Reprompt — What It Is, When a Bot Sends One, and How Many Before It Gives Up

Quick answer: A reprompt is the bot asking again. The bot asked a question, waited, and got either nothing or something it could not use, so it asks a second time, ideally in fewer words and with a hint about what a usable answer looks like. The term is native to voice: Amazon's Alexa documentation defines the reprompt field as the speech "to use if a re-prompt is necessary," played when the skill keeps the session open "but the user doesn't respond with anything that maps to an intent." Chat builders reuse the idea under other names. Google's Dialogflow CX calls the handlers that fire when a form field is not filled "reprompt handlers"; none of the Voiceflow documentation pages we searched (73 on a 10 September recount) uses the word at all, and they call the same thing a "follow-up message" or "reminder"; Landbot puts it in a field labeled Validation error message. In every case the mechanism is the same: a counter, a message per attempt, and a limit after which the bot stops asking.

The limit is the part worth knowing. Alexa's documented sequence, read 10 September 2026, has one reprompt; after it, if the user "still doesn't respond, the session ends." Dialogflow CX lets you write up to six numbered reprompts for each of two events, sys.no-input-1 through sys.no-input-6 and sys.no-match-1 through sys.no-match-6, with a default handler for whatever comes after, which is our reading of the docs (the Sources note says what Google's pages do and do not state). Voiceflow's global no-reply setting, which is "Off by default," waits once enabled "how many seconds of silence before the agent responds (default 10)" and repeats "how many times the agent will follow up before ending the session (default 5)." Rasa's two-stage fallback asks up to three times, confirm, rephrase, confirm the rephrase, before an "ultimate fallback action" that is usually a human handoff. None of the fifteen platforms we review uses the word reprompt in a way our reviews recorded; the tally is in Sources.

Two triggers, one mechanism

A reprompt fires on one of two events, and the vendors that separate them give each its own counter.

No input. The bot asked and heard nothing. In voice this is silence: Alexa's session page says that after the prompt "Alexa opens the microphone to indicate that the skill expects the user to respond," and "if a few seconds elapse without a response from the user, Alexa closes the microphone. If the skill specified a reprompt, Alexa reprompts the user to speak and opens the microphone for a few more seconds." Dialogflow CX's sys.no-input-default event is defined as "End-user input was not received," and it lists three ways that happens: empty text input, "empty end-user audio input or the input does not contain any recognized speech," and "a no speech timeout occurs before end-user audio input contains any recognized speech." So no input covers the speech-to-text layer hearing noise as well as hearing nothing.

No match. The bot asked, the user answered, and the answer did not fit. Dialogflow CX defines sys.no-match-default two ways depending on where it is attached: at flow or page level, "end-user input does not match any intents for handlers that are in scope"; at parameter level, "end-user input does not satisfy the form parameter." The second is the reprompt case proper, a question that was answered with the wrong kind of thing (a date field given a shrug), and Dialogflow CX names the handlers attached there: parameter-level event handlers "are also known as reprompt handlers" and "are specifically intended to handle invalid end-user input during form filling." The first is closer to a fallback intent. Alexa folds both into one session-ending reason, EXCEEDED_MAX_REPROMPTS, defined as "The user either did not respond or responded with an utterance that did not match any of the intents defined in your voice interface."

Chat has a third case voice does not: the message that arrives hours later. A WhatsApp or Instagram customer who does not answer is not silent in the voice sense, they have put the phone down, and a bot that "reprompts" after ten seconds on a messaging channel is nagging. Voiceflow's docs separate the two: the no-reply follow-up counts in seconds, while a chat session that has gone quiet is closed by a separate inactivity timer, "1-2,880 min (2 days), default 15 minutes." Intercom's Fin can send a follow-up when a chat goes quiet, and the wait is not yours to set: "The inactivity timer is set to 4 minutes over chat channels and isn't configurable," with a "No follow up" option if you would rather it sent nothing.

Reprompt, fallback, disambiguation

The three words get used for each other, and the difference decides which message you write.

Rasa's documentation shows all three in one flow. Its two-stage fallback fires on low intent confidence: first "the user is asked to confirm the intent" (a disambiguation question, generated in Rasa's source as "Did you mean '{intent}'?" with Yes and No buttons), then if they deny it "the user is asked to rephrase their message" (a reprompt, the documentation's example utter_ask_rephrase reading "I'm sorry, I didn't quite understand that. Could you rephrase?"), and if the rephrased message is confirmed as wrong too, "an ultimate fallback action is triggered (e.g. a handoff to a human)." Rasa's newer CALM patterns keep the split: a pattern_clarification that lists options ("I can help, but I need more information. Which of these would you like to do: {{context.clarification_options}}?"), a pattern_cannot_handle that plays utter_ask_rephrase ("I'm sorry I am unable to understand you, could you please rephrase?") or, for a request outside scope, utter_cannot_handle ("I'm sorry, I'm not trained to help with that."), and a pattern_human_handoff.

