Live Chat Software Demos: Ten Questions to Type Into Any of Them

The reason this matters: a chat widget looks identical on every vendor's website. Rounded bubble, bottom right, friendly greeting. You cannot tell them apart by looking. What separates them only shows up under specific questions, at specific times of day, on your own content. So the useful question is not "can I see a demo" but "what should I do with the twenty minutes once I have one".

Who lets you start today, and who wants a call

We make Asyntai, so read that row as a disclosure. The general point holds regardless of vendor: a free tier you keep beats a trial that expires, because the questions worth asking take longer than fourteen days to occur to you.

A trial that expires in 14 days tests the software for 14 days. A free tier tests it in your actual working life.

Why a vendor demo tells you so little

Watch what happens in a scheduled demo. Someone shares a screen showing a fictional shop with clean product data, an inbox with three tidy conversations, and a knowledge base written specifically to answer the questions they are about to type. Every answer is correct because the whole environment was built so it would be.

None of that is dishonest. It is just a demonstration of the interface, and the interface was never your problem. Your problem is whether it handles a customer asking about the compatibility of a product you discontinued last year, at 11pm, on a phone, in Spanish.

So treat a scheduled demo as what it is: a walk through the menus and a chance to ask pricing questions to a human. Do the real evaluation yourself.

The thirty-minute test, on your own site

This is worth doing before you talk to anybody. It works for any tool that lets you sign up without a call.

1. Install it on a real page, not a test page. Your busiest product page or your pricing page. A blank sandbox proves nothing because the widget has no context to get wrong.
2. Give it whatever content it accepts. Your URL, your FAQ, your policy pages. If a tool needs you to hand-write fifty question and answer pairs before it can say anything, that is a finding in itself, and it is your afternoon.
3. Type the ten questions in the next section. Not "hello". Not "what do you do". Real customer questions, including the awkward ones.
4. Close the laptop and test on a phone. Most of your visitors are there. Check the bubble does not cover the buy button.
5. Come back at 10pm and ask another question. This is the test almost nobody runs and it is the one that decides most purchases.

Ten questions to type into any chat demo

Copy these, adapting the specifics. They are chosen because each one exposes a different failure.

1. A question answered on your website. Baseline. If it cannot do this, stop here.
2. The same question in different words. Ask about "shipping" then "delivery", or "cancel" then "refund". Scripted tools match keywords and fall over on the synonym.
3. A question with a number in it. "How much is it for 40 users?" Watch whether it reads your real pricing or invents something plausible. Invented prices are the most damaging failure in this category.
4. Something you deliberately do not publish. Ask about a discount you do not offer. The correct answer is "I do not have that information". A confident wrong answer here means you cannot trust the correct-looking ones either.
5. A two-part question. "Do you deliver to Ireland, and how long does it take?" Many tools answer the first half and drop the second.
6. A follow-up with no context. Ask a question, then just say "and for the larger one?" This tests whether it remembers the conversation or treats every message as new.
7. The same question in another language. Type it in French or German. If your visitors are international, this one question can decide the whole purchase.
8. A complaint, not a question. "This is the third time I have asked about my order." Watch whether it tries to answer or recognises it needs a person.
9. A request to speak to a human. Ask directly. Time how long the handover takes and check where the message ends up.
10. Something about a page you published yesterday. This tests how quickly new content becomes answerable. Some tools re-read on a schedule you cannot control.

Ten questions takes about eight minutes per tool. Run the same ten against every shortlisted product and write the answers down. The comparison will be obvious and it will not match the feature grid.

The three tests people skip

The after-hours test

Look at your analytics for when sessions actually happen. On most consumer sites a large share arrive in the evening and at weekends. Then open your own site at that hour with the trial widget live and ask something.

If the answer is a form saying the team is away, you have learned the most important thing about that product: it covers your working day, and your working day is not when your customers are browsing. That is a legitimate choice, but make it knowingly.

