People Ignore the Buttons in Your DM Flow? How to Design for Free-Text Replies

Buttons are a convenience you offer, not a contract the other person signed. Build the flow so typing is a valid answer everywhere.

By Anthony Casauria · · 8 min read

Share
Man with a ginger beard smirks while typing on his phone at a cafe table against a pink plaster wall.

Key takeaways

  • People type instead of tapping because typing is the normal way to use a messaging app, so design each DM flow step so a typed answer has somewhere to go.
  • Instagram allows up to 3 buttons or 11 quick replies per message, never both in one step, with button labels capped at 20 characters.
  • Give every question step an explicit exit for replies that don’t fit, and wire the Skipped path somewhere useful instead of letting it end the conversation.
  • Layer keyword triggers under the flow as a net, then intent matching for reworded questions, because with nothing matching and no agent enabled, no automated reply is sent.
In this article
  1. Why people type instead of tapping
  2. What Instagram's message format actually allows
  3. How to design for free-text replies at every step
  4. Designing DM flows that expect free-text replies

A DM is a conversation, and nobody has ever been in a conversation where the other party refuses to acknowledge anything except three approved phrases. That mismatch sits behind every "why do people ignore my buttons" question. Your flow treats free-text replies as an error state. The person treats typing as the normal way to use a messaging app, because it is. Better button labels help a little. They don't fix it. What fixes it is designing each step so a typed answer has somewhere to go, and that takes four layers sitting under each other: fewer and clearer options, a keyword net, intent matching for the questions people word ten different ways, and something that can take the turn nobody anticipated. What follows is how each layer works, what Instagram's message format actually permits, and the order to build them in.

Short answer: design for free-text replies rather than taps. Keep options to three or fewer with concrete labels, give every question an explicit exit for replies that don't fit, put keyword triggers underneath the flow as a net, and add intent matching or an AI step for the questions people phrase differently every time.

Why people type instead of tapping

Tapping a button only wins if the button says the thing the person was already thinking. Most buttons don't.

Three patterns cover almost all of it. The question is more specific than any option on offer, so "Pricing" does nothing for "do you ship to Ireland and how much is express". Or the person is on an older thread where your buttons have scrolled out of view, or they replied straight from a notification where the card never rendered. Or they don't read your message as an interface at all, because it landed in the same inbox as messages from their friends, where typing is how everything works.

None of that is user error. The last one deserves a second look: the more natural your copy sounds, the more likely somebody answers it in prose. Writing well increases typed replies rather than reducing them. That's the trade you're making, and it's worth making.

What Instagram's message format actually allows

Know the constraints before you redesign anything, because some "nobody tapped" problems are really "nothing was sent" problems.

Instagram permits up to 3 buttons or 11 quick replies on a single message, and the two can't be mixed in one step. Button labels are capped at 20 characters, which is tighter than it sounds once you try to write something unambiguous. A step carrying both an image and buttons sends only the buttons, which is why HyperDM's builder refuses to publish that combination and asks you to move the image into its own step above. These are platform-level message constraints, not tool preferences, and the wider set of what Instagram's messaging surface supports is documented in Meta's Instagram Platform overview. The practical upshot: if you wrote five options, two of them were never delivered, and the people you think ignored your buttons may have been answering a message that looked half-finished. HyperDM's flow builder enforces these limits at design time instead of failing at send time.

How to design for free-text replies at every step

Six steps, in this order. The early ones reduce typed replies. The later ones catch the ones you can't prevent.

Step 1: Cut to three options and make the labels concrete

Fewer options get tapped more. That sounds backwards until you watch somebody read four near-identical choices and give up. Three is both Instagram's button ceiling and roughly the limit of what anybody will evaluate in a DM. Write labels as the answer the person would give, not as a category: "See the price" beats "Pricing", and "I need a size" beats "Product info". The 20-character cap forces that discipline anyway. Avoid labels that overlap, because two options that could both be right is the most reliable way to produce a typed reply asking which one you meant. And never make a button the only way forward on a step that asks a real question.

Step 2: Use quick replies when you genuinely need more choices

Quick replies are the right tool for a list, since Instagram allows up to 11 of them against three buttons. They behave differently in the thread, too. They sit as suggestions above the keyboard rather than as a card in the conversation, so the person is already looking at their keyboard when they choose and typing stays one tap away. That makes them the better pattern for anything resembling a menu. A step can't carry both quick replies and buttons, so pick one per message. Keep the question above them short. A long question with a wide set of options reads as a form, and forms get abandoned in DMs.