The practical test: if the bot is inside a question it asked, the message is a reprompt and should repeat the ask. If the bot is between questions, it is a fallback and should not pretend a question was pending.

How many times

Every platform that names the mechanism also caps it, and the cap is the design decision.

Two things stand out. Alexa, the platform that coined the word, documents exactly one reprompt, which is why the voice-design habit is to make it count: Amazon's own tip is "Include a specific reprompt that lets the user know what to say next." And Dialogflow CX's six numbered slots are not a recommendation to ask six times; they exist so that each attempt can say something different. Our editorial rule, which is reasoning rather than a measurement, is two asks at most, the second with a change of mechanism (buttons instead of free text), and then a handoff rather than a third ask; the fallback-rate guide states the same ladder in two steps, options on the first miss and a person on the second, and the copywriting guide sets out what each of those messages says.

What a reprompt has to say

A reprompt is not the prompt again. The user already heard or read the prompt and did not answer it, so repeating it verbatim tells them nothing new. Three properties follow from the vendor guidance and limits read for this entry.

Shorter than the prompt. In voice, every second of the reprompt is a second the microphone is closed; Alexa caps the audio files in a reprompt at 90 seconds against 240 for the main speech, and the practical length is one sentence. In chat the same logic applies to screen space on a phone.

Carries the expected answer. Amazon's tip above; Rasa's clarification pattern lists the options; Dialogflow CX's parameter-level handler knows which field is open and can name it ("What date? For example, 14 October"). A reprompt that says "Sorry, I didn't get that" without saying what would be gotten has wasted the attempt.

Changes between attempts. The reason numbered handlers exist. The first reprompt narrows the ask; the second changes the mechanism, typically from typed or spoken input to quick replies or buttons where the channel has them, with a person among the options; the third is not a reprompt but a handoff. Repeating one line is the pattern the fallback intent entry warns about, and Intercom's escalation guidance treats the customer's side of it as a trigger: "the customer repeats themselves across three turns" is one of the events on which Fin offers a human.

How platforms expose it

The voice-bot buyer's guide covers the voice platforms in more depth; for a chat builder, the question to ask a vendor is whether a question block has a per-attempt message and a retry limit, or whether you will be counting attempts with a variable, which is what our chatbot scripts guide assumes for the flow builders we review.

What our fifteen reviews record about reprompts

Nothing under that name. Searched 10 September 2026, case-insensitively, across the fifteen files matched by sample-reviews/*-review.md: the string reprompt|re-prompt appears in 0; no reply|no-reply|no input|no-input in 0; fallback in 9; didn't understand|don't understand|rephrase in 1, the Landbot review, which captured the Validation error message field on an Ask for an email block with its default retry copy. The Chatfuel review captured a Fallback message field on the AI behavior form (a fallback, not a reprompt, by the distinction above). The reviews time a ten-question FAQ bot and probe handoff, and none of the fifteen protocols asked a question, stayed silent and counted the asks; that is a gap in the protocol, recorded here, not a finding that the platforms lack the feature. Corrections to editorial@chatbotscape.com.

Related terms

• Fallback intent — the catch-all for a message that matched nothing, as distinct from an unanswered question.
• Slot filling — the form-filling loop in which most chat reprompts occur, with Lex's and Dialogflow's retry settings.
• Chatbot fallback rate — the metric that counts how often the catch-all fires.
• Human handoff — what the last reprompt should lead to.
• Turn-taking — the microphone-open window in which a voice reprompt fires.
• Voice bot — the channel where reprompt is a first-class field.

FAQ

What does reprompt mean in a chatbot?

The bot asking its question a second time because the first ask got no usable answer: silence, in a voice bot, or a reply that did not fit the question, in either. The word is Amazon's; Alexa's response format has a reprompt field that plays after "a few seconds" of silence, once, before the session ends. Chat builders implement the same thing under names like reprompt handler (Dialogflow CX), follow-up or reminder (Voiceflow) and validation error message (Landbot).

What is the difference between a reprompt and a fallback?

A reprompt happens inside a question the bot asked and repeats that ask. A fallback happens when the user sends something the bot cannot place and no question is pending; it recovers by stating what the bot can do and offering an exit. Dialogflow CX uses the same no-match event for both, but attaches reprompt handlers to form parameters and page- and flow-level handlers to the catch-all case.

How many times should a chatbot reprompt?