The speed test

Chat widgets are among the heaviest third-party scripts a site loads. Run your page through a speed test before you install the trial and again after, on mobile, and write both numbers down.

A widget that costs you half a second of load time on every page is not free, whatever the plan says. If two products score the same on your ten questions, this is a reasonable tiebreaker.

The removal test

Before the trial ends, remove it. Genuinely uninstall it and check nothing is left behind: no orphan script, no leftover styling, no cookie still being set. Doing this once during a trial saves an unpleasant afternoon later, and it tells you how the vendor treats people who leave.

What to ask if you do take the sales call

Sometimes you have to, particularly with outsourced providers where there is no self-serve option. In that case use the time for the things a trial cannot tell you.

1. What is the price at our exact size, on our billing frequency? Ask for one number in writing before the feature tour, not after.
2. What is metered, and what happens when we exceed it? Conversations, resolutions, contacts and seats are all meters in this market. Find yours.
3. Show me the product failing. Ask them to type a question it cannot answer. How they respond tells you a great deal.
4. What does the export look like if we leave? Transcripts are your record of what customers keep asking.
5. What is the notice period? Month to month exists in this market. Ask before you assume an annual term.

Score it, so the comparison survives the week

Three trials in five days blur together. By Friday you remember which interface you liked, which is the least useful thing you learned. Write the results down as you go, in a table like this one.

Two rules make this work. Score each tool on the same day you test it, not from memory later. And treat the invented-answer row as disqualifying rather than as points, because a tool that confidently makes things up will do it to a customer eventually, and every other score becomes irrelevant.

Demoing conversational AI, specifically

If the shortlist includes AI answering rather than only routing to humans, three extra tests apply. They take five minutes and they are where these products differ most.

Test what it does with a gap

Ask something your website genuinely does not cover. The correct behaviours are to say it does not know, to offer a person, or to point at the nearest relevant page. Anything confident and specific is a failure, and it is the failure that costs you customers rather than merely annoying them.

Test how it handles being corrected

Give it a wrong premise. "I read that you ship to Australia in two days, is that right?" A good system checks your actual content and corrects you politely. A weak one agrees, because agreeing is the easier pattern to produce.

Test freshness

Change something on a page, or publish a new one, then ask about it. How long until the answer updates? Some tools re-read on a schedule you cannot influence, which matters a great deal if your prices or stock move.

Ask the vendor the same question directly and compare the answer to what you observed. Where those two disagree, believe your own test.

Buying options after the demo

Once you have a winner, the purchase itself has choices worth knowing about before you click.

• Monthly first, annual later. Annual billing saves real money at most vendors, commonly 20% or more, but pay monthly for the first quarter. If the tool disappoints in week six, the discount was not worth it.
• Check what the meter is before the tier. Seats, conversations, contacts, resolutions and messages are all in use. Pick the meter that suits your shape, then pick the tier inside it.
• Ask what happens at the limit. Some plans stop serving, some upgrade you automatically, some bill overage. All three are survivable if you knew in advance.
• Keep the free tier account. Where a vendor has a genuine free plan rather than a trial, you can leave a dormant account in place and revisit later without starting over.

Common mistakes in a chat evaluation

What to do next

• Shortlist by pricing model first. Per agent, per workspace, per conversation, per resolution, per message. Rule out the models that do not suit your shape before you install anything.
• Start with the free tiers. tawk.to, Crisp, Tidio, Help Scout and Asyntai all let you begin without a call, so there is no reason a scheduled demo comes first.
• Write your ten questions once. They are your test suite. Reuse them on every tool and at every renewal.
• Decide the after-hours question before anything else. It eliminates half the market immediately, in one direction or the other, and everything after it is easier.

A demo is not a thing you watch. It is a thing you run, on your own site, with your own awkward questions. Half an hour spent that way will tell you more than three scheduled calls, and you can start it in the next two minutes without speaking to anyone.