Chatbot Suggested Questions: 40 Starter Prompts by Industry

Under the greeting of most chat widgets sit a few tappable questions. They are the shortest way into a conversation, because one tap replaces typing, which matters most on a phone. They are also easy to get wrong. In a March 2026 study of eight site chatbots, Nielsen Norman Group found that most site AI chatbots "do a poor job of communicating what they can actually help with", and its advice was blunt: "The chat's prompt and intro message should be specific to the situation."

This guide is for whoever sets up the chat widget: a store owner, a SaaS founder, a clinic manager or an agency building sites for clients. It covers how many suggested questions to show, eight rules in one table, 40 examples for seven industries with the source each answer needs, what to do when pages need different prompts, and how to measure which prompts start useful conversations. The greeting above the buttons has its own guide, the chatbot welcome message examples; this one is about the buttons.

PepoChat is our product; its behaviour below was checked in its code and documentation on 22 October 2026, and every outside source was read the same day. The examples work in any chat tool that shows suggested questions.

What are suggested questions in a chatbot?

Suggested questions in a chatbot are short, tappable prompts, usually two to four, shown under the greeting before the visitor types anything; one tap sends the prompt as the visitor's own first message, and the bot answers it like a typed question.

Suggested questions are a shortcut and a lesson at once. A visitor who sees "Can I book for this week?" learns that the bot can book, even if they came to ask about parking. Nielsen Norman Group's June 2025 article on designing use-case prompt suggestions puts it in one line: "Use-case prompt suggestions are examples of good AI prompts." It adds that "specific and targeted suggestions help users quickly determine relevance", while broad ones rarely work.

Vendors use other names for the same thing. Conversation starters, called prompt starters in Microsoft Teams, are suggested questions by another name. Quick replies are Meta's term for buttons under any Messenger message, which disappear once the next message arrives. Follow-up suggestions appear after an answer and propose the next question; this guide is about the starting set.

What a tap does varies. Messenger posts the tapped button's title as the user's message, and PepoChat does the same; Teams prompt starters put the text in the compose box so the user can edit it first. Inserting suits prompts that need a detail only the visitor knows, such as an order number.

In PepoChat, the Default suggestions on the Widget Customization page appear as rounded chips under the greeting and disappear once the conversation starts. With no published Help Center articles, the Home screen uses a chat-first layout, so the greeting, chips and message box are the first thing visitors see. With published articles, Home shows a help search and a Send us a message card, and the chips appear one tap later, under the greeting of the new conversation.

How many suggested questions should a chatbot show?

A website chatbot should show three suggested questions or fewer, because three cover the main reasons visitors open a chat, fit under the greeting on a phone, and match the cap that WhatsApp and Microsoft Teams set for suggestion buttons.

Limits from each platform's developer documentation and PepoChat's code, checked 22 October 2026.

WhatsApp's reply buttons let a business "send up to three predefined replies for users to choose from", according to Meta's developer documentation, and Teams displays only the first three suggested actions even when a bot sends more. Messenger and Google's RCS allow 11 to 13, where buttons often act as menus inside scripted flows, such as picking a size. The character caps tell the same story: 20 to 25 on messaging platforms, 40 in Chatbase's dashboard.

Research on choice gives a more careful answer than "fewer is always better". Choice overload is the drop in satisfaction or willingness to choose when people face too many options. A 2015 meta-analysis by Alexander Chernev, Ulf Böckenholt and Joseph Goodman in the Journal of Consumer Psychology, covering 99 observations from 7,202 participants, found it is not automatic. As the Kellogg School's summary of the paper explains, it grows with the complexity of the options, the difficulty of the decision, uncertainty about what one wants, and a goal of keeping effort low.

A visitor who opens a chat to avoid reading your pages arguably has that low-effort goal, and three distinct questions keep the comparison trivial. Space matters too: PepoChat's chat window is 400 by 600 pixels and shrinks to fit a phone, so each extra chip claims room the greeting needs. Three is a ceiling, not a target. Two questions your content answers well beat three with a weak one, and an AI agent always has a fourth option: the text box.

What are the rules for good chatbot suggested questions?

Good chatbot suggested questions follow eight rules: phrase them as visitors type, suggest only what your content answers well, give each a different job, keep them short, skip account-only tasks, keep people optional, match the site's language and avoid end dates.

