AI can fill a content calendar in an afternoon. It can also invent a second shop: wrong colors, softer claims, a new face every Reel. The difference is rarely the model. It is the brief.
This guide shows how product shops brief AI so photos, Reels, and captions stay on-brand. It sits on top of the brand kit and feeds on-brand product photos, Reels that sell, and character consistency. If the kit is empty, stop here and write the kit first.
A good brief is short, strict, and reusable. A bad brief is a mood paragraph and a hope.
Start from fixed facts, not vibes
Every brief should open with copy-paste facts:
- Product name, variant, true materials, true size cues.
- Shop sentence from the brand kit.
- Primary and secondary colors with values.
- Voice do’s and don’ts.
- Offer line and CTA if the asset is commercial.
- Character or face rule, or “no people / no mascot.”
Then add the job of this asset: product tile, proof, offer, Reel cover, Story FAQ. One job. The selling grid jobs work as brief labels.
Save this block on the brand profile so you are not rewriting the shop in every chat window.
Write bans as hard lines
AI follows negative space poorly unless you name it. Add a ban list every time:
- Do not recolor the product.
- Do not add logos or badges that are not ours.
- Do not invent ingredients, certifications, or testimonials.
- Do not change the mascot wardrobe or face.
- Do not use trend templates that hide the product in the first second.
- Do not write urgency we cannot fulfill.
Update bans after each review using metrics that matter. If a look got reach and damaged trust, it becomes a ban, not a style to scale.
Brief photos differently from Reels
For still product photos:
- Angle and crop that must match the catalog.
- Background: kit neutral or a named set, not “aesthetic vibe.”
- Lighting: soft, hard, daylight — pick one the shop already uses.
- Text on image: yes/no, and the exact words if yes.
- Reference: attach or describe the hero photo that is already approved.
For Reels:
- Silent story in one sentence. If sound dies, the product story must still work.
- First frame: product readable.
- Motion: one action, not five cuts of unrelated b-roll.
- Audio: brand bed or product-led sound, following Reel audio that sells.
- On-screen text: same words you are willing to put on the page.
UGC-style asks need extra bans so “handmade casual” does not become “wrong product in someone else’s kitchen.” Use UGC-style Reels without creators as the quality bar, not as permission to get sloppy.
Brief captions like search plus voice
Caption briefs should include:
- The exact product and the buyer question you are answering.
- Two or three search phrases a real buyer might type, from captions that rank and sell.
- Voice samples: one good caption, one bad caption.
- CTA that matches the page button.
- Length target: short for offers, longer for teaching posts.
Never ask AI to “make it more viral.” Ask it to answer a specific objection in the shop’s voice.
Use a two-pass review every time
Pass one is brand safety. Pass two is commerce.
Pass one — brand:
- Product still looks like the product.
- Colors inside the kit.
- Face or character consistent, or absent as ordered.
- No invented claims.
- Readable at thumbnail size.
Pass two — commerce:
- Job of the asset is obvious.
- Offer and link language match the page.
- Fits the grid pattern if it will live on the profile.
- Does not contradict Highlights or DM blocks.
If pass one fails, do not publish and “fix in the caption.” Regenerate or edit the image. Captions cannot repair a false product.
Lock characters before you scale faces
Any repeating person or mascot needs a sheet before a batch. Build it on the character sheet, then quote sheet details in every brief: wardrobe, colors, age look, what they never wear. Month-long batches without a sheet are how shops get a different “helper” every Tuesday.
If you only need hands or a product-only world, say so explicitly. Silence invites AI to cast extras.
Batch with one master brief and small deltas
For a month batch, write one master brief for the shop, then a one-line delta per asset:
- Master: kit facts, bans, character, voice.
- Delta: “Proof tile — show stitch detail on the navy tote, no text.”
- Delta: “Offer tile — restock Saturday, include price as on page.”
- Delta: “Reel — pour test, silent-first, brand bed only.”
This is faster than twenty independent chats and keeps the shop coherent. Preview the resulting grid in Instagram UI before you schedule.
Attach proof when the product must stay exact
For catalog-critical stills, attach the real product photo or a pack shot to the brief whenever the tool allows. Words alone are how AI “improves” a label into a fiction. Say which parts are locked: logo placement, label text, hardware color, stitch color. Say which parts may move: background, soft props, crop.
If the tool cannot take a reference image, describe the locked parts in measurements a stranger could check: “navy body, brass zipper, cream label centered under the rim.” Vague luxury adjectives are not measurements.
Stop brief habits that create brand drift
- “Surprise me.” Surprises belong in personal art, not in a catalog.
- References from five unrelated brands. Pick one approved shop reference set.
- Prompt stacking without facts. Long poetic prompts with no product specs.
- Accepting the first pretty fail. Pretty and wrong still fails pass one.
- Different briefs from different teammates. One owner for the master brief.
Add broken outputs to the ban list the same day. The brief should get stricter after mistakes, not vaguer.
Brief sound and UI the same way you brief pixels
If the asset needs music or a mock profile frame, extend the brief. For audio: mood in three words, instruments to avoid, and “product must stay audible or readable over the bed.” For UI previews: bio text exact, grid order, and which tile is new. Treat Instagram UI checks as part of pass two, not as decoration after publish.
Sound without a brief becomes a trend bed that fights the product. UI without a brief becomes a pretty phone frame around the wrong shop.
Tie briefs to launches and the first month
Launch weeks need a frozen brief. Use the offer card from the launch plan as the top of every prompt so AI cannot invent a second price. New shops should write the master brief during the first 30 days before they ask AI for volume.
When buyers ask the same question in DMs, add that question to the next teaching brief. AI is useful when it answers real demand, not when it invents topics from thin air.
Keep a failed-output library
When an image or Reel fails pass one, do not only delete it. Save one frame and write the reason: recolored product, wrong face, invented badge, unreadable type. Add that reason to the ban list. After a month you will have a sharper brief than any generic “best practices” thread, because it is built from your shop’s actual mistakes.
Share that library with anyone allowed to generate. New helpers should read bans before they open the studio.
Time-box generation so batching stays honest
Set a ceiling: for example, three regenerations per asset, then a human edit or a reshoot decision. Endless rerolls feel productive and still miss the week. The monthly batch method only works when AI is a fast first draft inside a fixed afternoon, not an open tab until midnight.
Run the brief inside Neurollamas on purpose
Keep the master brief on the brand profile, generate stills and motion in the studio, lock faces on the character sheet, and let Content Autopilot hold only assets that passed both review passes. Mag articles around this one are the curriculum; the brief is the daily instrument.
AI does not replace a product shop’s taste. It multiplies whatever brief you give it. Give it the shop you actually run — colors, claims, face, and bans — and the calendar fills without inventing a stranger’s brand. Keep the Mag cluster next to the brief so every new asset still serves the same pillar: a product shop strangers can trust.