The Idukki Prompt Pack

These run against a folder of customer photos and produce a structured tag JSON. Useful when you have eight thousand images and one merchandiser.

• Tag a customer photo for ecommerce search

  VisionGPT-4oClaude Sonnet

  You are tagging a customer-uploaded product photo for an ecommerce store's UGC gallery. Return a single JSON object with these fields:
  - subject: the main product visible (e.g. "running shoes", "ceramic mug")
  - colours: an array of up to three hex codes representing the dominant colours
  - scene: one of [indoor-home, outdoor-street, studio, kitchen, bathroom, gym, other]
  - occasion: a single tag from [everyday, party, gift, work, sport, holiday, wedding, none]
  - bodyOnly: true if a human body part is visible, false otherwise
  - safeForBrand: true unless the image contains hold marks, packaging tape across the product, visible price tags, or competitor logos
  - reasoning: one short sentence on why safeForBrand was set as it was
  
  Return only the JSON. No prose.

  Copy prompt

• Cluster a batch of UGC into editorial themes

  VisionBatch

  I will paste 30 image URLs below. Group them into 4 to 6 editorial clusters suitable for a brand's PDP modules. For each cluster, return: cluster name (2 to 4 words), shared visual signal, image indexes in this cluster, and a one-line caption suggestion. Return JSON, no prose.

  Copy prompt

• Generate alt text from a UGC photo

  VisionAccessibility

  Write alt text for the attached customer photo. Constraint: 12 to 18 words. Describe what is visible (subject, setting, action). Do not editorialise. Do not include the brand name unless it is visible in the image. Return only the alt text string.

  Copy prompt

• Detect a hold mark or product damage

  VisionModeration

  Inspect the attached customer photo of a product. Return JSON with: hasHoldMark (boolean, true if you can see fingertips or a hand visibly pressing or holding the product), hasDamage (boolean), hasTags (boolean, true if a price tag or hangtag is visible), confidence (low/medium/high), evidence (one short sentence). Return only the JSON.

  Copy prompt

• Rank a batch of UGC by PDP-readiness

  VisionCuration

  I will paste image URLs. For each, score from 0 to 10 on each of: lighting quality, product clarity, scene authenticity, brand safety. Return a sorted JSON array, highest total first, with the four sub-scores and a short reasoning. Aim for 30 seconds of work per image; do not over-explain.

  Copy prompt

These are the actual scripts our customers send when asking creators for repost permission. Response rates noted where measured.

• First-touch rights DM (short, friendly)

  DM38% response

  Hi {firstName}, this is {senderName} from {brandName}. We love your post of our {productName}, the {oneSpecificDetail} caught our eye.
  
  Would it be alright if we reposted it on our website's customer gallery? Full credit to @{handle}, link back to your post, you can ask us to take it down any time. Just reply "yes" and that's all we need.
  
  Thanks either way for the post.

  Copy prompt

• Second-touch DM (no response after 5 days)

  DMFollow-up

  Hi {firstName}, gentle nudge in case the first message got buried. No pressure either way, totally fine if it's a no. If you want to read what we'd do with the photo first, our usage page is at {brandRightsPageUrl}.

  Copy prompt

• DM for a paid-rights upgrade (whitelisting)

  DMWhitelisting

  Hi {firstName}, the post you tagged us in is performing really well on our gallery. We'd love to use it in a paid ad on Meta for the next 30 days. Standard whitelisting rate is {rate}, payable on day one, and we will not modify the post or remove your handle. Reply "yes to whitelisting" and our ops team will send a one-page agreement.

  Copy prompt

• Decline a low-quality submission politely

  DMSoft-no

  Hi {firstName}, thanks so much for tagging us in your post. We're not adding it to the gallery this round, the lighting is a little tricky for our PDP layout, but we'd love to feature your next one. Here's our brief guide on what works: {creatorBriefUrl}.

  Copy prompt

These are the prompts that turn a marketing-copy PDP into something an LLM can actually quote. The shift is from prose to structured fact-statements.

• Rewrite a PDP for citation-readiness

  CopyAEO

  Rewrite the product description below so it can be cited cleanly by ChatGPT, Claude, Perplexity and Gemini when a shopper asks "what is the best {category} for {use case}".
  
  Rules:
  - Lead with a single fact-statement of 18 words or less that names the product and its strongest claim.
  - Follow with three to five short paragraphs, each leading with a bolded fact-statement.
  - No marketing adjectives ("revolutionary", "premium", "ultimate"). Replace with measurable facts.
  - Every claim must be either directly verifiable or qualified ("typical", "up to", "in tests").
  - End with a one-line summary that a chatbot could quote verbatim.
  
