Pesquisa a partir de um briefing

Building a survey from a brief

Someone describes the job in words: "I need a survey for our repair shop customers, to rate the work and collect contacts for a follow-up visit". Turn that into a working survey in their account, not a list of questions in chat.

Reply to the person in the language they write in. Write the survey texts in the language of their audience, not the language of this instruction.

1. Ask three things if they were not given

Who is being asked, what decision the results serve, where the link will live. No more than three questions: everything else is faster to show as a draft and fix from feedback.

Do not ask about question types, styling or order. Choosing them is the assistant's job.

2. Check for something ready-made

search_quiz_templates by topic. The catalog has English templates but no Spanish or Portuguese ones: for those audiences search with lang set to en and translate. If a suitable template exists, offer to start from it with create_quiz_from_template and edit from there: it is faster and usually better than building from scratch. The choice stays with the person.

3. Create the survey and build the questions

create_quiz, then update_quiz_widgets.

A survey is created inside a folder, so start with get_folder_list and pick the one to use. If there are several and the person did not say which, ask instead of choosing for them.

Choosing types is the core of the work and where mistakes happen:

What you need	Type	Why not otherwise
Quality rating	rating or scale	not input with a number: free text cannot be summarized
Likelihood to recommend	nps	its own 0–10 scale and its own analytics
One choice from a short list	choice	not input: free text cannot be grouped in a report
Several choices	choice with multiple selection
One choice from a long list	dropdown	a long list of buttons is hard to scan
Name	fio	its own fields; switch the middle-name field off for most audiences
Phone, email	phone, email	format checks; otherwise half the contacts are unusable
Date and time	datetime	not text: otherwise it cannot be filtered or sorted
Several ratings on one set of criteria	matrix	instead of ten separate questions
File or photo	file
Numeric answer	number
Yes or no	yesno
Put in order	ranking

Start the survey with a welcome screen. The last widget is always submit; add a finish thank-you screen before it so the respondent sees what happens next instead of a bare confirmation.

Make required only the contacts the survey exists for. Every extra required question costs responses.

Set the phone question's default country to the audience's country, not a guess.

The full list of types and their fields is in references/widget-types.md.

4. Texts and styling

Headings and the thank-you text are fields of the welcome and finish widgets: write them with update_quiz_widgets. Button labels and other interface strings go through update_quiz_texts, by their code from get_quiz_texts. Write in the plain language of the audience.

get_theme_list and apply_quiz_theme: take an existing theme. Create a new one only if brand colors are explicitly requested.

5. Branching

update_quiz_logic, if the brief implies branches: "unhappy customers get a why question", "did not use the service, so skip the quality block".

The order is strict: get_quiz_structure first, then edit, then write. Conditions reference questions by identifier, and logic built blind leads respondents into a dead end.

6. Show the draft, then publish

publish_quiz only after the person agrees. Until it is published the survey does not open by its link, and that is the right state for review.

After publishing, hand over the link and briefly list what came out: how many questions, which are required, whether there is branching.

What not to do

• Do not publish unasked. A published survey can already be taken, and edits after the first responses break comparability.
• Do not invent filler questions. Long surveys are abandoned.
• Do not make everything required.
• Do not suggest a plan upgrade or lead to payment. If a limit is reached or a question type is not on the plan, state the fact and stop.
• Do not delete or rewrite existing surveys when asked to make a new one.

If the brief is large

For long briefs there is generate_ai_quiz: the whole build runs in the background. It returns the id of a survey that stays empty until the build is done, not a finished survey. Call generate_ai_quiz again with that quiz_id to see whether it is pending or ready, show the result, then edit with the usual tools.