AI for your product catalog, done right: 18 prompts that won't tank your Google rankings

In early 2024, Google rolled out a policy called scaled content abuse. Within months, ecommerce sites that had quietly published tens of thousands of AI-generated product descriptions saw their organic traffic collapse — some by 80% or more in a single update.

The lesson most people took away was wrong. They blamed AI.

What Google actually punished was the shape: identical, mass-produced, undifferentiated pages with no human in the loop and nothing a real shopper couldn't already get from the supplier's spec sheet. The stores that used AI as a drafting partner — and then edited, fact-checked, and shipped — kept their rankings and made their catalogs measurably better.

This guide is the second path.

You'll find: what Google actually rewards (and what it actually punishes), five rules for AI catalog content that survives algorithm updates, and 18 copy-paste prompts you can run on your store today to upgrade descriptions, meta tags, FAQs, alt text, category pages, translations, and stale-product refreshes — without making your store look like a content farm.

Use the prompts as templates, not gospel. The whole point is that the human still ships the work.

What Google actually penalizes (and what it doesn't)

The "AI content is banned" rumor isn't true. Google has stated, in writing, that AI-assisted content is fine when it's helpful, original, and people-first. What gets penalized is a pattern — and the pattern has a name.

Scaled content abuse (Google's own term) is when you generate content at scale without enough human curation or originality to make it useful. It applies whether the content is written by AI, by a freelancer farm, or by a templating system. The medium isn't the issue; the shape is.

In practice, Google looks at three things on your product pages:

1. Helpfulness. Does the page answer the actual question a real shopper has when they land there? "Helpful" is the literal name of the update ("Helpful Content"). Pages that read like specs-with-adjectives, but never answer "is this the right one for me?", lose ranking.
2. Originality. Does the page show first-hand experience or a point of view nobody else has? (Google calls this "E-E-A-T": Experience, Expertise, Authoritativeness, Trust.) Auto-spun copy from the supplier sheet shows none of those.
3. Pattern. If 5,000 of your pages share the same template, same adjectives, same paragraph structure, and the same two-sentence intro, Google notices. So do shoppers.

The fix isn't "stop using AI." It's: use AI to draft, ground it in real product data, edit every output, vary the shape, and add what an LLM can't.

The five rules

Five principles that separate AI-assisted catalog work that wins from AI-spam that gets demoted.

1. Ground every output in real data. Don't ask the LLM what your product is — tell it. Paste the spec sheet, the supplier description, the existing customer reviews, your photos, anything you have. The model's job is to organize and translate, not invent. The moment a model invents a fact — and they do — your description becomes a liability.

2. Edit every single output. Not "skim." Edit. Fix the one sentence that sounds like marketing fluff. Add the one detail only you know. Remove the adjective stack. If you can't justify spending two minutes editing a draft, you shouldn't publish it.

3. Don't mass-publish the same shape. A catalog where every product description opens with "Discover our premium…" or ends with "Add to your collection today" reads as templated — both to shoppers and to crawlers. Vary the structure. Some products should lead with a use case, some with a spec, some with a story. Variety is a quality signal.

4. Match shopper language, not LLM defaults. LLMs default to a mildly-corporate tone: "discover", "elevate", "indulge in", "premium quality". Real shoppers don't search that way and don't read that way. Tell the model the actual words your customers use — pulled from reviews, support tickets, and search queries — and force it to use them.

5. Add what an LLM can't. Your photos, your customer reviews, your shipping speed, your founder's reason for picking this product, the trade-off you made on materials, the screw-up you fixed last month. These are the things that make a page yours. The AI handles the volume; the human adds the distinctiveness.

The 18 prompts

Each prompt below is a template. Replace the {curly brace} placeholders with your real data before you run it. The prompts work with any major model (Claude, GPT, Gemini); copy and paste as-is.

Two universal tweaks that should go on every prompt:

• End every prompt with: "If a key buyer question is unanswerable from the data above, output [NEEDS: question] so I can fill it in. Do not invent."
• Tell the model your brand voice in one sentence: "Write in {your brand voice — e.g. "warm, direct, slightly nerdy; no marketing speak"}."

These two add-ons fix 80% of what's wrong with raw AI catalog copy.

A. The product page (prompts 1–6)

1. Master product description from raw specs

Use when: you have a supplier spec sheet and need a real product description.

You are writing the main product description for a {product category} sold on {store name}.

Here's what we know about this product:
- Brand: {brand}
- Specs (paste verbatim): {raw spec list}
- Materials: {materials}
- Intended use: {what shoppers actually use this for}
- The one thing it does better than alternatives: {your honest differentiator}
- Common shopper concern: {a real objection you hear, e.g. "is it warm enough for ski touring?"}

Write a 90–130 word product description that:
- Opens with the single biggest reason a shopper would want this (not a feature list)
- Uses concrete, sensory language ("dense enough to block wind") instead of stacked adjectives ("luxurious, premium, high-quality")
- Weaves 3–4 specific specs naturally into sentences — never a bullet dump
- Names the common concern above and answers it honestly
- Ends with one line about who it's NOT for, so the right buyer self-selects

If a key buyer question is unanswerable from the data above, output [NEEDS: question]. Do not invent.

