Customer feedback over iMessage: tapback surveys and voice notes

Who uses iMessage for feedback surveys?

iMessage feedback surveys suit teams that already talk with customers in Messages and want one score per interaction instead of another form. ZeroSettle is an iMessage API: a real phone number on real iMessage, running on Apple hardware we operate for you in a server facility, that your code texts customers from and hears back on through a bearer token and signed webhooks.

• Support teams scoring each closed ticket (CSAT). ZeroSettle's support docs put it plainly: A ❤️ on “Fixed, sorry about that” is your CSAT survey.
• Service businesses asking how a visit or job went.
• Product teams that would rather collect a 60-second voice note from opted-in users than book a 30-minute interview.

A tapback answer is one tap in a thread the customer already has open, with no link to open and no form to finish. More workflows are in the use-case hub.

How iMessage feedback surveys work, step by step

A closed ticket triggers one send, and the customer's tapback comes back to your webhook naming the survey message; a 👎 gets one follow-up that invites a voice note.

1. Ask once, in the same thread. Call POST https://portal.zerosettle.io/api/imessage/send with “How did we do? Tap 👍 or 👎 on this message.” and store the returned id with the ticket.
2. Score the tapback. ZeroSettle POSTs a signed reaction.received event whose message.reaction is like or dislike (count a love as a 👍) and whose message.target_id is the survey's id. The other tapback types are listed in what an iMessage API can send.
3. Keep the latest answer. Messages allows one tapback per person per message, which they can change or remove. A retraction arrives with the remove_ prefix (remove_like), so keep the newest event by message id.
4. Follow up on a 👎, once: “Sorry about that. What should we have done better? A quick voice note is fine.”
5. Transcribe the voice note. It arrives as message.received with kind: "audio", an audio.url to an .m4a (AAC) file and duration_ms. Fetch it with the same bearer token and run it through your speech-to-text provider, as with technicians' voice-note job reports.
6. Thank them with a ❤️ on the voice note (POST /api/imessage/reactions, "reaction": "love") and one closing text.

The same mechanics for user research

Product teams ask opted-in users “Got 60 seconds? Reply with a voice note: what almost made you quit last week?”, tag the transcripts and thank each person with a ❤️. The invite needs no link, which matters: a link in a first message to someone who has never texted your number is not tappable. Coaches use tapbacks the same way in client check-ins.

Answers that aren't a clean 👍 or 👎

Example conversation: a support survey after a fix

This thread shows a 👎, the voice-note follow-up and the thank-you, from the support team's side.

Setup: tapback scoring and voice-note transcription

The survey needs one send when the ticket closes and a receiver that scores tapbacks by target_id and fetches voice notes, built from the webhooks and attachments sections of the ZeroSettle docs.

Send the survey

BASE=https://portal.zerosettle.io
curl -s -X POST -H "Authorization: Bearer $ZEROSETTLE_TOKEN" -H 'Content-Type: application/json' \
  -d '{"to":"+15555550142","text":"Glad we got that sorted! How did we do? Tap 👍 or 👎 on this message."}' \
  $BASE/api/imessage/send | jq .id    # store this id with the ticket

The call returns once the line hands the message to Apple, normally in 2–5 seconds, so send from a background job. A recipient_opted_out error means the customer texted STOP: skip the survey.

Score tapbacks and transcribe voice notes

# feedback.py: run with  uvicorn feedback:app --port 8100
import hashlib, hmac, json, os, time
import httpx
from fastapi import BackgroundTasks, FastAPI, Request, Response

app = FastAPI()
BASE = "https://portal.zerosettle.io"
AUTH = {"Authorization": f"Bearer {os.environ['ZEROSETTLE_TOKEN']}"}
SECRET = os.environ["ZEROSETTLE_WEBHOOK_SECRET"]
SCORE = {"like": 1, "love": 1, "dislike": -1}  # a ❤️ on the survey counts as a 👍
seen = set()                                   # use your database in production

def verify(header, raw, tolerance=300):
    try:
        parts = dict(p.split("=", 1) for p in header.split(","))
        t, v1 = int(parts["t"]), parts["v1"]
    except Exception:
        return False
    if abs(time.time() - t) > tolerance:
        return False
    expected = hmac.new(SECRET.encode(), f"{t}.".encode() + raw, hashlib.sha256).hexdigest()
    return hmac.compare_digest(expected, v1)

