Prompt Engineering for E-commerce: Product Descriptions, Reviews, and Ads

E-commerce copy is SKU-specific, compliance-bound, and channel-formatted. Amazon truncates titles at 200 characters. Google Shopping rejects exaggerated claims. FTC cares about review authenticity and disclosure. Paste "write a product description" without a SKU block and models invent materials, dimensions, and five-star narratives you cannot ship. Prompt engineering for e-commerce means a structured SKU context block, FTC and Amazon compliance disclaimers in every prompt, separate templates for titles, bullets, descriptions, ads, and review responses, plus FAB (Features-Advantages-Benefits) architecture. GPT-5.5, Claude Opus 4.8, and Gemini 3.1 Pro / 3.5 Flash each fit different catalog workflows. This guide covers templates, compliance language, model routing, and FAQ.

Why generic product prompts fail

Weak: Write a compelling description for our wireless earbuds. Output: "Premium crystal-clear sound with industry-leading noise cancellation" — no driver size, no battery hours, no compatibility list, unverifiable superlatives. Strong: SKU block with verified attributes + compliance footer + channel FORMAT + FAB mapping. Every claim traceable to SKU facts or labeled [NEED SPEC].

The SKU context block

Foundation of every e-commerce prompt. Paste once per product.

SKU BLOCK:

SKU: [id]

Product name (internal): [name]

Category: [Amazon browse node / Google category]

Brand: [brand]

Price: [if allowed in copy]

Materials: [verified]

Dimensions / weight: [verified]

Compatibility: [devices, regions, voltage]

Certifications: [FCC, CE, organic, etc. — or NONE LISTED]

What's included: [box contents]

Warranty: [terms or link]

Restricted claims: [no medical, no #1 unless documented]

Competitor references: [none / naming policy]

Instruction: Do not state any attribute not listed. Missing attribute → [NEED SPEC].

FTC and Amazon compliance disclaimer

Add to every prompt — not optional for listings and ads.

COMPLIANCE:

• This is marketing draft, not legal approval
• No fake reviews, implied testimonials, or unsubstantiated #1 claims
• Amazon: follow category style guide; no promotional language in title (Amazon rules vary by category)
• FTC: material connections disclosed if copy implies endorser
• Reviews response: no incentives for positive reviews mentioned
• Superlatives require SOURCE in SKU BLOCK or omit

Human compliance review before publish. AI prompts reduce drafting time — not liability.

FAB architecture

Map features to customer outcomes before writing prose.

FAB TABLE (generate or paste):

Feature: [spec] → Advantage: [what it enables] → Benefit: [customer outcome in plain language]

Example:

Feature: 40mm drivers → Advantage: wider frequency response → Benefit: hear detail in vocals and instruments without max volume

Prompt: Convert FAB rows to bullets and description — lead with benefits, support with features. No feature without linked benefit.

Product title template

CHANNEL: [Amazon / Shopify / Google Shopping]

FORMAT:

Amazon title: max 200 chars (category-dependent), brand + key spec + size/color + model, no promotional phrases

Shopify: readable front-loaded keywords, 70 chars visible in many themes — front-load

Google: clear product identity, no all-caps

TASK: 2 title variants from SKU BLOCK

RULES: Include brand, product type, distinguishing spec. No FREE SHIPPING, no BEST.

OUTPUT: Variant A | Variant B | char counts

Bullet points template (Amazon-style)

FORMAT:

5 bullets, max 500 chars each (Amazon typical), capitalize first letter, no end punctuation on bullets

Bullet 1: primary benefit hook

Bullets 2–4: feature-benefit pairs from FAB

Bullet 5: what's in box / warranty

BANNED PHRASES: [Amazon prohibited terms for category — e.g. antibacterial if unapproved]

Prompt: Output bullets only. Each bullet must map to one FAB row.

Long description template

FORMAT:

Structure: opening benefit paragraph (80 words) → Features section (H3-style headers in plain text) → Specs table (markdown) → Care/use → Warranty

Tone: [brand voice reference]

Length: [300–500 words for DTC / Amazon A+ separate]

HTML: [allowed tags if Shopify rich text]

RULES: Specs must match SKU BLOCK exactly. Conflicts → flag [SPEC CONFLICT].