Why this works: the "who it's NOT for" line is the most underused move in ecommerce copy. It signals confidence, helps the right buyer convert, and is the opposite of mass-produced fluff.

2. Short product-card description (60–90 chars)

Use when: the one-line description shown on collection pages, search results, and inside meta tags.

Given this product description: {paste the full description from prompt 1}

Write 5 candidate one-line descriptions, each 60–90 characters, that:
- Lead with the biggest concrete benefit (not "stylish" — what does it actually do for the shopper?)
- Use 0 marketing adjectives ("premium", "luxury", "discover")
- Each candidate uses a different angle: use case, material, ideal buyer, problem solved, surprising detail

Number them 1–5. No emojis.

Why 5 candidates: pick the one that feels least like AI wrote it. The act of choosing is the editing.

3. Benefit-led bullet points

Use when: the bullet list under the product title. Most stores get this backwards — they list features, not benefits.

For this product: {paste description or specs}

Write 4–6 bullets in the format: [concrete benefit to shopper] — [the spec that delivers it].

Example: "Doesn't dent in checked baggage — 100% polycarbonate shell, not ABS"

Each bullet:
- Starts with what the shopper gets, not what the product has
- Names the spec that enables it (so it's verifiable, not marketing-speak)
- Is under 90 characters

Skip generic bullets ("high quality construction"). If a benefit can't be tied to a real spec from the data, drop it.

4. SEO meta title

Use when: the page's <title> tag — what shows in Google's blue link.

For this product page:
- Product name: {full name}
- Brand: {brand}
- Single biggest selling point: {one phrase}
- The store name suffix on every page: {e.g. " · Store Name"}

Write 5 candidate <title> tags, each ≤ 60 characters INCLUDING the suffix, that:
- Lead with how shoppers actually search (not the brand's marketing name for it)
- Include the most important distinguishing attribute (size, material, use, color)
- Don't start with the brand name unless the brand is the search driver

Show the character count next to each.

Why character count: anything over ~60 gets truncated in Google's results. You want the math visible.

5. SEO meta description

Use when: the page's <meta description> — the grey text under the title in Google.

For this product page:
- Title we picked: {meta title}
- Top 3 buyer questions the page answers: {list 3 questions from your reviews or support data}
- One reassurance shoppers need before clicking (free shipping, returns, in stock, etc.): {one}

Write 3 candidate meta descriptions, each 140–160 characters, that:
- Answer at least one of the buyer questions in the first 100 characters (so it shows in mobile previews)
- Include the reassurance
- Don't repeat the title's exact phrasing
- End with an implicit call to action — never literal ("Buy now") — like "See the full size range" or "Compare with the Pro model"

6. Alt text for product images

Use when: every product image. Most stores leave this blank — it's a free SEO and accessibility win.

I'll paste 3 image descriptions for the same product. For each, write alt text 80–125 characters long that:
- Describes what the image actually shows (color, angle, context, model wearing it if applicable)
- Includes the product name once, naturally
- Does NOT keyword-stuff — alt text is read by screen readers and Google, both of which penalize stuffing
- Each image's alt text is different (varying angle, context, or detail)

Image descriptions:
1. {describe image 1 — e.g. "Front view, navy color, on a wooden table"}
2. {describe image 2 — e.g. "Worn by a model walking through a wet street"}
3. {describe image 3 — e.g. "Close-up of the YKK zipper detail"}

B. Shopper decision support (prompts 7–10)

7. Product FAQ from real questions

Use when: generating a FAQ block — but only on products with enough volume to have real customer questions.

For this product: {paste description}

Here are real questions shoppers have asked about it (from support tickets, reviews, chat logs):
{paste 8–15 real questions verbatim}

Group the questions into 4–6 themes and write one FAQ Q+A pair per theme. Each answer:
- Is 1–3 sentences long
- States a clear answer in the first sentence (don't bury the lede)
- Names a specific spec or fact, not a marketing claim
- Where there's an honest limitation (only ships to US, sizing runs small, etc.), say so plainly

If a real question can't be answered from the product data we have, output [NEEDS: question] and skip it. Do not invent answers.

Why this is the killer prompt for SEO: product FAQs answering real questions get pulled into Google's "People also ask" boxes. Real questions, real answers, plus FAQPage JSON-LD schema, and you'll outrank competitors who copy-paste a generic 5-question FAQ.