async def call(method, path, body=None):
    async with httpx.AsyncClient() as c:
        return await c.request(method, BASE + path, headers=AUTH, json=body, timeout=90)

async def on_tapback(m):
    s = surveys.by_message_id(m["target_id"])  # your DB: survey message id -> ticket
    if s is None or m["id"] < s.last_reaction_id:   # last_reaction_id starts at 0
        return                                 # not a survey, or an older event arriving late
    s.last_reaction_id, r = m["id"], m["reaction"]
    if not r.startswith("remove_"):
        s.reaction = r                         # one tapback per person: the newest replaces the last
    elif r[7:] == s.reaction:
        s.reaction = None                      # they took the answer back
    s.score = SCORE.get(s.reaction)            # 1, -1, or None (unscored)
    if s.score == -1 and not s.followed_up:
        s.followed_up = True
        await call("POST", "/api/imessage/send", {"to": m["handle"],
            "text": "Sorry about that. What should we have done better? A quick voice note is fine."})
    s.save()

async def on_voice(m):
    s = surveys.awaiting_voice(m["handle"])    # a 👎 survey we followed up on, else None
    if s is None or not m.get("audio"):
        return surveys.store_comment(m)        # not a survey answer, or the file never arrived
    r = await call("GET", m["audio"]["url"])   # .m4a (AAC), same bearer token
    r.raise_for_status()
    s.voice_text, s.voice_ms = transcribe(r.content), m["audio"]["duration_ms"]
    s.save()
    await call("POST", "/api/imessage/reactions", {"message_id": m["id"], "reaction": "love"})
    await call("POST", "/api/imessage/send", {"to": m["handle"],
        "text": "Thank you, that's going straight to the team."})

@app.post("/webhook")
async def webhook(req: Request, tasks: BackgroundTasks):
    raw = await req.body()
    if not verify(req.headers.get("X-ZeroSettle-Signature", ""), raw):
        return Response(status_code=401)
    ev = json.loads(raw); m = ev.get("message")
    if m is None or m["direction"] != "in" or (m["id"], ev["event"]) in seen:
        return {"ok": True}                    # typing events, our own sends, retries
    seen.add((m["id"], ev["event"]))
    if ev["event"] == "reaction.received":
        tasks.add_task(on_tapback, m)
    elif m.get("kind") == "audio":
        tasks.add_task(on_voice, m)
    else:
        tasks.add_task(surveys.store_comment, m)   # typed answers: a person reads them
    return {"ok": True}

surveys and transcribe stand in for your database and speech-to-text provider; only a voice note answering your follow-up gets the thank-you. Register the receiver with PUT /api/imessage/webhook and keep the signing secret, which is shown once; how ZeroSettle webhooks are signed, retried and deduplicated covers the rest.

Limits and consent for feedback surveys

A survey sent into an existing thread uses none of ZeroSettle's new-contact allowance; only research participants who have never texted your number or heard from it count. Consent is the tighter limit.

Source: ZeroSettle iMessage API docs, “Errors, limits, and compliance”. The day runs 3 AM to 3 AM Eastern. A survey sent within 5 minutes of your closing reply counts as one message with it.

Inviting 60 opted-in users who have never texted the line takes twelve days on one Pro number (5 a day). On Growth each number takes 50 a day at 3 per 15 minutes, so an account with three numbers can invite all 60 the same day. A handful of new participants a day is a reasonable research pace anyway: each one is a recording someone has to hear.

When ZeroSettle is not a fit for feedback surveys

ZeroSettle is the wrong tool when a survey needs more than one tap and one recording, or when the sample matters as much as the answers.

• You need multi-question surveys with branching. Send a survey-tool link after the customer replies, when links in the thread are tappable.
• You need a representative sample that includes Android users. ZeroSettle reaches iMessage handles only, with no SMS fallback, so results lean toward Apple users; non-Apple phones get SMS or RCS, not iMessage.
• You would survey people with no existing relationship, such as a bought panel. That is unsolicited messaging, and ZeroSettle suspends lines used for it.
• You need anonymous feedback. Every answer arrives with the customer's handle.

How much does iMessage for feedback surveys cost?

Feedback surveys cost the plan price. ZeroSettle charges no per-message fees, and messages in both directions are unlimited (counted).

Both plans carry a 30-day money-back guarantee on the subscription, and this flow uses no AI credits. See the pricing page, read the ZeroSettle API docs, or get a number.