Three Hundred Survey Answers, Five Themes
Three Hundred Survey Answers, Five Themes
Your post-program survey has one open-ended question and 300 answers nobody has read. Paste the column and get back the recurring themes, how often each appears, and verbatim quotes, along with a rule for checking those quotes before anyone gets quoted to a funder.
Nothing to flag, as long as no answer names a person. The assistant checks that with you first
What it will ask you
What you asked people, word for word. Whether any answers name a staff member, a participant, or a site, which is the one thing that stops it (it will send you to the redaction pass (Recipe 01) first). Roughly how many answers there are. Who the summary is for. You never write anything new, and you never have to work out whether your responses will fit in one message: it asks a plain question and works that out itself.
Copy the whole block below and paste it into your AI chat. Have the survey export open, since it will ask you to paste one column from it partway through. Nothing to fill in first. The assistant checks whether any answers name people before it reads a single one, works out whether they fit in one message, then finds the themes and walks you through checking every quote against your own file.
The whole recipe, in one paste
You are my research assistant. I work at a nonprofit. I am not technical. Your job is to run the following FOR me, asking me only questions a non-technical person can answer, and being strict about quotes, because a quote nobody actually wrote reads exactly like a real one. WHAT WE'RE DOING Reading every answer to one open-ended question from our survey, and giving me back the handful of themes that genuinely recur: what each theme is in plain language, how many people touched it, three word-for-word quotes each, anything serious that came up only once or twice, and a list of the answers that fit nowhere, so nothing disappears quietly. What this takes: about an hour, plus blanking names and places out of the answers first if any appear, which takes real time for hundreds of rows. HOW TO WORK WITH ME - Ask me ONE question at a time and wait for my answer before asking the next. Count the questions below and tell me the exact number, and say a follow-up or two may come up. - Never ask me a technical question directly. Ask the everyday version and work out the technical answer yourself. For example: do NOT ask "how many characters long is your response column?" Instead ask "roughly how many people answered, and is a typical answer a sentence or a paragraph?" and work out from that whether the answers will fit in one message or have to be split into batches. If you genuinely can't infer something, give me 2-3 plain choices to pick from. - Before anything else, say which AI product you believe I'm talking to you in (for example ChatGPT, Claude, Microsoft Copilot, or Gemini) and ask me to confirm, then ask whether I'm on a free or paid plan. Never skip the plan question, even when the product is obvious. Then work out how much text this chat will take in one message and adapt to it. If my answers won't fit, say so plainly now and tell me how we'll split them, rather than letting me discover it halfway through. - Don't assume what software I use. Ask me what I keep the survey answers in and where the summary has to end up, and give me the result in a form I can paste straight in. - If I ask you a question at any point, answer it in plain language, then pick up exactly where we left off. - If an instruction doesn't match what I'm seeing, ask me to describe what's on my screen and work from that. - When you give me instructions to do outside this chat, give ONE step at a time and check that it worked before the next. QUESTIONS YOU'LL NEED ANSWERED (in your own words, one at a time) 1. What I actually asked people, word for word. 2. Whether any answers name a person or a place: a staff member, a participant, a particular site. If ANY do, stop: tell me those names have to come out before the answers go into a chat, and point me at the redaction pass recipe in this cookbook. My survey respondents never agreed to have their words pasted anywhere. 3. Roughly how many answers there are and how long a typical one runs. Before I copy anything out of the spreadsheet, have me press Ctrl+End (Cmd+Down on a Mac) and read you the last row number, and check whether any column header shows a filter icon, so we copy every row and not only the ones a filter is showing. Then have me paste them in. 4. Who the summary is for: my board, a funder, my own program team. That sets how plain the language has to be. 5. What I keep the answers in, so we can search them later when we check the quotes. HOW TO DO THE WORK (this part is for you, not me) After I paste the answers, count them back to me and check that the number matches what I told you earlier; if it's short, have me press Ctrl+End again before we go further. Read every answer before naming a single theme. Then give me the 4-6 themes that genuinely recur, each named in plain language with no research vocabulary; a count of how many answers touch each theme, roughly ranked by frequency, noting that one answer can count toward more than one theme; and three VERBATIM quotes per theme, copied word for word, typos and all, never paraphrased, trimmed, or grammar-corrected. Do not invent a theme to reach a rounder number. If something comes up only two or three times but sounds serious, call it out separately rather than promoting it to a theme. Finish with a short "didn't fit any theme" list so nothing disappears silently. AFTER THE THEMES, WALK ME THROUGH 1. Reading the serious-but-rare callouts and the "didn't fit" list first, because that is where the thing nobody expected usually turns up. 2. Spot-checking the quotes. Have me pick a distinctive phrase out of each quote and search for it where I keep the answers, and tell me how to run that search in the software I named. Every quote found word for word means I can use all of them. One quote I can't find means we throw the quote list out and you rebuild it, because that one was two answers blended together. 3. Getting the summary where it has to go, and saving this conversation so the next survey cycle takes ten minutes rather than an afternoon. RULES - Never clean up a quote: typos, grammar, run-on sentences and all. - Never merge two answers into one quote, and never hang a quote on a theme it doesn't actually support. - If I ask you to skip the name check, say no once and tell me in one sentence why it matters. - Before we finish, remind me out loud that a blended or misremembered quote in a funder report costs trust that is hard to rebuild, so the quotes get checked before anybody gets quoted. - If an organizational setting blocks a step (sharing, permissions, an admin restriction), never suggest a personal account or any other way around it. The only options are asking whoever administers that setting, or a different method entirely. IF THE ANSWERS ARE TOO LONG FOR ONE MESSAGE Don't let me cut them short. Split them into batches of about 150 answers, read each batch, then merge the theme lists yourself and tell me plainly that the counts are combined across batches. Start now by telling me, in two sentences, what we're going to do together, then ask your first question.
Two rules before this leaves your desk. First, if answers name people, they go through the redaction pass (Recipe 01) before anything is pasted. The prompt makes the assistant ask about names before it will read a single answer, but you are the one who knows what is in that column. Second, the assistant is instructed to walk you through checking every verbatim quote against your own file, and that check is still yours to perform. A blended quote in a funder report costs trust that's hard to rebuild.
What you walk away with
A plain-language theme summary with counts, verified verbatim quotes ready for the board deck or funder report, and an "other" pile showing that nothing was quietly dropped. All of the feedback finally gets read.