8. Sizing / fit guidance

Use when: anything with a size — apparel, footwear, furniture, even electronics with footprint constraints.

For this product: {paste product info + size chart}

Here's what real customers have said about fit in reviews:
{paste 5–10 review snippets mentioning fit, sizing, scale}

Write a sizing guidance section, ~150 words, that:
- Names the actual pattern from the reviews (e.g. "runs about a half size small for wide feet")
- Gives a specific recommendation for the most common scenario
- Names the exception (when to size up, when to size down)
- Includes one concrete measurement that's verifiable (e.g. "the chest measures 22 inches laid flat at size M")

Do not generalize beyond what the reviews and the size chart support.

9. "Which variant is right for me?" guide

Use when: products with multiple variants (Pro vs Standard, S/M/L, different colors with different specs).

For this product line:
- Variant A: {name, key specs, ideal buyer in one line}
- Variant B: {name, key specs, ideal buyer in one line}
- Variant C (optional): {…}

The shopper has just landed on the parent product page and is deciding. Write a 200-word section titled "Which one is right for you" that:
- Frames the choice in terms of the buyer's situation, not the spec ("If you commute 5+ miles daily" — not "If you need 500Wh of battery")
- Gives a concrete recommendation per scenario, naming the variant
- Calls out the one situation where you'd recommend AGAINST the most expensive variant (counterintuitive — but builds trust and conversion)

10. Honest comparison vs a specific competitor

Use when: you compete head-to-head with a known product and shoppers compare.

Compare these two products fairly:

OURS: {our product name and key specs}
THEIRS: {competitor product name and key specs}

Write a comparison section, ~200 words, that:
- Names 2 things the competitor does genuinely better (yes, really — if you can't, the section won't land)
- Names 2 things ours does better
- Concludes with the buyer profile for whom each is the right choice
- Uses no marketing language — just facts and tradeoffs

If a competitor spec is unknown, write [NEEDS: spec X for {competitor}] — never guess.

Why brutal honesty wins: shoppers who feel "sold to" leave; shoppers who feel "advised honestly" buy. Stating where the competitor wins is the single most trust-building move on a comparison page.

C. Category & navigation (prompts 11–12)

11. Category page intro

Use when: the short intro at the top of a collection page (e.g. /collections/running-shoes).

For the category page "{category name}":
- Total products in this category: {number}
- The 3–5 buyer personas who shop here: {list}
- The single most common shopper question on this page: {question}

Write a 120–160 word category intro that:
- Answers the shopper question in the first paragraph (don't bury it)
- Tells the shopper how to narrow down (mention the actual filter options on the page, e.g. "Filter by gait type if you over-pronate")
- Names one product worth checking first and why (this is internal linking + helpful at once)
- Does not list every product; the page already does that

No headings. Two short paragraphs.

12. Curated collection description

Use when: you build curated sets — "Best running shoes for flat feet", "Gifts under $50", "Home-office desk setups".

Collection: {name}
The promise of the collection: {one sentence — what does it deliver?}
The products in it: {list product names + a 1-line description of each}
Who shops this collection: {persona}

Write a 100-word collection intro that:
- Says what's IN the collection and what's deliberately OUT (the second is more useful)
- Names the criteria used to pick (helps shoppers trust your curation)
- Mentions 1–2 products by name with a single-line reason

End with: "Updated {date}, last revised after {trigger — e.g. spring stock refresh}." Trust signal.

D. Translation & localization (prompts 13–14)

13. Translate a product description (don't just translate — localize)

Use when: selling into a market with a different language.

Translate the following product description into {target language, e.g. German} for shoppers in {target country, e.g. Germany}.

Rules:
- Do NOT translate word-for-word. Translate the intent.
- Convert units to local norms (US to metric, USD to {{price}} placeholder)
- Replace any culturally-specific references (sports leagues, brands, holidays) with the local equivalent or remove them
- Keep brand and product names in English unless they have an established local name
- Use the formal/informal tone customary for ecommerce in {target country}

Source:
{paste English description}

14. Regional re-localization (English → English)

Use when: US store expanding to UK/AU, or vice versa.

This product description is written for {source market, e.g. US} shoppers. Rewrite it for {target market, e.g. UK} shoppers.

Adjust:
- Spelling (colour, organisation)
- Units (oz to g, inches to cm)
- Sizing (US 9 → UK 8, US apparel → UK apparel)
- Cultural references (sports leagues, holidays, brand examples)
- Pricing format and currency placeholder (USD → GBP)
- Reassurance lines (US "free returns" may need adjusting for UK consumer-rights baseline)

Source:
{paste description}

Output the rewritten version, then below it list every change you made so I can review.

E. Refresh & maintenance (prompts 15–16)

15. Refresh a stale product description (1+ year old)

Use when: the product is still selling but the description hasn't been updated.

