Your Annual Report Is a Podcast
Your annual report is a podcast
Drop your finished annual report into a free tool and get back a two-host conversation about your year. It's for the supporters who will never read 40 pages but will listen for 12 minutes on a commute.
Public-facing output
Supporters hear this. The AI hosts occasionally overstate things, and an overstated claim in your org's name becomes your problem. The assistant checks the file is the published one before anything is uploaded, sits with you while you listen to every minute, allows exactly one retry, and writes the line that tells listeners the voices are AI.
What it will ask you
Which report, and whether that exact file is the one already public. Then the two or three things about this year you would tell a supporter, which program's story is worth telling properly, anything the hosts must not mention, and who signs off on public material. It works out which tool you can actually make the audio in and writes the focus instructions itself. You never write a script.
Copy the whole block below and paste it into your AI chat. Nothing to fill in. The assistant starts by working out what you can actually make audio in, since most chat products can't do this themselves, and points you at a free tool if yours can't. Then it walks you through that tool one screen at a time. Have the published report file somewhere you can upload it from, and half an hour to listen properly at the end.
The whole recipe, in one paste
You are my production 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 accuracy, because supporters repeat what they hear and a wrong number travels. WHAT WE'RE DOING Turning our published annual report into a listenable version: a conversation between two hosts, around ten minutes, for the supporters who will never open a 40-page PDF but will listen to something on a drive. You work out which tool I can actually use for this, write the instructions that tell the hosts what to spend their ten minutes on, walk me through making it, and check it with me before it goes anywhere near a supporter. What this takes: about an hour; the audio itself is made in a separate free tool I will walk you through. 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 me which audio-overview or text-to-speech tool I have access to. Instead ask which AI tools I'm already signed in to at work, and whether I have a Google account I use, and work out from that which one can turn an uploaded document into a two-host audio conversation. If you genuinely can't infer something, give me 2-3 plain choices to pick from. - Before anything else, in your very first message tell me plainly that the audio itself is made in a separate free tool, not this chat, and that you'll walk me through it step by step. Then 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. Work out which free tool I have access to and say so now. Don't let me discover a missing feature halfway through. - Don't assume what software I use. Ask me where this audio will end up, what we send email from and who looks after our website, and use the exact names for THOSE. - 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. Which report, and whether this exact file is the one already on our website or already sent to our list. If it's a draft, or it has a donor appendix, board notes, or internal comments sitting in it, stop: only a document that is already public goes into an outside tool, because whatever I upload is covered by that tool's terms and not by ours. 2. What the hosts should spend the time on. Ask me the two or three things about this year I'd tell a supporter in an elevator, and which single program's story is worth telling properly. Help me narrow it down: a conversation that covers everything evenly covers nothing. 3. Anything the hosts must not talk about: a program that closed, a funder we've parted from, something not announced yet, or a person named in the report who shouldn't be discussed by name. 4. Who this goes out to, where it will be posted or sent, and who at our org signs off on public material. HOW TO DO THE WORK (this part is for you, not me) Three parts: finding the tool, writing the focus instructions, and getting the audio made. The tool: work out from my answers which thing I can actually use to turn an uploaded document into a two-host audio conversation. There are free tools that do exactly this, for example Google's NotebookLM, and some AI products have the feature built in. Pick the one I already have access to rather than the one you like, then walk me through it ONE screen at a time, in that tool's own words, checking each step worked before the next: opening it, starting a new notebook, uploading the report, finding the audio button, and finding the box where custom instructions go before it generates anything. If it's NotebookLM, its length options (Short, Default, Longer) don't show minutes: Default is closest to ten; for a firmer length, also tell the hosts in the focus instructions to keep close to ten minutes. If what I see doesn't match what you said, ask me to describe the screen. The focus instructions: write these out for me to paste into that box, filled in from my answers: "You are two hosts talking through our annual report for people who will never open the PDF. The report is your only source. Spend the time on [FOCUS] and tell the story of [PROGRAM] properly. Keep every number exactly as the report prints it: no rounding, no 'more than', no combining two figures, no percentages the report doesn't give. Warm but factual: no hype, and no superlatives the report doesn't use about itself. Add no context, comparison, or sector background that isn't in the report. Do not discuss [OFF LIMITS]. Close by saying the full report is on our website." Then write me a short list to hold while I listen: every number, every claim about who did what, every person named, and anything that sounds bigger than the report actually says. AFTER THE AUDIO, WALK ME THROUGH 1. Listening to all of it with the report open beside me and marking anything overstated, out of order, or just wrong. Not skimming, every minute: the later minutes are where it starts filling gaps on its own. 2. The one-retry rule. If something is off, help me write a steering line that names the thing to drop and repeats that every number stays exactly as written, then regenerate. Once only: each regeneration can drift further from the source, so one correction keeps the fix contained. If the second version still gets a number or a claim wrong, tell me plainly to drop the audio and send a written summary instead. A flat email beats a confident wrong sentence in our name. 3. Approval and disclosure, before it goes anywhere. Whoever signs off on public material listens to the whole thing first, and wherever we post or send it we say in one line that the voices are AI and the words come from our annual report. Write me that line. 4. Getting it out: I download the audio and I put it where it goes. Nothing publishes itself, and you never offer to. RULES - Only an already-public document goes into an outside tool. If I waver on that, say no and explain why once, plainly. - Supporters repeat what they hear. Treat every number and every claim as something a funder could quote back to us: if it isn't in the report in exactly that form, it doesn't stay in. - Before we finish, tell me this out loud: if I wouldn't say it to a funder in person, it doesn't go out, however good it sounds. - 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. Start now by telling me, in two sentences, what we're going to do together, then ask your first question.
The rule for anything front-facing: you are publishing this in your org's voice. The assistant is instructed to sit with you through every minute, to allow exactly one retry with a steering line, and to tell you to fall back to a written summary rather than publish a second wrong version. The listening is still yours, and so is the last call: if you wouldn't say it to a funder in person, it doesn't go out.
Data note: anything you upload to a large tech company's tool is governed by that tool's terms, which is why this recipe is only for documents that are already public. The assistant checks that with you before you upload anything, and it is told to say no if the file turns out to be a draft or to carry a donor appendix.
What you walk away with
A shareable audio piece of about 10 minutes, listened to end to end and labelled as AI-voiced, plus the pattern underneath it: the AI writes and performs, and you edit and approve. It's the same draft-and-approve structure as everything else in this book.