Quizzes have a poor reputation among people who take them and a good one among people who run them, which tells you something. They convert well compared with a plain form, they generate email addresses, and they are recommended constantly.
They also, in the common implementation, ask eight questions that have nothing to do with the answer and then demand an email address before showing a result that could have been written without the quiz. People finish them, give a disposable address, and never open anything.
The difference between the two versions is not the mechanics. It is whether the result is genuinely worth the eight questions, and that is a design decision made before anything gets built.
Why they work at all
A quiz converts better than a form because it inverts the transaction. A form asks for something and promises a reply later. A quiz offers something and asks for payment at the end, by which point the person has invested eight answers and wants to see the result.
There is also a genuine appeal to being told something about yourself or your situation. Which of these four types are you is an old format because it works, and it works because people are interested in themselves in a way they are not interested in your services.
None of this is manipulative in itself. It becomes manipulative when the promise of a useful result is not kept, which is the failure mode that has given the format its reputation.
A form asks for something and promises a reply. A quiz offers something and collects at the end.
Design the result first
The single decision that determines whether a quiz is any good is what the person gets at the end, and it should be settled before a single question is written.
Work backwards. If the result is a recommendation of which of your three packages fits, then the questions are whatever distinguishes those three, and there will be about four of them. If the result is a score against a benchmark, the questions are whatever the benchmark measures.
Quizzes that are designed forwards, starting from what would be interesting to ask, end up with questions that do not affect the outcome. People notice this, because the result reads as generic, and the generic result is what teaches them that quizzes are not worth finishing.
Every question should change the answer
The test to apply to each question is whether two people who differ only on this question could get different results. If not, the question is there for engagement rather than for the outcome, and it is costing you completions.
This is worth being strict about, because engagement questions accumulate. They come from a reasonable instinct, which is that a two-question quiz feels thin. The fix for that is a better result rather than more questions, and a sharp two-question quiz with a genuinely useful answer outperforms a padded eight-question one.
There is one legitimate exception, which is a question whose purpose is qualification rather than scoring. Keep those to one, put them near the end, and be honest with yourself that it is there for you rather than for them.
Scoring that survives contact with reality
Two approaches cover nearly every useful quiz, and the choice between them determines how much work the thing is to maintain.
Points-based scoring adds up numeric values and sorts the total into bands. It is simple, it handles a lot of questions gracefully, and it produces results that shade into each other, which suits anything measuring a degree of something. The risk is that the bands are arbitrary, so they need testing against real cases rather than picked because thirds seemed reasonable.
Category-based scoring counts which type of answer somebody picked most and assigns them to that category. It suits which-of-these-are-you formats and produces cleaner narratives. The risk is ties and near-ties, which need a defined tiebreak or the result feels arbitrary to exactly the people it describes least well.
Test the edges before publishing
Scoring bugs are invisible in the editor and obvious to the person who receives a wrong result. Walk the extremes deliberately.
Answer everything at the lowest value and check the result reads sensibly rather than insultingly. Answer everything at the highest and check the same. Then construct the mixed case that lands exactly on a boundary, because boundaries are where the off-by-one errors live and where two similar people get very different answers.
If your quiz has optional questions, check what happens when they are skipped, because a scoring system that treats an unanswered question as zero will push everybody who skips towards one end of the scale.
When to ask for the email
This is the contested decision and the honest answer is that it depends on what the result is worth.
Asking before the result maximises addresses and minimises quality, because a share of people will give something disposable and a share will abandon. Asking after gives fewer addresses and much better ones, because the person has seen the value and is choosing to continue rather than paying a toll.
The version that works best in practice is to show the headline result immediately and offer the detail by email. You have kept the promise, the person knows the result is real, and what they are giving an address for is the fuller version rather than for access to something they already earned.
The result has to be specific
Whatever the format, the result fails if it could have been written before the quiz existed. You are a strategic planner, here are three general tips, is the shape that has trained everybody to distrust the format.