  Product description:
  {pasteHere}

  Copy prompt

• Generate a JSON-LD Product block

  SchemaAEO

  Given the PDP content below, write a JSON-LD block of type Product. Include: name, image (placeholder URL), description (max 250 chars), sku, brand, offers (price, priceCurrency, availability), aggregateRating (if review data present), and review (one to three, only if real review data is in the source). Output valid JSON-LD only, in a script tag.
  
  PDP content:
  {pasteHere}

  Copy prompt

• Extract product facts as Q and A pairs

  Q&AAEO

  Read the product page below. Generate 8 Q and A pairs that an AI shopping agent would ask before recommending. Each question must be one a real shopper would type. Each answer must be 25 words or fewer, fact-shaped, no marketing adjectives. Return as JSON array of {q, a} objects.
  
  Product page:
  {pasteHere}

  Copy prompt

Reviews are evidence. These prompts turn a review corpus into something a shopper can act on, and a reply into something that does not sound like a brand-bot.

• Summarise a SKU's review corpus

  ReviewsPDP

  Below are the 1- to 5-star reviews for one SKU. Write a 60-word summary suitable for the PDP. Constraint: include the average rating, the count, the single most-praised attribute, and the single most-criticised attribute. No filler. Do not invent. If a class of complaint is in fewer than 5% of reviews, do not mention it.
  
  Reviews:
  {pasteHere}

  Copy prompt

• Reply to a 2-star review (brand voice)

  ReplySupport

  Write a reply to the 2-star review below, in the brand voice described. The reply must: acknowledge the specific issue named (not a generic apology), name one concrete thing being done about it, offer a single next step the reviewer can take, be 60 words or fewer, and avoid the words "sorry to hear", "we apologise", "your satisfaction is".
  
  Brand voice: {voiceNotes}
  Review: {pasteHere}

  Copy prompt

• Convert a review into ad copy

  ReviewsAds

  Below is a 5-star customer review. Produce three ad copy variants based on it, each under 90 characters, suitable for Meta. Variants: one that quotes the strongest sentence verbatim with attribution, one that paraphrases the most useful claim with no quotes, one that frames the review as a question the shopper is also asking.
  
  Review:
  {pasteHere}

  Copy prompt

Scripts that hold attention past three seconds. The prompts below produce video plans tied to a specific SKU, not generic templates.

• Twenty-second shoppable video script for a single SKU

  VideoScripts

  Write a 20-second shoppable video script for the SKU below. Structure: 3-second hook (specific to this product, no "hey guys"), 12 seconds of demonstration (what the product does, narrated by the user), 5-second close (one concrete benefit + a clear next step). Tone: a customer, not a salesperson. Output the script as timestamped lines.
  
  SKU: {productName}
  Notes: {productDetails}

  Copy prompt

• Hook line bank for one product

  VideoHooks

  Given the product details below, generate 12 hook lines for a 3-second shoppable video opener. Each hook must do one of: ask a question the buyer is already asking, name a surprising fact, show a contrast ("most X do Y, this does Z"), or open mid-action. No hook may use "did you know", "let me show you", or "hey guys".
  
  Product: {pasteHere}

  Copy prompt

• Caption + first comment for an Instagram Reel

  InstagramCaptions

  Write a caption for the Reel described below. Constraints: 80 characters or fewer, no hashtag spam (max 2 tags inside the caption), one specific question to drive a reply. Then write a separate "first comment" with up to 8 relevant hashtags. Output both as plain text, labelled.
  
  Reel: {pasteHere}

  Copy prompt

These are the moderation passes our customers run before publishing. Cheap to run, expensive to skip.

• Brand-safety pass for outbound copy

  Moderation

  Read the copy below. Return JSON with: containsClaimRisk (boolean, true if there is a health, legal, financial or absolute-effectiveness claim), containsCompetitorRef (boolean), containsRegionRisk (array of any region-specific compliance triggers spotted, e.g. ASA, FDA, ASCI), containsAITell (boolean, true if the copy uses any of: "crucially", "notably", "moreover", "let's dive in", "game-changer", "seamless", "in today's digital landscape"), suggestedFixes (array of one-sentence fixes per issue).
  
  Copy: {pasteHere}

  Copy prompt

• Tone-match an existing brand voice

  Voice

  I will paste 6 examples of {brandName}'s real on-brand copy. Then I will paste a new draft. Rewrite the draft to match the voice of the 6 examples. Do not change facts. Do not invent product attributes. Keep the new draft within 10% of the original word count. Return only the rewritten draft.
  
  Examples:
  {paste 6 examples here}
  
  Draft:
  {pasteHere}

  Copy prompt