A+ Content / enhanced brand content prompt

MODULE: [comparison table / brand story / feature grid]

SKU BLOCK: [paste]

FORMAT: Headline max 80 chars per module, body max 300 chars, no competitor names

Comparison table: our SKU vs generic category — not named competitors unless legal cleared

OUTPUT: Module-by-module copy with image suggestions labeled IMAGE SUGGESTION:

Google Shopping ad copy template

FORMAT:

Title: 150 chars max

Short description: 250 chars

Highlight: price if allowed, shipping, return policy from SKU facts only

No promotional superlatives Google policies disallow

OUTPUT: title, short description, character counts

Meta / paid social ad template

FORMAT:

Primary text: 125 chars visible before see more — hook + benefit

Headline: 40 chars

Description: 30 chars

CTA button: [Shop Now / Learn More]

SKU facts only; add DISCLAIMER if health/beauty category

OUTPUT: labeled fields matching Ads Manager

Email product launch template

FORMAT:

Subject: max 50 chars, no spam triggers

Preview: max 90 chars

Body: hero benefit, 3 bullets, social proof from SKU (awards in BLOCK only), single CTA

TASK: launch email for SKU

No fake urgency (countdown) unless FACT: sale end date in SKU BLOCK

Review response templates

Positive review response:

TASK: Thank customer, mention one specific product attribute from SKU (not generic thanks), invite to reach support for issues, max 80 words, no incentive for more reviews

Negative review response:

TASK: Acknowledge issue, apologize without admitting liability, offer support channel, no argument, no PHI, max 100 words

Paste review text + SKU BLOCK. Output response only.

Q&A and customer question template

CUSTOMER QUESTION: [paste]

SKU BLOCK: [paste]

RULE: Answer only from SKU. If unknown: We do not have that specification listed. Contact support at [channel].

No guessing on compatibility or safety.

Shopify SEO and collection page template

CHANNEL: Shopify

FORMAT:

Product title: front-load primary keyword, 70 chars visible in many themes

Meta description: 155 chars, benefit + CTA, no duplicate title text

Collection intro: 100–150 words above grid — category story + internal links

TASK: Generate title, meta, collection intro from SKU BLOCK

RULES: No keyword stuffing. One primary keyword from KEYWORDS block.

Amazon backend search terms template

FIELD: backend search terms (250 bytes max, no repetition of title words)

SKU BLOCK: [paste]

RULES: lowercase, comma-separated synonyms and alternate spellings not in title. No competitor brands. No subjective claims.

OUTPUT: single line + byte count estimate

Subscription and replenishment copy

SKU BLOCK + SUBSCRIPTION FACTS: interval options, cancel policy, first-order discount if in SKU

TASK: bullets emphasizing convenience and transparency — no hidden terms

CTA: Subscribe and save [X]% only if percentage in SUBSCRIPTION FACTS

FTC: clear billing cadence language if prompted for email or landing page

PROMO FACTS:

Sale price: [if in SKU]

Dates: [start-end]

Stacking: [not combinable / etc.]

Overlay on title/description prompts: Include promo only in PROMO FACTS window. Otherwise omit.

Returns and shipping snippet block

SHIPPING FACTS: [free over $X, regions, delivery window]

RETURNS: [window, condition, who pays return shipping]

TASK: 2-sentence trust block for product description footer — facts only

Amazon: many categories forbid shipping claims in title — footer only unless category allows

LOCALE: [de-DE]

SKU BLOCK: [English facts]

RULES: Translate for local marketplace (metric units, local compliance phrases). Do not translate brand name. Keep SKU numbers.

Amazon DE title rules differ from US — note locale FORMAT.

Variant and matrix products

PARENT SKU: [id]

VARIANTS: [color/size matrix with per-variant specs]

TASK: shared description + variant-specific title suffix

RULE: Variant-specific claims (size, color) only on matching variant — no bleed

Bundle and kit copy

BUNDLE CONTENTS: [SKU list with quantities] | TASK: emphasize complete solution benefit without implying items sold separately unless true | List each component per FTC bundle disclosure — human legal review.

Walmart and marketplace extension

CHANNEL: Walmart Marketplace / Target Plus | FORMAT: shorter titles than Amazon, plain spec language, no promotional symbols | SKU BLOCK required — reject unverifiable environmental claims | TASK: adapt Amazon title + bullets to marketplace FORMAT with compliance footer

BUNDLE CONTENTS: [SKU list with quantities]

TASK: emphasize complete solution benefit without implying items are sold separately unless true

List each component per FTC bundle disclosure expectations — human legal review.

DTC landing page hero block

FORMAT: headline 8 words max, subhead 20 words, hero benefit paragraph 60 words, 3 icon bullets from FAB, single CTA button text

SKU BLOCK + brand voice | No countdown timers unless sale end in SKU | OUTPUT: labeled fields for Webflow/Shopify section paste

Comparative advertising legal guardrails

If COMPARISON approved: name competitor only in approved doc types; never in Amazon title; use neutral spec comparison | Add: comparative claims require legal clearance tag [LEGAL CLEARED] in prompt or omit

Category-specific compliance notes

Supplements: structure/function claims only from approved list. Children's: choking hazard, age grade from SKU. Electronics: FCC ID from SKU if required. Apparel: fabric content exact match. Add CATEGORY COMPLIANCE line per vertical.