Alexa allows one. Dialogflow CX lets you write six per event, which is a ceiling, not advice. Our editorial rule is two, each different from the last and the second switching to buttons instead of free text, followed by a handoff to a person rather than a third ask. Rasa's two-stage fallback is the same shape with one more ask: confirm, rephrase, confirm the rephrase, then hand off.

Why does Alexa end my session with EXCEEDED_MAX_REPROMPTS?

Because the user either stayed silent through the prompt and the reprompt, or answered with something that matched none of the skill's intents. Amazon's request-types page defines the reason exactly that way. The fix is usually the interaction model (the utterances users actually say are missing) or a reprompt that does not tell the user what to say.

Does a reprompt apply on WhatsApp or Instagram?

Only loosely. Messaging is asynchronous, so a customer who has not answered in ten seconds has not failed to answer; they have gone. A reprompt on those channels is a follow-up message sent later, subject to the channel's messaging-window rules, and a bot that repeats its question within seconds reads as nagging. The no-match reprompt (an answer that did not parse) applies as it does anywhere.

Sources

• Amazon Alexa Skills Kit, Request and Response JSON Reference (developer.amazon.com/en-US/docs/alexa/custom-skills/request-and-response-json-reference.html), read 9 September 2026: the reprompt property, "OutputSpeech to use if a re-prompt is necessary. Used if your service keeps the session open after sending the response (shouldEndSession is false), but the user doesn't respond with anything that maps to an intent defined in your voice interface while the microphone is open. The user has a few seconds to respond to the reprompt before Alexa closes the session. If the reprompt doesn't have a value, Alexa doesn't reprompt the user."; the shouldEndSession description ("Alexa opens the microphone for a few seconds to listen for the user's response. Include a reprompt to give the user a second chance to respond"); the size limits ("The combined total time for all audio files in the reprompt property of the response can't be more than 90 seconds," 240 seconds for outputSpeech, "The outputSpeech response can't exceed 8000 characters").
• Amazon Alexa Skills Kit, Manage Skill Sessions and Session Attributes (developer.amazon.com/en-US/docs/alexa/custom-skills/manage-skill-session-and-session-attributes.html, last updated 20 March 2026), read 10 September 2026: "the session stays open, and Alexa opens the microphone to indicate that the skill expects the user to respond"; "if a few seconds elapse without a response from the user, Alexa closes the microphone. If the skill specified a reprompt, Alexa reprompts the user to speak and opens the microphone for a few more seconds. If the user still doesn't respond, the session ends"; "Tip: Include a specific reprompt that lets the user know what to say next"; the screen-device extension ("between 20 and 30 seconds" after a failed reprompt).
• Amazon Alexa Skills Kit, Request Types Reference (developer.amazon.com/en-US/docs/alexa/custom-skills/request-types-reference.html, last updated 28 November 2023), read 10 September 2026: SessionEndedRequest reasons, EXCEEDED_MAX_REPROMPTS: "The user either did not respond or responded with an utterance that did not match any of the intents defined in your voice interface"; "Your skill cannot return a response to SessionEndedRequest."
• Google Cloud, State handlers, Dialogflow CX documentation (docs.cloud.google.com/dialogflow/cx/docs/concept/handler; stamped "Last updated 2026-08-26 UTC" in the copy fetched 9 September 2026 and "Last updated 2026-09-03 UTC" when re-read in a browser on 10 September; the quoted text was identical in both), read 9 and 10 September 2026: the built-in events table (sys.no-match-default: "For flow or page level: end-user input does not match any intents for handlers that are in scope. For parameter level: end-user input does not satisfy the form parameter"; sys.no-match-[1-6]: "If you provide handlers for any of these numerically ordered events, they are invoked instead of sys.no-match-default and in order"; sys.no-input-default: "End-user input was not received" with the three triggers; sys.no-input-[1-6]; sys.long-utterance: "End-user input exceeds the maximum allowed length (256 characters) ... If not provided, Dialogflow CX treats long user utterance as no-match"); "Every flow has event handlers for the no-match and no-input built-in events. These event handlers are automatically created when you create a flow, and they cannot be deleted"; "Parameter-level event handlers are event handlers that are applied to a form parameter. They are also known as reprompt handlers. These event handlers do not allow custom events, as they are specifically intended to handle invalid end-user input during form filling"; the evaluation order (parameter-level "reprompt handlers" first, then page, then flow). The handlers page itself does not say what happens after the sixth numbered handler; Google's Parameters page, linked from it (form-filling reprompt handlers section, read 10 September 2026), says reprompt handlers can be used "to transition to another page after N failed attempts" and that "If no reprompt handlers are defined, the initial prompt will be used to reprompt the end-user as needed." The reading that the default handler covers whatever follows the numbered ones is ours, from the "invoked instead of ... and in order" wording.
