse, because that list is what makes the rest safe to switch on.

The five that always come to you

Complaints, money, promises, uncertainty, and requests for a person

These are handed to you rather than answered. The wording in the app is direct about why: complaints are yours, and Soclo will not try to smooth one over. Anything about money — a refund, a price, a deposit — goes to you. Any promise about what you will do goes to you. Anything it is not confident about goes to you.

And if someone asks to speak to a person, that message reaches you unanswered. It does not reply first and then hand over, because a reply from software to "can I speak to Sarah please?" is precisely the moment a customer decides you are not worth dealing with.

That is a deliberately conservative list, and it covers most of what actually matters in a small business inbox. What is left is the repetitive half: opening hours, delivery areas, whether you do a particular thing.

Two kinds of answer, and only one of them sends alone

This is the distinction to understand before switching anything on.

  1. Your exact answers. A question you write out, with the reply you want given, word for word — "What time do you open?" answered by "We open at 9am, seven days a week." These are yours, so they can go out unattended.
  2. Everything else. Soclo drafts from your Brand Kit and your instructions, and then waits for your tap. It does not send its own writing to a customer on its own initiative.

So the safety model is not "the AI is careful". It is that the only text sent without you is text you wrote yourself.

Setting it up

The screen is organised as four questions, and it is worth doing them in order.

  • Which accounts it may answer from. Pick explicitly — nothing is assumed, and if nothing is connected there is nowhere to answer from yet.
  • What it should say. Write the exact answers first. This is the part that does the work, and the app tells you to start here.
  • When it should come to you. The escalation rules — asked for a person, sounds unhappy — on top of the five that always escalate.
  • What it has done. The log. Check it early and often at the start.

Alongside the exact answers you can give general instructions, and the example in the app is a good model for the shape: "We're open 9 to 5, Monday to Saturday. Be friendly and brief. Never quote a price — say I'll confirm." Note what that does — it sets tone, and it forbids a specific thing.

Rehearse it before anyone real is involved

There is a no-send rehearsal, and it is the single most useful thing on this page. It runs the safety checks against sample messages — an opening-hours question, a delivery question, a refund request, someone asking for a named person, someone plainly upset — and shows you what would have happened.

Nothing is sent. The result is captured locally, and the app says so explicitly rather than leaving you to wonder. Run it, read every outcome, and confirm the escalations behave the way you would want in front of a customer you cannot afford to lose.

The test that matters: does the refund question come to you, does the upset message come to you, and does "can I speak to Sarah" arrive with no reply attached?

The limits that stop it running away

  • A daily limit per conversation. One customer cannot be answered indefinitely; the thread hits a safe limit and stops.
  • Duplicate protection. Platforms occasionally deliver the same message twice. The second one is recognised and not answered again.
  • Changes mid-draft cancel the draft. If you edit the setup while something is being written, that reply is abandoned rather than sent under the old rules.
  • A malformed setup blocks rather than guesses. If the saved configuration cannot be read, nothing is sent.
  • Stop everything. One control halts all automatic replying at once, without unwinding settings first.

Each of those is a refusal rather than a feature, which is the right way round for something operating unattended.

What to write exact answers for

The best candidates are questions where the answer never changes and getting it slightly wrong would be embarrassing:

  • Opening hours, and bank holidays.
  • Whether you cover an area — including the areas you do not.
  • Whether you do a particular service, and what you have stopped doing.
  • Parking, access, what to bring.
  • How to book, with the link.

Do not write exact answers for prices. Prices move, an out-of-date figure sent automatically is the worst possible version of a pricing error, and the instruction example in the app forbids quoting for exactly that reason.

Read the log for the first fortnight

The "what it has done" section is not decoration. Read every reply for the first two weeks and ask one question of each: would you have sent that? Where the answer is no, the fix is usually an exact answer you have not written yet, or an instruction that needs to be narrower.

After a fortnight the pattern is established and a weekly glance is enough.

What this does not replace

Automatic replies handle the repetitive layer so that the messages that matter reach you sooner and less buried. They are not a substitute for answering people. The commercially valuable part of an inbox — the quote, the awkward question, the person deciding between you and someone else — is deliberately routed to a human, and that is the design rather than a limitation.

Where to go next

Setting up your Brand Kit matters here, because drafts are written from it. What Soclo will never do without asking covers the wider boundaries, and handling enquiries from social media covers what to do with the messages that reach you.

Common questions

Will Soclo answer a complaint on my behalf?

No. Complaints are always handed to you. The app states it plainly: complaints are yours, and Soclo will not try to smooth one over. The same applies to anything about money, any promise, and anything it is unsure of.

What happens if someone asks to speak to me personally?

It stops. A message asking for a person goes straight to you, unanswered — Soclo does not attempt to stand in for you.

Does it send everything it writes?

No. Only the exact answers you have written and approved go out on their own. Anything else it drafts waits for your tap.

Can I test it without messaging real customers?

Yes, and you should. There is a no-send rehearsal that runs the safety checks against sample messages and captures the result locally without sending anything.

Can it reply endlessly to the same person?

No. Each conversation has a daily safe limit, and the same incoming message cannot be answered twice even if the platform delivers it twice.

How do I stop it immediately?

There is a stop-everything control that halts all automatic replying at once. It does not require unwinding any settings first.

Which accounts can it answer from?

Only the ones you choose. You pick the account explicitly, and every reply stays bound to that business and that provider account.

Still stuck? Tell us the screen or address, what you expected and what happened. Never send a password, verification code or full card details.

Email Soclo support →