Specificity usually comes from referencing the answers. Because you said you handle this in-house and your volume is around this, the thing that will bite first is this. That requires writing more result variants, which is the actual work in building a good quiz and the part most people skip.
A useful minimum is that each result should contain at least one sentence that would be wrong for somebody in a different band. If every sentence would apply to everybody, the quiz has not done anything.
Length and the abandonment curve
Quizzes hold attention better than forms, and not indefinitely. The drop-off is gentle for the first four or five questions and then steepens noticeably, which is a different shape from a form's cliff at the start.
Somewhere between five and eight questions is where most quizzes should land. Below five it is hard to produce a result specific enough to feel earned. Above eight, the abandonment starts costing more than the extra precision is worth.
A progress indicator helps more here than on an ordinary form, because the promise of a result makes people willing to continue if they can see how far it is. Without one, every question could be the last or the tenth, and uncertainty is what makes people stop.
What to do with the answers afterwards
A quiz produces far richer data than a contact form and most businesses throw all of it away except the email address, which is the least interesting part.
The answers tell you what the person's situation is, in structured form, which means the follow-up can be specific in a way a generic sequence cannot. Somebody who scored low on one dimension and high on another has a describable problem, and an email that names it will outperform anything generic by a wide margin.
It also aggregates. A few hundred quiz responses is a survey of your market that you did not have to run, and the distribution of results is frequently more informative than any individual one.
The honesty test
There is a simple check that separates the quizzes people recommend from the ones they resent. Would the result be useful to somebody who never bought anything from you?
If yes, the quiz is a genuine piece of value and the email address is a fair exchange. If the result is essentially an advertisement dressed as a diagnosis, people will work that out within two sentences and the address you collected is worth nothing.
This is not only an ethical point. It is the practical difference between a list of addresses that open your emails and a list that does not, and the second kind actively damages your sending reputation.
Where quizzes are the wrong tool
Not every business should build one. If your customers arrive knowing exactly what they want, a quiz is an obstacle between them and buying it, and the conversion numbers people quote do not apply.
They work best where the customer has a problem and is not sure which solution fits, which covers a lot of services and very little retail. They also work where there is a genuine diagnostic to perform, which is why they are common in health, finance and anything consultative.
If you cannot name what somebody would learn, do not build one. A plain form with a good first question will serve you better and takes an hour instead of a week.
Do not fake the scarcity
A specific pattern worth avoiding, because it is common and it works briefly. Quizzes are frequently dressed with artificial urgency: a countdown on the result, a claim that only a few slots remain, a score presented as worryingly low.
This converts in the short term and it poisons the thing you were building, which is a relationship with somebody who trusted the result. People who feel manipulated by a diagnosis do not become customers, and they are considerably more likely to tell somebody about it than a satisfied respondent is.
If the result genuinely is urgent, say so plainly and explain why. If it is not, the quiz is doing enough work already, and a result that stands on its own converts better than one propped up with theatre.
Rerunning it, and what changes
A quiz is not a one-off asset. The distribution of results is data about your market, and it moves, which means the quiz should be reviewed rather than left running for three years.
The signal to look for is a band nobody lands in, or one that everybody lands in. Either means the scoring is mis-calibrated against the people who actually take it, and the fix is to move the boundaries rather than to change the questions. A quiz where four out of five people get the same result has stopped diagnosing anything.
It is also worth reading a sample of completed responses alongside what those people went on to do. If the band that was meant to indicate a good fit does not convert any better than the others, the quiz is measuring something that is not related to buying, which is worth knowing before you build a second one.
A reasonable first version
If you are building one, this shape is hard to get wrong. Decide on three or four results you can describe specifically. Write five questions that distinguish between them and nothing else.
Show the headline result immediately, on the page, with two or three sentences that could only apply to that band. Offer the detailed version by email, unticked and optional. Keep the quiz answers alongside the address so the follow-up can reference them.
Then take it yourself three times, answering as three different kinds of customer, and check that you get three answers you would be comfortable defending. If any of them feels like it could have been anybody's, the questions are not doing enough work yet.