Telegram Poll Answers: Updating CRM Choices and Retractions
A custom-integration policy for separating poll totals from personal answers and handling changed choices without creating false enquiries.
To store a person's Telegram poll answer in a CRM, use a non-anonymous poll sent by the bot itself and process poll_answer updates. A poll update describes the poll's state; it does not by itself connect a person to a selected option. Before writing anything, check the poll's origin, the voter identity and the authorized link to an enquiry. A changed selection should replace the previous answer. An empty selection represents a retracted vote.
This is a proposed policy for a custom Telegram Bot API integration. It is intended for a service owner preparing an implementation with a developer. You need access to application code, durable event storage and a test environment. It is not a verified built-in 24go.site feature or a no-code configuration guide. Documentation was checked on November 12, 2026. The acceptance procedure below is a test plan for your environment, not a report of a live bot or CRM experiment.
Decide whether you need an identifiable answer
Aggregate results are sufficient for many surveys. If a team asks which meeting time suits a group, it may only need an option and a vote count. Writing personal answers to a CRM needs a defined purpose, an established link to the relevant enquiry and an appropriate basis for processing the data. Do not try to identify anonymous voters by correlating voting times with nearby messages.
Before running an identifiable survey, explain who can see the answers and why they will enter your enquiry-handling system. Do not describe that collection as anonymous. A Telegram identifier is not a legal basis for processing, permission to send marketing messages or proof that someone can act for a company. Selecting an interest does not constitute an order, a mailing subscription or authority to change a deal.
For example, a person with an existing enquiry may choose which topic they want to discuss. Store that selection as an answer to that particular question. Creating a manager task, moving a sales stage or sending another message should each have separate eligibility rules. A corrected answer can then be recorded without creating another lead or triggering an unwanted repeat conversation.
Distinguish poll totals from a person's selection
The official Update documentation defines separate poll and poll_answer fields. New votes through poll_answer are limited to non-anonymous polls sent by the bot itself. A poll created manually by an employee in a group must not automatically be treated as a source of personal answers for your bot.
A PollAnswer includes poll_id and the selected options. Its user field is optional: it identifies the user who changed an answer when the voter is not anonymous. The object may instead contain voter_chat, identifying the chat on whose behalf the answer changed. A chat identity must not silently become a person's contact record. Preserve the voter type and route unsupported types for review.
| Received information | Safe application decision |
|---|---|
An aggregate poll update only |
Update poll statistics; do not assign votes to contacts |
A registered poll_id, a user and an authorized enquiry link |
Validate the selected options and replace this participant's answer |
A voter_chat without a personal user |
Do not write a guessed answer to a person's CRM record |
| A poll missing from your registry | Hold the event and investigate poll ownership and registration |
| No selected options | Record a retraction after validating the poll and participant |
When events do not arrive at all, first check update subscriptions and routing. An allowed_updates fix addresses delivery configuration. It cannot turn an anonymous survey into an identifiable one or reconstruct a missing poll registry.
Register the question before applying answers
Maintain an internal poll registry. After a successful sendPoll response, store the returned Message and its poll.id. The successful response provides the data for the sent poll. Your registry should include the bot or connection, tenant, poll_id, question purpose, schema version, authorized enquiry association and option map. Question text is not a suitable unique key.
You can create an internal sending intent before the request. Until a successful result is available, that intent is not a confirmed Telegram poll. If the sending outcome is unknown, do not repeatedly create another poll as an automatic recovery loop. Investigate the ambiguous outcome first. An answer that arrives before registration completes can be persisted in a held queue and processed once the verified mapping exists.
The sendPoll documentation makes anonymity enabled by default. Explicitly set is_anonymous: false for an identifiable workflow. Also decide whether the poll is regular or a quiz, whether multiple selections are allowed, whether participants may change their selection, and whether options can be added or shuffled. Adding options is not supported for anonymous polls or quizzes. Regular polls allow revoting by default; quizzes have a different default. Do not inherit these decisions accidentally from an old library example.
Store the actual configuration, not just what the developer intended. Even if you intended to prohibit added options, a missing parameter check does not make unknown option mappings safe. A first implementation can deliberately fix the option set and predefine every CRM meaning. That is an application policy chosen to reduce complexity, not a universal limitation of Telegram polls.
Keep option identity separate from its position
option_ids contains zero-based option numbers. Those numbers belong to a particular poll; they are not universal CRM values. Option zero might mean integration in one question and reporting in another. A rule such as "zero means strong interest" is unsafe without the question's registered mapping, even when the options never change.
The current API also documents option_persistent_ids on an answer and persistent_id on a PollOption. Persistent identifiers survive option additions and deletions. When your workflow permits changes to the option set, bind business meanings to persistent identifiers and a mapping version, rather than a current position or displayed label. Check that your installed library exposes these fields instead of discarding them during parsing.
A legacy integration that only supports positional numbers can operate under a narrower contract: a fixed option set, the saved original map and rejection of unknown changes. Do not describe that as support for editable poll options. If the map is lost or an unfamiliar persistent identifier arrives, retain the event for investigation. Similar wording is not sufficient evidence to invent the missing business meaning.
