A Botpress website bot that stays silent can have an unpublished change, a broken Webchat configuration, a usage restriction or a workflow waiting without sending a reply. Test the published bot separately from your website first, then compare it with Studio. That split tells you whether to inspect the embed, publication, account usage or conversation logic.
The Complete Excel Course with AI Integration
From formulas and pivot tables to Power Query, DAX, VBA and AI with Copilot, ChatGPT and Claude — hands-on from the first lesson.
By Sagnik Bhattacharya and Virginia Thorn, Primetime Emmy Award winner
Buy nowSave the exact question, page address and visible error before changing anything. “The chatbot is broken” is too broad to diagnose. “The chat opens, accepts two messages and stops after the service choice” is useful evidence. Keep your ordinary enquiry route visible while investigating so a chat fault does not block customers from contacting the business.
How do you find out why your Botpress bot is not responding?
Test three surfaces to find out why a Botpress bot is silent: your actual website, Botpress's published preview and the Studio Emulator, and record the outcome on each. Open the preview from your agent's Overview section in the Dashboard by selecting Preview, or use Studio's Share button, which gives a hosted Webchat that is already the production bot. They answer different questions. The website includes your site's scripts and security settings. The published surface checks the version customers receive. The Emulator, in Studio's right-hand panel, helps inspect the workflow under development.
| Observed symptom | Test first | Evidence to retain |
|---|---|---|
| No chat bubble on the website | Open the published bot independently | Page address and whether the standalone bot loads |
| Bubble opens to an error screen | Read the exact error label | Configuration, quota or network error text |
| Studio works, published bot does not | Confirm the latest intended version was published | Expected wording and published wording |
| Stops after one particular choice | Trace that branch in the Emulator | Last completed step and next expected reply |
| Replies to greetings but not FAQs | Check knowledge access and answer delivery | Question, source and retrieval status |
An illustrative pest-control firm checks 12 customer-style sessions after adding a new service menu. Eight stop after “inspection”, while four finish the general FAQ route. The standalone published bot fails the same way; the website is therefore a poor first target for repair. In Studio the inspection choice reaches an unfinished branch. Connect that branch to the intended reply and repeat the same 12 sessions.
After the illustrative repair, require all eight inspection cases to receive either the approved inspection explanation or an explicit request for more information, while the other four retain their previous answers. Twelve out of twelve expected outcomes is an acceptance result for this tiny test set, not a claim about overall accuracy. Keep the original failures as regression checks, meaning checks that ensure an old fault does not return.
If every surface is failing suddenly, check the official Botpress status page before editing the bot. A service incident and a local workflow fault need different responses. The page lists Webchat, Studio, Integrations and the API as separate components, so a fault on one row can explain a silent widget while everything else shows as operational. Record what the status page actually says at the time of your check; do not assume that a green page proves your particular bot is configured correctly.
Why does the Botpress chat widget not appear on your website?
A Botpress chat widget needs a published bot and the current Webchat snippet, copied from Webchat > Deploy Settings in the Dashboard and installed on the page customers visit. Copy that snippet from the intended bot rather than borrowing a script from an old tutorial. The documented installation places it in the head section of the website's HTML. If the site runs on WordPress, Wix or Webflow, Botpress has a separate installation page for each of those builders, so follow that one instead of pasting code into a random theme field.
Ask the website maintainer to verify the snippet on the actual page customers visit. A header change in a draft website theme may never reach the published theme. A landing page may use a separate template. A caching layer may still serve an earlier page. These are hypotheses to test by inspecting the delivered page, not reasons to keep pasting additional snippets.
For example, an illustrative cleaning company has chat on its main service page but not on its campaign page. The published standalone bot answers on both test devices. The website maintainer finds that the campaign uses a separate page template without the approved embed. Apply the snippet to that template once, then test the campaign address directly instead of repeatedly republishing the bot.
Check for duplicate installations. A manually added snippet and a second copy inserted by a website extension can make the page difficult to reason about. Ask the maintainer to identify every place the Webchat loader is added and retain one intended installation. Do not delete unfamiliar website scripts broadly; another business function may depend on them.