• Voiceflow documentation, Behaviour, Buttons, Card and Listen (voiceflow.com/docs/documentation/build/behaviour and the steps pages under the same path; the HTML rendering's JSON-LD dateModified read 31 July 2026 on each when fetched 9 September, and the Markdown rendering carries no date), read 9 September 2026: Global no-match, "How the agent responds when it can't match any skill, user intention, or knowledge base result," with Generative and Scripted options; Global no-reply, "How the agent responds when the user says nothing. Off by default," "Inactivity time - how many seconds of silence before the agent responds (default 10)," "Max no-reply messages per turn - how many times the agent will follow up before ending the session (default 5). For example, with 10 seconds and 5 attempts, the session ends after 50 seconds of total silence"; "Time in minutes before an inactive chat session ends (1-2,880 min (2 days), default 15 minutes)"; the Buttons step settings ("No match: When enabled, handles situations where the user's response doesn't match any button label. Your agent can send a message or follow a path"; "No reply: When enabled, handles situations where the user doesn't respond within a specified time. Your agent can either send messages periodically or follow a path after the specified time"); the Listen step ("Configure how long to wait and what message to send as a reminder"). A case-insensitive search of every documentation page listed in the site's llms.txt index (73 on a recount of 10 September) for "reprompt" returned no match on 9 September 2026.
• Rasa Open Source documentation, Fallback and Human Handoff (legacy-docs-oss.rasa.com/docs/rasa/fallback-handoff, last updated 25 April 2025), read 10 September 2026: the Two-Stage Fallback sequence ("A user message is classified with low confidence – The user is asked to confirm the intent"; "If they deny, the user is asked to rephrase their message"; "If they deny, an ultimate fallback action is triggered (e.g. a handoff to a human)"); utter_ask_rephrase "I'm sorry, I didn't quite understand that. Could you rephrase?"; utter_default "Sorry I didn't get that. Can you rephrase?"; action_default_fallback "will send the response utter_default and revert back to the state of the conversation before the user message that caused the fallback." The affirmation wording "Did you mean '{intent}'?" is not printed on the docs page; it is read from rasa/core/actions/action.py on the project's main branch the same day. Rasa Pro documentation, Patterns (rasa.com/docs/reference/primitives/patterns, last updated 9 September 2026), read 10 September 2026: pattern_cannot_handle, pattern_clarification, pattern_human_handoff; utter_cannot_handle "I'm sorry, I'm not trained to help with that."; utter_ask_rephrase "I'm sorry I am unable to understand you, could you please rephrase?"; utter_clarification_options_rasa; "By default, clarification will continue to check on the users intentions regardless of the number of times it has asked. This can be avoided by counting the number of clarification requests and linking to the human handoff pattern if this count reaches some threshold"; the example flow's slots.clarification_count > CLARIFICATION_LIMIT branch linking to pattern_human_handoff.
• Intercom Help, Manage Fin AI Agent's escalation guidance and rules (article 12396892), read 10 September 2026: "Loop: The customer repeats themselves across three turns." Intercom Help, Use Fin AI Agent in Workflows (article 10032299), read 10 September 2026: "The inactivity timer is set to 4 minutes over chat channels and isn't configurable"; the follow-up options, including "No follow up: Fin doesn't send any additional messages after responding." Intercom Community, Can I customize the customer-facing message Fin sends when an escalation rule happens? (8 January 2026), read 10 September 2026, quoting Fin's FAQ (fin.ai help article 10763810, not fetched directly): "While this default message itself cannot be directly edited or removed, you can influence this process by creating Guidance in Fin."
• Manychat Help, Default Reply in Manychat (updated 3 December 2025), read 9 September 2026: the trigger fires "every time a user sends a message" or "once every 24 hours for the same user."
• Chatbotscape review corpus (the fifteen platform reviews listed at /reviews), searched 10 September 2026 from the repository root. Denominator: ls sample-reviews/*-review.md | wc -l returns 15. grep -liE 'reprompt|re-prompt' sample-reviews/*-review.md returns 0; grep -liE 'no[- ]reply|no[- ]input' sample-reviews/*-review.md returns 0; grep -liE '\bfallback' sample-reviews/*-review.md returns 9; grep -liE "didn.t understand|don.t understand|rephrase" sample-reviews/*-review.md returns 1 (landbot). Passages cited: landbot-review.md lines 708 and 714 (the Validation error message field and its default copy); chatfuel-review.md line 577 (the Fallback message field); manychat-review.md line 672 (the Default Reply hook on the Telegram channel page).
• Ahrefs Keywords Explorer, US overview, queried 9 September 2026 — the demand, difficulty, CPC, global-volume and parent-topic figures in this entry's keyword note.
• Chatbotscape evaluation methodology. /methodology (continuously updated).