Verified against Gemini · 2026-07-27
Turn a photo of a whiteboard brainstorm into an organized action plan
A multimodal prompt that reads the handwriting, arrows, and groupings in a photographed whiteboard and turns the visual structure itself — not just the words — into an organized, de-duplicated action plan.
The prompt
Ready to copy — highlighted parts are example details you can swap.
Look at the photo I've attached of a 45-minute product roadmap brainstorm with the design and engineering leads. Read the actual spatial layout, not just the words — arrows, circles, groupings, and crossed-out items all carry meaning here. Don't flatten a structured board into a random flat list; that throws away exactly the information that made it worth photographing instead of just listing the ideas from memory. INSTRUCTIONS 1. Transcribe everything legible first, noting anything you genuinely can't read as [illegible] rather than guessing a plausible-sounding word to fill the gap. 2. Reconstruct the groupings the way they're actually drawn on the board — circled clusters, boxes, arrows connecting one idea to another. Tell me what connects to what, and don't invent a connection that isn't actually drawn just because two ideas seem related. 3. Treat anything crossed out or struck through as rejected, not as a live idea — list it separately under "Discarded," and don't quietly drop it from the notes entirely, since knowing what was considered and rejected is often as useful as knowing what survived. 4. Merge near-duplicate ideas written in different handwriting or in different spots on the board into one item, but tell me explicitly when you merged something and what the original separate items were, so I can un-merge it if you merged wrong. 5. Turn the surviving ideas into an action plan grouped by: this quarter vs. next quarter vs. someday. Each action gets an owner placeholder and a rough priority (high / medium / low). Context that isn't visible on the board but matters here: we already committed to shipping the mobile redesign this quarter, so anything competing with that engineering time should be flagged If the photo is angled, glare-affected, or cuts off part of the board, say which part is affected before presenting the plan, so I know whether to take another photo. HANDWRITING FROM MULTIPLE PEOPLE If different handwriting styles are visible, note when an idea appears to have been added later or by a different person than the surrounding cluster — a note squeezed into a margin in different handwriting is often a late addition or a dissenting comment, and treating it as part of the original cluster it was squeezed next to loses that distinction. NUMBERS AND ESTIMATES ON THE BOARD If any numbers, dates, or estimates are written on the board (a rough timeline, a headcount, a budget figure), carry them into the plan exactly as written rather than rounding or reinterpreting them — a number scrawled on a whiteboard during a brainstorm is often a deliberately rough placeholder, and representing it with more precision or confidence than it actually had misrepresents what was decided in the room. ARROWS THAT POINT BUT DON'T CLEARLY CONNECT If an arrow's start or end point is ambiguous — it could plausibly connect to either of two nearby items — say so rather than picking the more sensible-looking interpretation and presenting it as certain; a wrong guess about which two ideas are actually linked can misrepresent the group's actual reasoning.
Customize
Optional — swap in your own details for the highlighted parts above.
Why this works
Gemini's multimodal image encoder reads spatial relationships directly from the pixels — what's circled, what an arrow connects, what's boxed together — rather than running OCR and discarding the layout, which is why the prompt explicitly demands groupings be reconstructed 'the way they're drawn' instead of flattened into a list; a plain OCR-then-summarize approach loses exactly this structure, since OCR by design only extracts text, not the drawn relationships between text blocks. The [illegible] instruction targets a specific and common failure with handwriting photos: a blurry or ambiguous word gets confidently transcribed as the closest plausible-sounding word rather than flagged, which quietly corrupts an action item into something nobody in the room actually wrote or said. Requiring crossed-out items to be filed as 'Discarded' rather than dropped or treated as live prevents rejected ideas from silently resurfacing in the final plan — a real risk since a merged or reorganized list has no visual strikethrough left to signal an idea was rejected the way the original board did. The explicit merge-disclosure rule matters because de-duplication is a judgment call, not a fact, and a model that silently merges two genuinely distinct ideas because they use similar words removes the one signal (the original separate wording) a human would need to catch and correct that mistake. Flagging handwriting that appears to be a later addition or a different contributor's note leans on the same spatial/visual reading the grouping instruction does — a margin note squeezed in afterward is visually distinguishable from the original cluster by spacing and often by pen pressure or size, and a model that flattens everything on the board into one undifferentiated pass loses the chance to notice that a late addition might represent a dissent or afterthought rather than a idea the room settled on together in the moment. Preserving numbers and estimates exactly as written, rather than reinterpreting them with more apparent precision, matters because a number scrawled during a live brainstorm is frequently a deliberately rough placeholder the group intends to revisit — presenting 'roughly 15 people?' as a firm headcount of 15 in a downstream action plan quietly launders a guess into a commitment nobody in the room actually made.
Verified against
Gemini Gemini 3 Pro · 2026-07-27
Changelog
- 2026-07-27 — Initial publish, verified against Gemini 3 Pro on a photographed product-roadmap whiteboard session.
Need this built into your business?
If a prompt isn't enough — what Scult builds, built and maintained for you — that's Scult's day job.
EXPLORE WHAT SCULT BUILDS