If the page requires consent before optional scripts run, test both permitted and declined paths. Do not bypass the site's existing choices to make the chat appear. Ensure the ordinary contact form or contact page remains usable in either case. A deliberate script restriction is different from a faulty bot and should have a clear customer alternative.
For site-specific embedding detail, use the Botpress website-embed testing tutorial where relevant. Keep a short request for the maintainer: “Please check that this page loads the current snippet from this bot once, and send the exact blocked-resource or configuration error.” That is more actionable than asking them to reinstall the entire website integration.
What do Botpress Webchat error messages mean?
When Botpress Webchat hits an error it replaces the chat with a dedicated error screen. The official Webchat error reference lists each one with its cause: missing or invalid identifiers, inaccessible configuration, a blocked bot, quota errors and network disconnection. Treat the wording as a clue. An invalid client ID does not call for uploading the FAQ again, and a disconnected network does not prove that the knowledge source is empty.
If the client ID or bot ID is missing or invalid, return to the intended bot's current embed configuration. If the config URL is invalid or access is denied, have the maintainer compare the installed snippet with the source configuration. Preserve the exact error without publishing access tokens or unrelated customer data in a help request.
A furniture maker's illustrative test page shows “Invalid config URL” immediately after the bubble opens. The copied configuration address was shortened while moving it into the website editor. Replace it with the intact current snippet and retest the standalone and website versions. Changing the bot's tone, model or opening message cannot repair an address the browser cannot load.
Two error screens point away from the website entirely. “Bot is blocked” means the bot has been temporarily paused, and the reference sends you to the bot's status in the Dashboard. “Quota exceeded” means the bot has reached a usage limit, so the remedy sits in the workspace's Usage menu, covered below, not in the embed code. If a maintainer starts editing the website after seeing either message, stop them and save their time.
For “Network error” the documented fix is to check the internet connection, and for “SSE disconnected” it is to refresh the page. SSE (Server-Sent Events) is the connection Webchat uses to receive a stream of updates from Botpress. If either keeps returning, test another connection or device. If the problem occurs only on one managed network, give its administrator the exact failing resource and time. Do not ask an owner to disable security protections across all company devices.
When “Not a participant” appears, the vendor reference suggests clearing browser storage and refreshing. Test that in a controlled browser session, understanding that clearing storage can remove local session state. Avoid instructing customers to erase all their browser data. First establish whether it is a single stale session or a reproducible problem affecting fresh visitors too.
How do you check which version of the bot is answering?
Confirm the bot was published, then open a new published session and ask a question whose approved wording recently changed, because a changed draft and the customer-facing bot may differ. Do not use only the welcome message if the failing branch is several steps later. Record an innocuous diagnostic difference, such as “inspection request” versus “inspection enquiry”, rather than adding customer-visible technical labels.
Imagine a garage updates its opening response to offer three categories, but the public bot still offers two. The draft's third category works in the Emulator. Confirm that the right bot was published and that the website points to that bot before recreating the third branch. Rebuilding correct draft logic would add risk while leaving the wrong public version in place.
Check account and bot identity carefully where several similar projects exist. A test bot and a live bot can have almost identical names. Record the bot identifier privately in the maintenance sheet, together with the live page address and person allowed to publish. The aim is to make “which bot did you change?” an answerable question rather than a guessing exercise.
Botpress's Versions menu, the clock icon in Studio's left-hand menu, holds earlier bot states. A version is saved automatically every time you publish, and you can add one manually with the + button. The documentation warns that reverting completely overwrites the current bot, including unsaved changes, and that pay-as-you-go users have access to three saved versions (Team plan subscribers 15). Three is not many when every publish creates a version, so before experimenting, check which versions you still hold, annotate the last accepted one with its test result, and restore that known target rather than working back through the list at random.
If you want to keep the maintenance sheet and test log in a spreadsheet, try The Complete Excel Course with AI Integration on Coding Liquids.
