Short answer: pick the three questions your customers actually ask most, phrase them in your customers' words, and make sure the assistant answers all three well. Suggested questions are not decoration. They are a promise, and the fastest way to lose a visitor is to suggest a question your assistant cannot handle.
Why suggested questions carry so much weight
Opening a chat panel is easy. Typing the first message is not. There is a small hesitation where the visitor has to compose a question, worry it is too vague, and commit. Plenty of people close the panel during that pause.
Suggested questions remove the pause. A tap replaces a sentence, and once the first exchange has happened the conversation flows because the ice is broken. On mobile this matters more, because typing on a phone is genuinely tedious and a well-chosen chip can carry someone to an answer without a single keystroke.
They also do a quieter job: they teach scope. A visitor reading three suggestions learns instantly what this assistant is for. That is the same job the greeting does, done visually and faster.
Where to get the questions from
Do not invent them. You already have the data in four places.
Your inbox. The last fifty emails to your contact address. Whatever appears four times is a suggested question.
Your phone. What does whoever answers the phone repeat every day? That is your top question, and it is usually about price, availability, or whether you cover a particular area or need.
Your search console. The queries bringing people to your site are literal phrasings of what people want to know. They also come pre-written in customer language.
Your existing chat transcripts. If you already run a widget, this is the best source of all, because it is the same channel and the same intent. Read a month of them and the pattern is obvious.
Phrasing rules
Use their words, not yours. You call it "restorative dentistry"; they ask "how much for a crown". You call it "discovery engagement"; they ask "what does a project cost". The chip should be phrased the way a customer would type it, because that is what makes it feel like a shortcut rather than a menu.
Make them questions, not categories. "Pricing" is a navigation label. "How much does a call-out cost?" is something someone wanted to know. Questions get tapped; categories get scanned and skipped.
Keep them short enough not to wrap. A chip that wraps onto three lines on a phone looks like an error. Aim for something that fits on one line at mobile width, two at most.
Do not stack similar questions. Three variants of a pricing question wastes two slots. Spread across the distinct things people want: what you do, what it costs, and how to get started is a solid default trio.
Three is usually the right number
More suggestions do not produce more engagement. They produce a menu, and a menu invites reading and comparison rather than action. Three is enough to teach scope and cover the common intents while staying scannable at a glance on a phone.
If you have four genuinely distinct top questions and one of them is seasonal, rotate rather than adding a fourth. Which brings us to the most useful habit here.
Change them when reality changes
Suggested questions should not be a set-and-forget decision. They are the cheapest lever you have for steering conversations, and the right set changes with your business.
- Seasonal shifts. A garden business should suggest different things in March than in November.
- Promotions. If you are running an offer, suggest the question people will have about it.
- Recurring confusion. If your transcripts show the same misunderstanding four times a week, promote it to a suggestion so it gets answered before it becomes an objection.
- New services. A new offering nobody knows about gets discovered faster through a suggestion than through a nav item.
Editing them in Clerkzo takes a minute in Widget Studio, with a live preview so you can see how they sit before publishing. No redeploy, no developer.
Suggested questions and lead capture
There is a tempting move here: make one suggestion a booking or quote request, so the panel doubles as a conversion path. This works, with one condition. The suggestion has to be phrased as something a visitor wants, not something you want.
"Get a quote" is your goal. "How much would my job cost?" is theirs. Both lead to the same place; only one gets tapped by someone who has not decided yet. The pattern is covered further in lead capture chatbot, and the comparison with a static form is in chatbot vs contact form.
Do not make all three suggestions conversion prompts. A panel that only offers ways to give you your details reads as a sales trap, and people close it. Two informational, one transactional is a comfortable balance.
Test that the answers are good
This is the step people skip. A suggested question is a promise that the assistant handles this well. If the answer is vague, hedged, or wrong, you have taken your most confident visitor and disappointed them at the first exchange.
So before publishing, tap each suggestion yourself and read the answer as a stranger would. If the answer is thin, the fix is usually on your website, not in the widget: the assistant trains by crawling your own site, so a question your site does not really answer will produce a reply your assistant cannot really give. Adding a proper pricing paragraph or a coverage-area page fixes the answer at the source.
What suggested questions are not
They are not a menu system. Some widgets turn suggestions into a decision tree where every answer produces another set of buttons and the visitor never types anything. That is a phone menu wearing a chat costume, and it frustrates people who arrived with a specific question that is not on the list.
Suggestions should be a starting nudge, not a rail. The input field stays available, free-text questions get real answers, and the assistant handles the ninety percent of questions you never thought to suggest. That is the whole advantage over a form.
Frequently asked questions
How many suggested questions should I show?
Three. Enough to teach scope and cover distinct intents, few enough to scan instantly on a phone.
Should suggestions be phrased as questions or labels?
Questions. Labels read as navigation and get skipped; questions read as shortcuts and get tapped.
Can I change them without touching my site's code?
Yes. In Clerkzo they are edited in Widget Studio with a live preview, and publishing does not touch your embed. See how it works.
Should one of them be a booking or quote prompt?
One of three, phrased from the visitor's point of view. All three being conversion prompts reads as pushy.
What if the assistant answers a suggested question badly?
Fix the source. The assistant learns from your site, so a weak answer usually means the page behind it is thin. Improve the page and the answer improves with it.