Skills / Run a survey
Run a survey
Run a real survey and report what it says. Questions that don't lead, a consent line, honest analysis of what came back. Use when they say "I want to survey people", need opinions for a decision, or claim "everyone thinks X".
Use this skill. Nothing to install.
Don't have an agent? · Raw file: skills/run-a-survey.md
Watch the first two minutes
This is how a survey session goes when someone arrives with the conclusion already written. Click through it.
Scripted example of a real session.
What if the answer comes back boring, or against me?
Then the survey worked. The discipline this skill teaches is accepting whatever comes back, including a result that contradicts what you hoped, and writing it up plainly. Questions are attacked before sending so they cannot lead, scales get a genuine middle and a "not sure", and personal details are only collected if the decision needs them. If you like asking groups of people things, these projects run on exactly that:
What you end up with
Not just responses in a sheet: a one-page finding, delivered to whoever owns the decision. Here is the whole shape of one.
Preferred day Sat 9 Sun 3 keep Wed 2
Would attend more often if moved:
yes 8 same 4 not sure 2
- Response rate is part of the result. 12 answers from 20 teammates means something different from 12 from a 400-person server, so the finding names who was reachable and who replied.
- Distributions before averages. A split crowd and a lukewarm crowd can share a mean, so every counted question is reported as its spread (COUNTIF per option).
- Claims stay inside the data. "9 of 14 who answered preferred Saturdays" survives scrutiny; "the team wants Saturdays" does not. The limits section is written with the same care as the results.
- Delivered, with a next step. The summary goes to the person the decision belongs to, and the suggestion is checkable against future data.
Questions people actually ask
How many responses do I need?
Enough from the people the decision is about, honestly counted. The skill has you survey people you can actually reach and report n and response rate as part of the finding, so 14 of 20 teammates is a solid basis for a team decision.
Should it be anonymous?
By default, yes, and the consent line says so. You also collect only the personal details the decision needs: if age or name would not change the outcome, the question gets cut.
How do I stop my questions from leading?
Draft, then attack each draft: one thing per question, neutral framing, balanced scales with a genuine middle and a "not sure". Then pilot on two people and reword anything they hesitate over.
What tool does it use?
Google Forms to collect and a spreadsheet to count. Nothing paid, nothing installed.
What if barely anyone responds?
You send once, remind once, and then the low response rate becomes part of the finding rather than a reason to quietly resend until the numbers look better. Two of twenty answering tells you something real about interest.
Where to go from here
Responses in hand? Make more of them:
Like collecting opinions? These projects feed on them:
Curious? Read the full skill — the exact instructions your agent gets
--- name: run-a-survey category: data description: Run a real survey and report what it says. Questions that don't lead, a consent line, honest analysis of what came back. Use when they say "I want to survey people", need opinions for a decision, or claim "everyone thinks X". --- # run-a-survey Help someone run a real survey: real questions, sent to real people, analyzed honestly. The respondents are people they can actually reach — classmates, teammates, a Discord, customers. Use Google Forms for collection and a sheet for analysis. The finished product is a short written finding, and the discipline is accepting whatever answer comes back, including a boring one. ## Decide what it is for One decision or one question the survey should settle: what should the club do next term, would anyone pay for this, when can the team practice. Write it down. Every question on the survey must serve it; a question whose answer would change nothing gets cut. Aim for five to eight questions — completion drops with every extra one. ## Write questions that do not lead Draft together, then attack each draft question: - No loaded framing. "How much do you love the new schedule?" assumes the answer. Neutral: "How do you feel about the new schedule?" with a balanced scale. - One thing per question. "Was it fun and useful?" is two questions wearing one checkbox. - Closed questions (choices, 1–5 scales) for anything you will count; one open "anything else?" at the end for what you did not think to ask. - Balanced scales with a genuine middle, and "not sure" where honest people might need it. Forcing an opinion manufactures data. - Ask only the personal details the goal needs. If age or name does not change the decision, do not collect it. Anonymous by default. Put a consent line at the top: what this is for, who sees results, that answers are anonymous (if true) and participation is optional. Two sentences, plain words. Then pilot it on two people and watch them fill it in — every hesitation is a question that needs rewording. Do not collect sensitive health, identity, financial, disciplinary, or precise location data. This skill is for low-risk informal feedback, not formal human-subject research, medical screening, or a high-stakes decision about an individual. Follow school, workplace, community, and guardian rules, especially when respondents are minors. Escalate formal research to the responsible institutional process. ## Distribute and analyze Send it where the real people are, once, with one reminder a few days later. Note roughly how many people saw the link, because response rate is part of the result: 12 answers from 20 teammates means something different from 12 from a 400-person server. When responses land, pull them into a sheet and count: per question, the distribution (COUNTIF), not just the average — a split crowd and a lukewarm crowd can share a mean. Read every open-text answer; quote short ones anonymously if they sharpen the finding. ## Write the honest summary One page or less: the question, who was asked, how many answered, what the numbers show, and what they cannot show. Keep claims inside the data: "9 of 14 who answered preferred Saturdays" — never "the team wants Saturdays" — and say who was missing (people who skip surveys may differ from people who answer). If the result contradicts what they hoped, the write-up says so plainly; that is the survey working. Deliver the summary to whoever the decision belongs to. ## Done - A form with a consent line, neutral questions, piloted before send - No unnecessary sensitive data collected; applicable institutional and minor consent rules checked - Real responses in a sheet, counted as distributions - A one-page summary with n, response rate, findings, and limits - The summary actually sent to the person or group deciding Then: chart-the-truth for the one chart the summary deserves.