Before vs after: title

Weak: Best Wireless Earbuds 2026 - Amazing Sound!!!

Strong from SKU: Brand Model X True Wireless Earbuds, Bluetooth 5.3, 36Hr Battery, IPX5, Black

Verifiable specs, no superlative, within char limit.

Before vs after: bullets

Weak: Crystal clear audio experience for audiophiles

Strong: Up to 36 hours total battery with charging case — 8 hours per charge for all-day listening (per SKU test conditions)

Benefit + spec + qualification.

Keyword research injection

KEYWORDS (use naturally, no stuffing):

Primary: [term]

Secondary: [2–3 terms]

Backend search terms (Amazon): [250 bytes separate field]

Prompt: Include primary in title if natural. Secondary in bullets/description. No repetition spam.

Competitor comparison (careful)

Only if legal approved:

COMPARISON: [our SKU] vs [category generic benchmark]

No competitor trademarks in Amazon titles. Comparison chart in A+ only with approval.

Accessibility and inclusive language

RULES: alt text describes product accurately; avoid ableist idioms; size/color inclusive descriptors from SKU only

TASK: review generated description for accessibility — list issues + revised paragraphs

E-commerce copy affects screen reader users and ad policy reviewers alike.

IMAGE: [describe shot]

ALT TEXT: max 125 chars, describe product + color + key visible feature

No keyword stuffing in alt — accessibility first.

Model routing for e-commerce

GPT-5.5: title/bullet variants, JSON SKU-mapped output, ad field batches

Claude Opus 4.8: long descriptions, brand voice, nuanced FAB prose

Gemini 3.1 Pro: large SKU catalogs, multi-locale in one session

Gemini 3.5 Flash: bulk title drafts — mandatory compliance pass on GPT-5.5 or Opus after

JSON structured output for PIM

OUTPUT JSON:

{ "title": "", "bullets": [], "description_html": "", "meta_description": "", "needs_spec": [], "compliance_flags": [] }

needs_spec lists every [NEED SPEC] the model wanted. compliance_flags lists hedged claims.

Feeds PIM or human review queue.

Batch catalog workflow

Paste 5 SKU BLOCKs. TASK: title + 5 bullets each. OUTPUT table SKU | title | bullets

Registered PromptMake users (5/day) batch 5 SKUs; guests (3/day) prioritize hero SKU.

Common failures

No SKU block → invented specs. Superlatives in Amazon title → listing suppression. Review response argues with customer → brand damage. Same description for all marketplaces → rule violations. FAB missing → feature dumps. Ignoring [NEED SPEC] → wrong compatibility claims.

PromptMake workflow

Store SKU BLOCK template in /text preset. Generate title + bullets in one run, description in second (guest budget). Registered: full listing set per SKU. Guests: 3 /text generations per day. Registered: 5/day. Prioritize hero SKU and compliance-heavy categories first.

FAQ

What must be in every e-commerce prompt?

SKU BLOCK, compliance disclaimer, channel FORMAT, and rule: no facts outside SKU.

Can AI write Amazon reviews?

No. AI drafts review responses and Q&A — not fake customer reviews. FTC and Amazon prohibit fake reviews.

How do I handle missing specs?

Model outputs [NEED SPEC] — never guess. Fill SKU BLOCK and regenerate.

Amazon title character limit?

Often 200 characters; varies by category. State limit in FORMAT. Front-load brand + product type.

Best model for 200-SKU catalog?

Gemini 3.1 Pro for batch with JSON schema; Flash for first pass then Opus compliance scrub on hero SKUs.

FAB vs features list?

FAB forces benefit linkage. Feature lists without benefits convert poorly and read like spec sheets.

Health and beauty claims?

Only approved structure/function phrases from legal. Prompt CATEGORY COMPLIANCE: supplements.

Review response tone?

Professional, specific, non-defensive. Never confirm medical outcomes or offer refunds outside policy in AI draft without human edit.

Google vs Amazon same copy?

No. Different title rules and length. Separate FORMAT blocks per channel.

PromptMake daily limits?

Guests 3/day, registered 5/day on /text. Plan one SKU per generation for full listing coverage.

Related articles

prompt-engineering-marketing-copy-aida-pas, ai-prompts-consistent-brand-voice, structured-output-prompting — ad frameworks, brand voice, JSON PIM pipelines

Bottom line

E-commerce prompt engineering is SKU-grounded copy with compliance footers and channel-specific FORMATs.

FAB before prose. Titles, bullets, descriptions, ads, and reviews each get templates. GPT-5.5 for variants, Opus for voice, Gemini for catalog scale. Human compliance review always.