Channels

A Fly survey is written once and can run on more than one messaging app. The questions, the logic, the timeouts, the payments and the exported data are the same whichever app a respondent is using — but the apps themselves are not the same, and a handful of differences will stop a survey that works on one from working on the other.

This page is the list of those differences. Read it before you write a survey that will run on WhatsApp.

The channels

Instagram appears in a few places in this documentation and in the dashboard. There is no way to run a survey on it today: nothing receives Instagram messages. Treat it as not supported.

A survey does not choose its channel. A form has a shortcode, and the same shortcode can be started on either channel. What decides the channel is how the respondent arrived — which page or which WhatsApp number they were sent to. One form can therefore serve a Messenger arm and a WhatsApp arm of the same study at once, and the channel each respondent used is recorded with their answers.

The differences that break surveys

These are the ones worth checking before you launch. Everything in this section is a send failure or a hard limit, not a matter of taste.

Number of choices

An 11-point Opinion Scale (0–10) is eleven options and will fail on WhatsApp. Use 0–9 or 1–10.

If your survey may ever run on WhatsApp, keep every Multiple Choice, Dropdown, Picture Choice, Rating and Opinion Scale question to 10 options or fewer.

Button labels on links and videos

The buttonText of a tracked link, video or plain link is capped at 20 characters on WhatsApp, and a longer one does not get truncated — the whole message is rejected and the respondent's survey stops with an error.

Messenger has no such cap. A buttonText that has worked on Messenger for years can stop a WhatsApp survey the first time it is sent.

Phone, email and SMS links

tel:, mailto: and sms: destinations work on WhatsApp only through a tracked link ("type": "link_tracking"), which turns them into an ordinary web link that redirects. A hand-written webview pointing straight at tel: is an error on WhatsApp.

On Messenger both forms work. Use link_tracking and the question works on both.

Captions on attachments

On WhatsApp, an attachment question's Typeform title is sent as the caption and appears with the image, video or document. On Messenger there is no caption and the title is not shown at all — if the text matters, send it as a separate Statement before the attachment.

For documents, the filename is what a WhatsApp respondent sees. Name the file properly before uploading it to the Media library.

attachment_id — the old way of referencing a file pre-uploaded to a Facebook page — does nothing on WhatsApp and will fail to send. Use a Media Library URL instead.

Messaging after 24 hours

Both platforms restrict what you may send to someone who has not written to you recently, and both solve it the same way: a pre-approved template, created in the dashboard's Message Templates tab and used with a utility_message question.

The window is 24 hours on both. What differs is how the template gets approved, and what a mistake looks like:

So any wait longer than 24 hours, any follow-up nudge and any delayed payment result needs an approved template on either channel. A survey whose long waits are followed by ordinary statements stops there for every respondent — noisily on Messenger, silently on WhatsApp.

Features that exist on one channel only

Receiving files from respondents

Not supported on either channel. See Upload.

The differences that do not matter

Worth stating so you do not go looking for them:

• Question types. Every question type other than handoff behaves the same way. You do not write a WhatsApp version of a question.
• Media. One Media Library URL works on every channel and every connected account. There is no per-channel upload.
• Hidden fields, logic jumps, seeds and validation. Identical.
• Timeouts, bails and payments. Identical, subject to the 24-hour rule above.
• Exports. One export contains every respondent, whichever channel they came through, with platform available as a column.
• Custom messages. The Typeform messages listed under Messages are used on both.

How respondents start a survey

This is the biggest practical difference between the two, because on WhatsApp there is no link you can simply hand out that carries a shortcode invisibly.

The consequence on WhatsApp is that the entry reference must be set as the ad's autofill message, and if it is not, every person who clicks the ad starts the default form instead of your survey — with no error anywhere. This is the single easiest and most expensive way to misconfigure a WhatsApp study.

The full account is on WhatsApp Surveys.

One ad, either channel

Virtual Lab can create a multi-destination ad that opens either Messenger or WhatsApp, with Meta choosing per respondent. One ad, one survey shortcode, and respondents arriving on both channels.

Two things follow, and both matter for survey design:

1. Your survey must work on both channels — every limit on this page applies, because you do not know which channel any given respondent will get.
2. You cannot use it to compare channels. Meta assigns the channel by predicted responsiveness, not at random, so the two groups are not comparable. To compare channels, run two separate destinations in a destination experiment. See Destinations.

A pre-launch checklist for a WhatsApp survey

1. No question offers more than 10 choices.
2. Every buttonText is 20 characters or fewer.
3. Every tel:, mailto: or sms: destination uses link_tracking, not webview.
4. Every wait longer than 24 hours is followed by a utility_message with an approved template on the number's account.
5. No handoff questions on the path a WhatsApp respondent takes, and no notify questions anywhere — it is retired on Messenger too.
6. Attachment titles read as captions, and document filenames read as filenames.
7. The ad's autofill message is the entry reference, verbatim.
8. You have started the survey yourself, from a real phone, before spending anything. See Testing.