The rules are a checklist drawn from the sources above, not a measured ranking.

The three-job mix is one question that helps a visitor decide, one that helps a customer do something, and one that settles a worry. Decide questions come before a purchase or booking: fit, price, coverage. Do questions are tasks: an order, a booking, a setup step. Trust questions remove risk: returns, cancellation, guarantees, credentials.

Every chip also needs an answer source, the page, document, Q&A pair or connected action that the agent answers it from. Before you publish a chip, find the sentence that answers it, and write one if it is missing; the knowledge base starter kit lists 50 questions a site should answer. In PepoChat, the surest fix is a Q&A pair whose question is the chip's exact wording, because the question is stored with its answer and retrieval matches that phrasing. Q&A pairs do not count as knowledge sources.

A chip that offers a person is optional: PepoChat's agent hands the conversation to the team inbox whenever a visitor asks for one. Add it only when many chats genuinely need a person and someone will answer.

40 chatbot suggested questions by industry

The 40 chatbot suggested questions below cover seven kinds of business, eight each for online stores and SaaS, six each for clinics and courses, and four each for agencies, home services and hotels, each tagged with its job and the answer source the agent needs.

If you need only three, start with this table, then read the detail for your industry.

Swap in your own places and products, and read each table's last column as a to-do list: a chip without that source behind it is a promise the bot cannot keep.

Suggested questions for online stores

In PepoChat, "Where is my order?" works once Shopify or WooCommerce is connected: the agent asks for the order number and reports the order's status, though Shopify lookups return no tracking number. The ecommerce chatbot guide covers twelve store questions in depth.

Suggested questions for SaaS products

SaaS visitors split into evaluators on the pricing page and customers stuck in setup. How-to chips are safe because the agent can explain any documented step; do-it-for-me chips are not, unless an action exists. PepoChat's Stripe actions, for example, find a customer, check a subscription and cancel at the end of the period, but cannot change a plan or refund.

Suggested questions for clinics, salons and appointment businesses

Booking chips need a live calendar; in PepoChat that is a connected PepoSmart calendar whose open slots the agent offers in the visitor's local time. Keep diagnosis out: "Do you treat lower back pain?" asks about a service, while "Is my back pain serious?" asks for advice the agent should decline.

Suggested questions for agencies and consultancies

Agency pricing is usually bespoke, so a cost chip needs a page that explains how quotes work and gives typical ranges, never a number the agent has to invent.

Suggested questions for online courses and coaches

Login trouble makes a fair chip when your course platform publishes the reset steps, because the agent can walk through them; repairing an account still needs a person.

Suggested questions for home services and trades

A service-area page that lists towns or postcodes turns "Do you cover my area?" into an instant answer instead of a callback.

Suggested questions for hotels and vacation rentals

Rules that vary by rate, such as cancellation terms, belong in the knowledge base under each rate's name, so the agent can say which rule applies instead of quoting the strictest one.

Should suggested questions change by page?

Suggested questions should ideally match the page, because a visitor reading pricing and a visitor stuck in the docs need different first questions; whether you can do that depends on what your chat tool supports.

Several tools let a page change or prefill the conversation, according to their own documentation:

• Chatbase's chat bubble API can replace the suggested messages for the current page load, up to four of at most 200 characters, and open the bubble with a message that is sent straight away.
• Intercom's Messenger JavaScript methods include showNewMessage, which opens the Messenger with the composer pre-filled, and startConversation, which sends the message immediately.
• Crisp's JavaScript SDK pre-fills the message box with message:text, and Tidio's messageFromVisitor sends a message on the visitor's behalf, which Tidio suggests for a button on your page.

PepoChat does not offer this. As of October 2026, a PepoChat workspace has one greeting and up to three suggested questions, shown on every page where the widget runs; there is no per-page set and no API to open the chat with a prefilled question. The JavaScript API has four methods, show(), hide(), destroy() and init(), and init() takes the workspace, position and preload setting, not suggestions.

What works within that limit:

1. Write the set for the page where chats start. Your web analytics shows where people open the chat; choose chips that also make sense elsewhere.
2. Put page-specific prompts on the page. A line such as "Questions about sizing? Ask our assistant" beside a button that calls EchoWidget.show() opens the chat with the visitor already knowing why.
3. Let the Help tab serve the docs. With published Help Center articles, Home shows a help search and your top four articles on every page.
4. Split genuinely different sites. A shop and a booking site with different audiences can be separate workspaces, each with its own chips. Each account gets one Free workspace; further workspaces need a paid plan.