Has your Botpress bot run out of conversations or AI spend?
A visible Botpress widget can still face an account usage restriction, so open the workspace's Usage menu (the pie-chart icon) and check both conversations and AI Spend. At the time of writing, the pricing page lists Free with 100 included conversations, no extra packs and no auto-recharge, and it does not label that allowance with a monthly renewal period. Do not infer the reset date from an older tutorial. The workspace documentation says AI Spend covers generating text and retrieving answers from knowledge bases.
Botpress's pricing FAQ counts an exchange with at least two end-user messages in the billing month as a conversation. It also says a conversation spanning two months counts in both, and anonymous sessions count separately. Those rules make repeated testing relevant to capacity. They also explain why the number of visible chat messages is not the same as the number of billable conversations.
Testing documentation needs care. A note on Botpress's Emulator page says Emulator conversations are not saved and bot invocations are not counted against your usage quota, though AI operations are; the pricing FAQ, by contrast, says emulator conversations are counted the same as any other. Budget using the pricing wording and check the account meter. Do not run hundreds of fresh test sessions on the assumption that all emulator work is exempt, especially when AI operations are involved.
For an illustrative allowance calculation, 76 conversations already recorded plus 18 new qualifying test conversations gives 94. If the relevant allowance is 100, that leaves six before any other usage. This is arithmetic for planning, not evidence of your current balance. Open the account usage view to confirm the actual number and whether the tests are being counted as expected.
The pricing page lists Plus at $150 a month billed annually, with 250 conversations a month and $25 of included AI usage. Paid plans are designed not to stop: at 95% of the conversation quota a pack of 100 is added automatically ($65 on Plus), with notices at 75% and 100%, and at 95% of the AI spend limit Botpress buys a $10 credit grant on your behalf. The FAQ says that credit-grant auto-recharge is not available on Free. So a silent bot on a paid plan is less likely to be a quota problem than one on Free. Do not upgrade impulsively as a diagnostic step; identify which meter caused the restriction first.
A slow first reply is a different symptom from silence. Botpress starts a fresh compute instance when a conversation begins, which its documentation calls a cold start, and the Always Alive option, available on Plus and above, keeps an agent running so the first message is processed much faster. If visitors report that the bot answers only after a pause and some give up before it replies, time the first response in a fresh session before deciding the bot is broken.
The Botpress plan-cost tutorial helps assess the longer-term arrangement. For a silent bot, the immediate task is narrower: establish whether the dashboard reports a quota restriction, retain that evidence and use the correct account remedy. A new FAQ document cannot restore replies blocked by account capacity.
Why does a Botpress conversation stop after one message?
A Botpress conversation stops when its branch is waiting, has ended or has no appropriate route to the next action. A node is a step in the conversation; cards inside it perform tasks such as sending a message or capturing information. Standard Nodes execute cards in order, while Autonomous Nodes use AI to decide how to respond or use their available tools.
Reproduce the failure in the Studio Emulator (Ctrl or Cmd + E opens it) and read the in-line statuses under each message. They include “Waiting for user input”, “Answer found in Knowledge Base”, “No content matched the query”, “Error occurred” and “Conversation ended”. That last one is the classic cause of silence: the documentation says a conversation ends when it reaches an End node or when there are no possible transitions left, so a branch with no outgoing link simply stops. Ask what the next observable event should be. If the bot expects the user to click a choice that never appears in Webchat, typing more greetings does not test the intended transition.
In an illustrative cleaning flow, the bot says “Choose a service”, but no choices were configured on the capture step. Staff in the editor know to type “regular”; customers do not. Add the intended choices or a clear typed response instruction and test fresh sessions. The repair concerns the conversation design, not the AI's general intelligence.
Another illustrative branch ends immediately after capturing a preferred day. The owner expected a confirmation sentence, but the workflow goes straight to End. Add an explicit message before the ending and distinguish a captured preference from a completed booking. “You requested Thursday; please contact the team to confirm” is honest when no calendar action has occurred.