For discussion, imagine a registry entry called demo-poll. Its demo-integration option maps to an integration-interest field and demo-reporting maps to a reporting-interest field. These are synthetic internal labels, not real Telegram identifiers or a network request example. A production map must be built from the actual data received for the poll you sent.
Store current answers separately from history
An answer key should include the tenant, bot or connection, poll_id and a typed participant identity. For a person, that identity includes user.id. Supporting a chat voter requires a distinct branch, if your business process permits it at all. Names and usernames may help an operator recognize a participant, but must not be the only link to a CRM record. A known Telegram ID still does not prove ownership of a particular enquiry; your authorized process must establish that association.
A newly accepted answer supplies the participant's current set of selections. Replace the previous set with that set. Do not append every received choice to an ever-growing list of interests. In a multiple-answer poll, someone may remove one selection while retaining another. A union of old and new choices would preserve a preference they already withdrew.
The API permits empty option_ids and option_persistent_ids when a vote is retracted. Model that as a distinct retracted-answer state. It is neither malformed JSON nor a new interest with an empty label. A retraction must not delete the person's entire CRM record. Remove the contribution of this particular question and reassess dependent automation under a separate policy. Work already performed by a person may not be appropriate or possible to undo automatically.
Keep an event history for diagnosis: what arrived, which checks passed, which mapping version was used and which state resulted. Retain only the data necessary for the purpose, with appropriate access controls and retention limits. Real participant answers and identifiers do not belong in public logs, demonstration files or search analytics.
Prevent retries and delayed workers from restoring old choices
Persist the incoming event before applying it to an answer projection - the current record destined for the CRM. A repeated delivery of the same update_id for the same bot must not create another task. The underlying separation is covered in the article on repeated webhooks and duplicate CRM leads. Poll answers additionally need version control at the participant-answer level.
A queue with concurrent workers can finish an older change after a newer one. Serialize work for an answer key or use an atomic version check in storage. The completion times of CRM HTTP requests are not a reliable history of the participant's actions. Neither the receiver's timestamp nor the operator's record timestamp is, by itself, a dependable user-action order.
Telegram describes update_id as useful for restoring update order. After a week without new updates, the next identifier may be chosen randomly. Do not turn "the larger number is newer" into an everlasting record rule. An initial implementation can preserve intake order and explicitly track delivery-epoch breaks. When the ordering context is lost, hold automatic promotion and reconcile the state. This is a conservative application policy, not an API promise to recover a complete voting history.
A timeout while changing a CRM record may leave the write outcome unknown. Before repeating a side effect, read the available destination state or use an idempotency facility actually supported by your CRM. A successful webhook response alone does not establish exactly-once business processing.
Acceptance procedure for your test environment
Use a synthetic participant, a separate enquiry and a registered test poll. For each case, record the input event, prior answer, map version, processing decision, resulting record and side-effect count. The outcomes below are acceptance criteria. They are not claimed results of a live network test performed for this article.
- Submit a known answer with one selection. Expect one question result for the authorized participant. An already linked contact must not become a new lead merely because it answered a survey.
- Replay the identical event with the same
update_id. Neither the answer nor the task count should change. - Submit a subsequent answer with a different selection. The old option should disappear from this question's current answer while the change history remains subject to the retention policy.
- For multiple answers, submit two choices and then only one. The current state should contain only the latest complete selection set.
- Submit a retraction with empty arrays. Expect a retracted state, an intact contact record and no new outbound message.
- Submit an unknown poll, an unknown option and a
voter_chatevent withoutuser. Each case should enter its defined rejection or review branch, without guessing a personal answer. - Delay old work until after newer work, restart a worker and simulate an unknown CRM write outcome. An old choice must not silently become current; ambiguity must remain visible to the operator.
- If changing option sets is supported, delete an option and add another in your test environment. Verify the meaning through persistent identifiers. A library that loses those fields cannot qualify that mode for production use.
Then perform an authorized integration test with your own bot. This covers what synthetic events cannot prove: the effective poll configuration, the fields exposed by your actual library and the behavior of your CRM. Do not send experimental surveys to real customers without an agreed scenario. Keep request secrets out of the test report and record destination reads separately from local handler assertions.
When to keep personal automation on hold
Use aggregate results or manual review when a poll is anonymous, was not sent by your bot, is unregistered, has lost its option map or lacks a verified participant-to-enquiry association. The same applies to unknown event ordering and ambiguous CRM writes. A held queue should expose a reason and an owner, so a safety stop does not become a silently lost enquiry.
Start with one question and a limited test flow. Agree on the allowable meanings, retraction policy and manager actions before executing the acceptance procedure. Expand only after checking the final state and the failure branches. Evaluate services on 24go.site against their actually confirmed capabilities. Hosting this guide does not mean that the described handler is already available in the product.
Sources
- Telegram Bot API: Update - conditions for
pollandpoll_answer, andupdate_idbehavior. - Telegram Bot API: PollAnswer - voter identity, selected options and retraction.
- Telegram Bot API: PollOption - persistent option identity.
- Telegram Bot API: Poll - poll state and configuration.
- Telegram Bot API: sendPoll - sending a poll and parameter defaults.