Step 3: Give every question an explicit exit for answers that don't fit

This is the step that matters most and the one almost nobody configures. A question step waits for a typed reply and can end three ways: the person answers in the format you asked for, they reply with your skip word, or their reply still doesn't fit after the re-asks you allowed. With HyperDM's default retry limit of 1, that third case fires on their second unusable reply. There's also a no-response timeout that simply ends the flow. Turn on a separate exit for skips so the question has distinct Answered and Skipped paths on the canvas, then wire Skipped somewhere useful rather than letting it end the conversation. One caution: keep opt-in, sequence-subscription and webhook steps on paths that only Answered reaches, or people who never answered will trip them.

Step 4: Put keyword triggers underneath the flow as a net

Keyword automations catch typed replies more reliably than anything else because they're literal: a keyword anywhere in the message matches, no scoring, no model involved. That makes them right for the handful of things people will definitely type no matter what your flow asked. Build separate keyword automations for "price", "ship", "size", "refund" and whatever else dominates your inbox, and they'll answer the people who went off-script mid-flow. The trade-off is coverage. A keyword can't anticipate phrasing it has never seen. Treat this layer as insurance on your top five questions, not as a system for understanding language.

Step 5: Add intent matching for questions worded ten different ways

Intent recognition compares an incoming message against a description you write plus example messages you review, and sends your saved reply when the wording is close enough. It handles the long tail that keywords miss. Be realistic about the ceiling: HyperDM labels the builder experimental, and its own documented test is instructive. "How much does the course cost?" matched a description written as "how much is the course?", while "whats the price of the course" and "how much is it?" did not. Very short questions and shorthand usually miss. Only the DM keyword trigger offers it, since comment and story triggers match on keywords alone, and it runs after your keyword automations and before the AI agent. Publishing requires a should-match and a should-not-match test to pass, which is a sensible gate to copy by hand even in tools that don't enforce it.

Step 6: Let an AI step take the turn nobody could script

The layers above handle the variety you anticipated. Real free-text replies include questions you'd never have written a branch for, and the honest options there are escalating to a human or letting a model answer that one turn. In HyperDM an AI step is a node on the canvas, available on Growth and above, so a deterministic flow can hand over a single unpredictable turn and then resume on whatever the model concluded. Worth knowing what the alternative looks like: when nothing in the flow matches and no agent is enabled, no automated reply is sent at all, which from the other side is indistinguishable from being ignored. Decide which of those two outcomes you want before launch. Our comparisons of AI agents versus keyword bots and an AI DM agent versus a flow builder go deeper on where each belongs.

Designing DM flows that expect free-text replies

Treat free-text replies as the default case and buttons as the shortcut, not the other way round. That reframing changes what you build: three concrete options instead of five vague ones, quick replies where a real list is needed, a wired path out of every question for answers that don't fit, literal keyword automations under the whole flow for your top questions, intent matching for the long tail, and one escalation route for the turn nobody could have predicted. Test it by typing the wrong thing on purpose from a second account. A flow that handles deliberate nonsense gracefully will handle real customers fine. The advanced flow nodes start on Growth, and HyperDM's plans list what each tier includes if you want to cost out the layers before building them.

Frequently asked questions

Three at most, which is also Instagram's limit per message. Use quick replies when you need more, since Instagram allows up to 11 of those. More options reduce tapping rather than increasing it, because people stop evaluating a list in a chat window and either type or leave.
It depends entirely on your configuration. A question step will re-ask up to its retry limit and then take its unfit path, which ends the flow if you left that path empty. Outside a question, the message falls through to keyword automations, then intent matching, then an AI agent if one is enabled. With none of those, nothing is sent.
Not yet, and tools that claim otherwise are overselling. Close rewordings still miss, and short messages like "how much?" are the hardest cases. Use intent matching to widen coverage on questions that matter, and keep a literal keyword trigger behind anything that must always get a reply.
They do different jobs. Buttons suit two or three decisive actions and can carry links or a purchase. Quick replies suit menus of up to eleven options and keep the keyboard in view, which makes typing feel permitted rather than wrong. A step can carry one kind or the other, never both.
No. Buttons reduce effort for the majority who do want the obvious path, and a link button is still the cleanest way to send someone to a checkout. The goal is making typing a valid alternative at every step, not banning the shortcut that works for most people.
HyperDM

Run the automations. Add AI when you’re ready.

Comment-to-DM, keyword replies and follow gates — set them up, test them, then add optional AI answers from your real catalog. Free on one DM channel, 50 conversations a month, no card.

More stories worth reading