Here's the current product description: {paste current}

Here's what's changed since it was written:
- New review / feedback patterns: {paste recent review themes}
- New use case shoppers are buying for: {if any}
- New competitor in the category: {if any, with how we compare}
- Anything we've improved in the product itself: {if any}

Rewrite the description (90–130 words) so it reflects the new information. Keep what was working; replace what's stale. Below the new description, list every sentence you removed and why.

Why this matters for SEO: Google's "freshness" signal counts. Pages that get refreshed with real updates (not just date changes) tend to climb. The trick is meaningful refreshes, not cosmetic ones.

16. Seasonal / context refresh

Use when: the same product is relevant for a new season, occasion, or use case.

Product: {paste current description}
Upcoming season/occasion: {e.g. "back-to-school", "winter commute", "holiday gift"}

Write a 100-word seasonal angle to add to the product page (don't replace the main description) that:
- Names the season/occasion explicitly in the first sentence
- Connects the product's existing strengths to the new context
- Does not invent new features or claims
- Has a date range it's relevant for (e.g. "Updated for the 2026/27 ski season")

Place this as a dated callout box on the product page, removable when the season ends.

F. Brand voice & differentiation (prompts 17–18)

17. Rewrite in your brand voice

Use when: AI's default voice is generic. You need every page to sound like you.

Here is our brand voice in 3 examples. Study them.

Example 1 (Instagram caption): {paste real example}
Example 2 (email subject lines): {paste 5 real subject lines}
Example 3 (how our founder talks about the products): {paste a real quote or interview snippet}

Rules our voice follows:
- {rule 1, e.g. "Never say 'discover' or 'elevate'"}
- {rule 2, e.g. "Use one piece of dry humor per piece"}
- {rule 3, e.g. "We always cite a specific person, place, or number"}

Rewrite this generic AI-produced description in our brand voice, keeping the facts identical:
{paste the generic version}

Show the rewrite, then a 3-bullet list of the specific brand-voice moves you made.

18. Find the differentiator nobody else mentions

Use when: before you write any of the prompts above, find your real angle.

Here's our product: {full specs + description}
Here are 3 competitors' descriptions of similar products: {paste all 3 verbatim}
Here are real customer reviews of OUR product: {paste 10 review snippets}

Tasks:
1. List 5 things our competitors all emphasize (the table stakes of this category).
2. List 3 things OUR customers mention in reviews that the competitors don't talk about at all.
3. Pick the strongest one from list 2 and write a single sentence describing the angle.

That sentence becomes the spine of every product description for this product. Don't make it up — only use what's actually in the reviews.

Why this is the most important prompt in the article: if you only run one, run this one. It tells you what your product page should be about — and that's the question 90% of catalogs don't answer.

The workflow — how to run these on one product

Doing all 18 on a single product is overkill. For a new product, run them in this order (skip what doesn't apply):

1. Prompt 18 (find the angle) — 5 minutes, done once
2. Prompt 1 (master description) — uses the angle from 18
3. Prompt 3 (bullets) — derived from 1
4. Prompts 4 + 5 (meta title + meta description)
5. Prompt 2 (short card description)
6. Prompt 7 (FAQ) — only once you have real shopper questions; skip on day 1, run in month 2
7. Prompt 6 (alt text) — for every image
8. Prompts 9, 8 (variant guide, sizing) — if the product needs them

For a category page, run 11. For a curated collection, 12. For translations, 13–14. For an annual refresh, 15. For seasonal pushes, 16.

For your whole catalog: do not run any of these in bulk without a human in the loop. The whole point of this article is that the human is what saves you from the "scaled content abuse" trap.

Antipatterns — what NOT to do

• Don't bulk-publish without editing. "1,000 product descriptions overnight" is the headline that gets you demoted.
• Don't let the LLM invent specs. It will, confidently. Always ground in real data, and always use the [NEEDS: question] trick.
• Don't keyword-stuff. Modern Google reads context, not density. Stuffing "running shoes" 12 times in a 100-word description is a 2013 SEO move that now hurts you.
• Don't reuse the same template across all products. Vary the structure, the opening, the length. Templated catalogs read as templated catalogs.
• Don't translate without localizing. A literal translation is worse than English in many markets.
• Don't skip the FAQ schema markup. If you generate a great FAQ with Prompt 7, wrap it in FAQPage JSON-LD on the product page — that's how it ends up in "People also ask."

Your catalog is a content asset

A product catalog is the most underrated content asset in ecommerce. Most stores treat it as a database — list the specs, ship the page, move on. The stores that beat them treat each product page as a small, helpful answer to a real shopper's real question.

AI is the lever that makes the second approach scalable. It is not the approach itself.

Use these prompts to draft. Edit every output. Vary the shape. Ground every claim in real data. Add the photo, the review, the founder's note, the trade-off — the things only you can add.

Then ship.