How do you measure which suggested questions work?

Measure suggested questions by counting how many conversations begin with each chip's exact text and how those conversations end; as of October 2026, PepoChat does not report widget opens or chip taps as a number.

We found no published, method-stated benchmark for how often website visitors tap suggested questions, and PepoChat publishes none, so compare chips with each other over time. A month of conversations is enough to start:

1. Export the conversations. In the team inbox, Export CSV takes a From and To date. Each transcript lists the messages in order as assistant:, user: and operator:, starting with your greeting.
2. Count the chip starts. A conversation began with a chip when its first user: line matches the chip word for word. A spreadsheet count such as =COUNTIF(H:H,"*user: Where is my order?*") is close enough.
3. Work out the share. Divide each chip's count by the rows with a message count above 1, the conversations where the visitor wrote something.
4. Read the silent rows. On the standard Home screen, Send us a message creates the conversation before the visitor types, so a row with a message count of 1 is usually someone who saw the greeting and chips and left. The chat-first layout creates no row until the first message.
5. Check the outcomes. For each chip, count its conversations that were escalated, got a thumbs-down vote, or show the visitor rewording the question.
6. Mine the typed openers. The first messages visitors typed instead of tapping are the candidates for your next set.

Two biases need correcting. The Topics card in Analytics labels each conversation from its first message, so a popular chip inflates its own topic; judge demand from typed openers. Position matters too, and NN/g's prompt-suggestion article warns: "Engagement rates are often influenced by placement, not just content." It suggests randomizing the order; PepoChat always shows Suggestion 1 first, so swap the order for the next period and compare.

The export also includes chats from the Test the agent drawer, so finish tapping your chips there before the measuring window opens. Change the set when your content changes, and otherwise about once a quarter, never halfway through a measuring window.

What should a chatbot never suggest?

A chatbot should never suggest a question it cannot answer from your content or connected tools, a task that needs the visitor's account, advice it must refuse, or three versions of the same question.

• Questions without an answer source. "Is this back in stock?" without live stock data, or "When exactly will it arrive?" when only the carrier knows. The chip advertises a dead end.
• Tasks behind an account the agent cannot reach. "Change my plan" invites a request the agent can only redirect; "How do I change my plan?" is the answerable version.
• Advice your instructions forbid. "Is my back pain serious?" or "Do I have a case?" sets up a refusal. Put that boundary in the agent's rules, as the support chatbot instruction examples show, and keep it out of the chips.
• Duplicates, labels and small talk. Three phrasings of one shipping question, one-word labels such as "Pricing", or "Tell me a joke" waste the space, and in PepoChat every tap spends an AI reply.
• Chips behind a form wall. With lead capture set to ask before the first message, a PepoChat visitor who taps a chip sees "Please enter a valid email address" until the form is filled. The lead generation guide explains when asking later works better.

How do you set suggested questions in PepoChat?

In PepoChat, suggested questions are set on the Widget Customization page under Default Suggestions, saved with Save Settings and tried in the Test the agent drawer, on every plan including Free.

1. Open Widget Customization and find Default Suggestions in the General Chat Settings card, under the greeting.
2. Type up to three questions in Suggestion 1 to Suggestion 3, in the order you want them shown, and leave a field empty to show fewer. If all three are empty, the chat-first screen shows three built-in chips, "What can you help me with?", "How do I get started?" and "I'd like to talk to a person", while a conversation opened from the standard Home screen shows none.
3. Click Save Settings, after checking the chips in the live preview.
4. Click Test the agent and tap each chip. Test chats are free and stay out of analytics, and saved suggestions reach an open drawer without reopening it.
5. Compare each reply with its answer source, and fix the source before you blame the chip.

Each live tap is a real first message, so the agent's reply counts toward your monthly AI replies, 200 on Free and 2,000 on Starter, while the greeting never does. Every field is described in the widget customization docs.

What to do next

• Pick the starter three for your industry, confirm each has an answer source, and tap them in a test chat before they go live.
• For the rest of the widget, from placement to accessibility, work through the website chatbot best practices.
• To try suggested questions on your own content, create a free workspace: the suggestions, the test drawer and every feature on the free plan, with no credit card.