Use the Debugger and Logs to inspect the sequence when the owner-level checks do not reveal the problem. The official documentation describes logs as a step-by-step record of what happened before an error. Studio's own Logs tab sits in its bottom panel; for the live bot, open the Dashboard, select the workspace and bot, then the Logs tab, which holds the production logs from the deployed bot. Retain only the relevant error and non-sensitive identifiers for a helper. Do not paste the complete raw conversation history into an unrelated tool just because the debugging panel exposes it.
If you want help tracing one failed conversation, I offer one-to-one AI consultations for small businesses.
Why does the bot answer greetings but not FAQs?
A Botpress bot that answers greetings but not FAQ questions may have a knowledge problem rather than a broken embed. Choose one question with a known answer in an approved source. For the pest-control example, use “Does the inspection include treatment?” and a source sentence that clearly says it does not. Keep invented service terms labelled as illustration when practising.
Check that the source is available, current and connected to the node that should answer. A file existing in the workspace is not proof that the active conversation step can search it. The Emulator documents “Knowledge disabled on that node” and “No content matched the query” as different statuses. The first points to access; the second points to the question, source wording or retrieval. For the second case, the knowledge-base troubleshooting notes say the logs show the exact query the bot sent to the knowledge base and any matches, with source name and a content preview. If the query reads “inspection treatment included” while your source says “The inspection fee covers the survey only”, rewrite the source in the words customers use.
The Knowledge Agent has an Answer Manually option. When enabled, generated answers are placed in a variable instead of being sent automatically. An illustrative owner turns it on while experimenting, sees evidence of a retrieved answer and assumes customers receive it. Either deliberately send the stored answer in the workflow or restore the intended automatic behaviour; verify the actual visible reply afterwards.
Repair an uncertain answer with a useful fallback. For example: “I could not confirm that from the approved service information. Please use the enquiry form and the team will check.” Do not promise a staff callback unless you have a working capture and handover process. The beginner FAQ-bot tutorial covers preparing sources and designing that first bounded bot.
Close the incident with a small evidence pack: the original failing question, the cause found, the change made and results on the website, published bot and Emulator. Repeat one unrelated FAQ and one fallback case to check for side effects. If you need a helper, use the Botpress hiring-scope tutorial and request a reproducible fix with a handover, rather than an unexplained replacement bot.
Put a time boundary on investigation so it does not become an afternoon of unrecorded edits. A suggested owner-led pass is ten minutes to capture symptoms, ten to compare the three surfaces and ten to check publication and account usage. These are planning allowances, not a promised repair time. If the failure remains unexplained, the evidence collected is the useful output to give a helper.
A filled-in handover could say: “Inspection branch stops after the second visitor message on the live service page and published standalone bot. General FAQs respond. The same inspection input waits at the service-choice node in Studio. Usage shows capacity remaining. No website changes made.” That narrows the job substantially without exposing customer details or asking someone to inspect every part of the website.
After recovery, remove any temporary diagnostic wording and check that the contact alternative remains visible outside chat. Ask a colleague to perform the original customer journey without coaching. If they need to know an undocumented phrase to make the bot continue, the silence may be technically fixed but the customer experience still needs repair. Record that as a separate task rather than closing the issue on the strength of one successful greeting.
Further reads
- Botpress Calendly Integration: How a Website Bot Books Calls — Check a booking handoff after chat replies recover.
- How to Connect Botpress to Google Calendar to Book Appointments — Separate a calendar-action failure from a chat failure.
- How to Connect Botpress to WhatsApp: What You Can Do Alone — Understand a different channel's setup boundary.
- ManyChat vs Botpress: Which Is Easier for a Non-Technical Owner? — A decision guide and a one-afternoon ease test for choosing between ManyChat's no-code flows and Botpress's more controllable builder.
Browse all AI-for-business tutorials
Sources: Botpress Webchat error handling and Quickstart; Botpress Studio Preview, Emulator, Nodes, Knowledge Bases, Knowledge Agent, Versions, Debugger and workspace usage documentation; Botpress pricing page; Botpress status page.
