Survey results summary

Answers "so what did the survey say" with figures and a conclusion instead of a raw export

Free agent skill for Claude Code, Codex, Cursor and Gemini CLI. It works through the SurveyNinja MCP server, with your permissions.

Category: Understand results
Tags: report, open answers, comparison
Works with: Claude Code, Codex, Cursor, Gemini CLI
MCP tools: 9, 1 of them change data
Version: 1.1, October 20, 2026
Made by: SurveyNinja
Price: Free
See also: AI survey analysis Survey reports
Skill file: SKILL.md

What it does

Conclusion first, numbers second. Instead of "question 3 has this distribution" you read "two thirds are unhappy with turnaround" and then see the breakdown.
Every percentage has a base. You always know whether a share is taken from all completions or only from finished ones.
A file only when you ask. In chat you get numbers. An Excel export appears only on request.

Example

You type

What did the survey show? Give me the short version

What the assistant brings back

412
Started: 318
Finished77%: 4.1
Average scoreout of 5: 186
Left a contact45% of those who started

Very satisfied 134 (42%)
Somewhat satisfied 92 (29%)
Neutral 48 (15%)
Somewhat dissatisfied 29 (9%)
Very dissatisfied 15 (5%)

The headline is turnaround, not price

In the open answers 41 of 96 respondents mention waiting time and 12 mention price. The percentages above are taken from 318 finished completions; abandoned ones are excluded.

The numbers above are an example. The digest is built from your real responses.

How it works

Starts from the summary and the reportDistributions are already calculated in SurveyNinja. Raw completions are needed only for a specific person or to quote wording.
Checks whether the data is completeIf many completions were abandoned, they are counted separately instead of being folded into one number with the finished ones.
Slices when a condition is named"September only", "only people who picked the premium option", "only from the link I emailed". The skill filters by dates, answers, tags or link sets instead of counting by eye.
Leads with the conclusionYou read "two thirds are unhappy with turnaround" first, and then the base that percentage is taken from.

Try also

Show results for September only, and only for people who came from the link I emailed
Summarize the survey and tell me what deserves attention

What it will not do

Do not export a file unasked
Do not recompute by hand
Do not mix completed and abandoned
Do not suggest a plan upgrade or lead to payment

The skill works with your permissions. Tools that change data are marked, so you check what the assistant is about to do before you confirm. To cut access, remove the connector in the assistant or delete the API key.

In short

Where does the skill start?

With the survey summary and the per-question report, where distributions and percentages are already calculated. It reads individual responses only when you ask about a specific one or it needs a quote.

Can I get results only for September or for one channel?

Yes. Name the condition and the skill builds a slice by dates, by answers to a question, by tags or by a set of unique links. Date, answer and tag slices are not on every plan. To separate channels, each one needs its own set of unique links; a label in the link is counted from the responses that carry it.

How does it handle incomplete responses?

Separately. It first checks how many people finished, and if many dropped out it says so. Finished and abandoned are never mixed in one figure without a note.

What about open-ended answers?

The skill does not retell them one by one. It groups them into themes and gives a couple of typical quotes.

Will I get conclusions even with few responses?

Not as facts. With around ten responses the skill says plainly that the data is too thin to conclude anything.

Does the skill change anything in my survey?

No. It only reads the summary, the report and the responses, and it makes an export file only if you ask for one. The download link works for one hour.

The full skill

The SKILL.md file the assistant reads, exactly as it is in the archive.

Show the full text

What the survey showed

The answer is a few numbers with an explanation of what they mean. A dump of raw responses is not an answer.

Reply to the person in the language they write in.

Which tool for what

This is where the wrong tool gets picked most often:

What you need	Tool	What it returns
The funnel	`get_quiz_summary`	visits, completed, abandoned, completion rate, screened out, average time, by day, devices, countries
Answers per question	`get_quiz_report`	distributions for every question, completed responses only
A slice by answers, tags or a set of unique links	`generate_filtered_report`	the same report over the selected responses
Individual submissions	`get_quiz_answers`	raw responses one by one
Free text of one question	`get_quiz_report_inputs`	text answers, 50 per page
A file for a human	`export_answers_xlsx` and the other `export_*` tools	a download link

Start with the summary and the report, not with raw answers. Raw answers are for a specific submission or for quoting examples.

Order

get_quiz_summary: how many visited, how many finished, how many abandoned. If many abandoned, say so and treat them separately.
get_quiz_report: the distributions. It counts completed responses only, whatever is_complete says. Unfinished ones are read one by one with get_quiz_answers with is_complete false, if the plan keeps them; if it does not, that call quietly returns completed ones, so check is_complete on each response.
If a condition was named, such as "September only" or "only people who chose the premium option", build a slice: dates in get_quiz_report, answers and tags in generate_filtered_report. Dates, answers and tags apply only when the plan includes results filters; otherwise the report covers everything and does not say so. Check get_workspace_tariff (the workspace_id comes from get_workspace_list) or compare the totals with the summary. For "only from the link I sent by email", check how that link was marked: a set of unique links filters directly, on any plan. A link label such as utm_source is in the extra_fields of each response from get_quiz_answers: count the responses with that value there. No tool builds the whole report for one label value; in the app that is the Hidden Variables filter in Results.
Open answers: get_quiz_report_inputs with the report_uuid from the report and the question's widget_id and widget_hash from get_quiz_structure. It returns 50 answers per page: page with offset to read them all. The texts follow the report filters saved in the app, not the slice built here.
Export a file only if a file was asked for. The link lives for one hour.

How to answer

Conclusion first, numbers second. Not "question 3 has this distribution" but "two thirds are unhappy with turnaround times, here is the breakdown".
Always name the base for a percentage: all submissions or completed ones.
Do not present conclusions from ten responses as fact.
Do not retell free text one by one. Group it into themes with a couple of representative quotes.

What not to do

Do not export a file unasked. Chat needs numbers, not a download link.
Do not recompute by hand what the summary or the report already provides.
Do not mix completed and abandoned in one figure without saying so.
Do not suggest a plan upgrade or lead to payment. If a limit is hit or a feature is not on the plan, state the fact and stop.

More skills

All skills

Connect SurveyNinja to your assistant

The MCP server works on every plan, Free included. Sign up, allow access and ask your assistant to build the first survey.

Start free About the MCP server