This is a template, not the training
Every training I run is tailored to the client: your products, your marketplaces, your markets and the people in the room. We work on your own listings and the questions shoppers ask about them, not on these generic examples.
Use this deck to see the approach. The live session, built on your listings, is a different thing.
● TOMEK SIKORA · PRACTICAL TRAINING
Marketplace content
for AI assistants
How Amazon’s Rufus and other shopping assistants read a listing, and how to write titles, bullets, A+, FAQ and attributes they can trust and quote. For e-commerce and marketplace content teams.
AmazonRufusListingsFAQA+Attributes
GEO · GENERATIVE ENGINE OPTIMISATION
FROM BEING FOUND TO BEING RECOMMENDED
Shoppers increasingly ask an assistant instead of typing keywords: Rufus / Alexa for Shopping on Amazon, retailer assistants, ChatGPT, Gemini. GEO means making product content easy for these assistants to read, trust and quote.
Shoppers now ask “which hand cream is best for cracked skin?”. Map the real questions per category. Keywords stay the base; questions set the structure.
Title and first bullets say what the product is, who it is for and which problem it solves. Benefit and use case first, brand story after.
Numbers, materials, skin type, compatibility. Assistants quote concrete facts and pass over adjectives like “premium” or “advanced”.
Short, self-contained answers to pre-purchase questions: suitability, usage, proof, compatibility. Often the biggest single gap on a listing.
Explain how products in the range differ and who each one is for. “Which one should I buy?” is exactly what assistants are asked to settle.
The same names, claims and attributes on Amazon, retailer sites and the brand site. Nothing invented; every statement traceable to approved data.
GEO builds on SEO, it does not replace it. Keywords still decide where a product ranks in Amazon search; structure and clarity decide whether the assistant recommends it. Both are worth tracking.
PART 01● HOW AN ASSISTANT READS A LISTING
Text first, images last, attributes as the cross-check
Reading order on a product page
Title → Bullets → A+ text → Attributes → Reviews → FAQ
Text inside images is read last, if at all. The assistant answers from these fields.
Most common gaps in listing audits: no FAQ at all, A+ content that is 100% images, attributes that contradict the copy.
- Two layers beat one: a claim in a bullet and an FAQ answer, worded differently, is picked up more often.
- Attributes cross-check the copy: a multipack with item count 1 or a product filed under the wrong type may be left out of the answer.
- Names must match: one spelling of brand and product in every field.
- Voice reads the first sentence: Alexa often reads only one or two sentences aloud.
- Completeness is rewarded: five full bullets, A+ text and an FAQ give the assistant more to work with.
PART 01● THE ASSISTANT TELLS YOU WHAT TO WRITE
Read the AI prompts on the product page
What to capture
- “Ask Rufus” question chips on your page and on 2–3 competitor pages — the assistant’s own signal of what shoppers want to know.
- Screenshot them with a date. They change.
- Outside Amazon: People Also Ask, AI Overviews and real prompts in ChatGPT or Perplexity.
- Note what the assistant gets wrong or leaves out — that gap is closed with a clear statement of fact.
| Category | Typical prompts |
|---|
| Beauty | Can it be used on sensitive skin? · Is the sunscreen water-resistant? |
| Bags | Does it fit a 15-inch laptop? · Is it waterproof? |
| Pet food | Is it grain-free? · Is it suitable for puppies? |
| Kitchen | Is it dishwasher-safe? · Does it work on induction? |
| Toys | What age is it for? · Does it need batteries? |
These questions are the best starting list for your FAQ and for the topics your bullets must cover.
PART 02● FIELD BY FIELD
Every field has a job
| Field | Its job | How to write it | Typical limit |
|---|
| Title | Be found and recognised at a glance | Brand + product/line + key attribute + who it’s for. Highest-value words first. No claims or promo words. | Often 75–200 characters |
| Bullets | Answer the main questions in 10 seconds | Five bullets, each with its own role. Short header + 1–2 sentences. | ~150–250 characters each |
| Description | Tell what bullets can’t hold | Open with the need or the product, then how it works. Key facts in the first ~500 characters. Never clone across variants. | Up to 2,000 characters |
| A+ content | Show and explain | Text modules next to images: a numbered how-to and a feature spotlight. Image-only claims repeated in text; alt text filled. | Module-dependent |
| Comparison chart | Help choose within the range | One master table per range, so a product carries the same data on every page. | Module-dependent |
| FAQ | Give assistants ready-made answers | 5 questions from the page’s AI prompts and review hesitations. Answer first, product named. | Q ≤120, A ≤250 characters |
| Backend search terms | Catch what doesn’t belong in visible copy | Synonyms, spelling variants, typos, other-language phrases. No title repeats, no competitor names. | Under 250 bytes |
| Attributes | Feed filters and cross-checks | Correct and consistent with the copy. Fix wrong values before writing anything else. | — |
Limits vary by category and marketplace and change over time. Check them at the start of every project.
PART 02● FIVE BULLETS, FIVE ROLES
One new piece of information per bullet
1
Identity & strongest claim
What it is, who it’s for, the main benefit with its proof.
2
How it works
The mechanism, feature or key ingredient.
3
Character & moment
The experience and occasions — only those the brand states.
4
How to use
Steps, amount, frequency.
5
Responsibility & range
Formulation and ethics claims, what’s in the box, where it sits in the range.
✕ KEYWORD BULLET“PREMIUM HAND CREAM – best hand cream dry hands cream for cracked hands moisturising hand cream for women and men”
✓ ROLE 1 · IDENTITY“FOR VERY DRY, CRACKED HANDS – This fragrance-free hand cream with 10% urea softens rough skin and helps restore the skin barrier, even with frequent hand washing.” (illustrative)
Adapt the roles to the category, and confirm the format (ALL-CAPS headers or not, length) against the local benchmark. None of the bullets opens with the brand name.
PART 02● KEYWORDS ACROSS A RANGE
One primary phrase per product, no cannibalisation
01
Reverse lookup
Run your own products plus 10–20 competitor products in a keyword tool to see which phrases drive their traffic and how much demand each has.
02
Seed expansion
Expand 5–10 seeds (category + key attribute). Clean out competitor names and attributes the product doesn’t have.
03
Assign per product
When several products share a phrase, give it to one: “vanilla body mist” to the vanilla variant. The others use secondary phrases.
Visible copy
Higher levels (product name, line, brand + category) in the title and bullets; broader phrases in the description and FAQ — always as natural sentences.
Backend only
Typos, spelling variants and other-language phrases. If a phrase can’t sit naturally in a sentence, it goes here.
No competitor names anywhere — not in the copy, not in the backend, not as “alternative to X”. Win those searches with descriptors instead.
GEO FOR DIGITAL ASSETS · IMAGES, A+, VIDEO, PRODUCT DATA
DIGITAL ASSETS: SEEN BY SHOPPERS, READ BY AI
Images, A+ modules and video still win the shopper’s attention, and they should stay. Assistants can read them too, but text is what they read most easily, so the safest setup is to have both say the same thing.
Images & infographics — Keep the claims on the image, in clear, legible type, and repeat the key ones in bullets or description. Fill in alt text / image keywords where the field exists.
A+ & enhanced content — Where the format allows, pair banners with live-text modules. One module answers one shopper question, with the benefit in the headline.
Comparison charts — Where possible, build them as real tables with attribute rows. Assistants draw on them for “which one is right for me?”
Video — Give it a descriptive title, and let the key claims said or shown on screen also appear in the page copy.
Product data & attributes — Complete every field: skin type, size, pack size, compatibility. Assistants match products to a request on these attributes.
AI can read images. Text simply makes it certain.
Reading a claim inside a graphic is possible for assistants, but it takes more computing power than reading plain text, so today it tends to come lower in their priorities.
Visuals are therefore worth optimising too. Repeating the key claims in text is the way to be sure they are picked up.
Build for consistency. No one knows how much weight assistants will give to visuals tomorrow. One set of claims carried across copy, images, enhanced content and retailer feeds works today and holds either way, on Amazon and beyond.
PART 03● THE FAQ MODULE
The most commonly missing piece on a listing
Five questions that cover
- Suitability — is it for me?
- Proof — does it work?
- Usage — how do I use it?
- Safety and ethics — is it safe, is it responsible?
- Choice — which one in the range?
How long does the battery last on these headphones?
The wireless headphones play for up to 30 hours with noise cancelling on. A 10-minute charge gives up to 3 hours of listening.
Is this single malt whisky smoky?
No. The 12-year-old single malt is unpeated, so it has no smoky character. Matured in sherry casks, it tastes of dried fruit and honey.
What age is the building set for?
The building set is for children aged 6 and over. It contains small parts, so keep it away from children under 3.
Don’t copy the platform’s system questions (returns, delivery). Every answer opens with the answer, names the product and adds one fact not worded the same way in a bullet. Examples are illustrative.
VISIBILITY & SEARCH · LISTING AUDIT
THE AI GAP IS CLEAR —
HERE’S HOW TO CLOSE IT
What a typical first AI-visibility audit of a marketplace brand shows (illustrative).
The fix lives in the listing text — and you can prove it before scaling.
THE GAP — IN THE LISTINGS
✔ From the listing audit: descriptions are thin for AI
✔ No FAQ, key claims locked in images, no comparison in words
✔ The brand is visible, but doesn’t win the recommendation vs its main rival
✔ Optimise the text layer on a subset of hero products
✔ Measure before / after visibility in Rufus & LLMs
✔ Check classic Amazon Search results too — not only AI
✔ Report the impact of every change
✔ Plan the next round of optimisations
✔ Stand up a recurring visibility tracker
OPTIMISE → MEASURE (RUFUS, LLMs + CLASSIC SEARCH) → REPORT → PLAN NEXT
PART 03● PRE-PUBLISH CHECKLIST
Before a listing goes live
- Every claim is on the fact sheet; nothing from reviews, competitors or another variant
- Questions from the “Ask Rufus” chips are answered somewhere in the copy
- Primary keyword in the title and bullet 1, in natural sentences
- Title opens with the brand and identifies the product, no claims or promo words
- Five bullets, five roles; none opens with the brand name
- Description not cloned from a sibling variant
- Every image-only fact repeated in text; alt text filled
- FAQ: 5 questions, answer first, within module limits
- Names identical in every field; attributes match the copy
- Local spelling, units and form of address checked by a native speaker
Contradictions and unconfirmed claims go back to the brand as open questions — never quietly resolved by the writer.
Who this session is for
E-commerce managers, marketplace content teams, copywriters and agencies that write Amazon and retailer listings. It works on Amazon and on other marketplaces with AI shopping assistants; the field names change, the logic does not.
It is one module in my SEO & AI Search training. Copywriters can go deeper in GEO & SEO for copywriters.
