Should a product description tell a story or list facts? For years the answer depended on who you asked: copywriters wanted narrative, SEO people wanted keywords and specs. AI shopping assistants have settled a large part of that argument — but not in the way either side expected.
I've worked on product content for hundreds of listings across haircare, spirits, eyewear and consumer health, on brand sites and on marketplaces. This is what I've learned about writing product descriptions that AI can understand and recommend, without turning them into spec sheets nobody wants to read.
How AI reads a product page
An AI assistant doesn't read your description the way a customer does. It extracts facts to answer a question: “Is this shampoo safe for colour-treated hair?”, “Which of these is better for a small kitchen?”. It then checks those facts against other parts of the page and other sources.
- Text beats images. A fact that appears only in an image or infographic is effectively invisible to an assistant.
- Structured fields are trusted most. Attributes, specs and filters are used to cross-check what the text says. A wrong attribute undermines a correct description.
- Consistency is rewarded. The same claim in the title, bullets, description and FAQ gives the model confidence. Contradictions make it hedge or pick a competitor.
- Specifics win. “Reduces frizz for up to 72 hours” is quotable. “Beautiful, manageable hair” is not.
Structured vs narrative: what each one does
Two layers of a product description
| Structured layer | Narrative layer | |
|---|---|---|
| Form | Attributes, specs, bullets, FAQ | Paragraphs, story, tone of voice |
| Main reader | AI assistants, filters, comparison | The customer deciding to buy |
| What it answers | “Does it fit my need?” | “Why this one, and why this brand?” |
| Typical mistake | Missing, incomplete or wrong fields | Adjectives with no facts inside |
It isn't structured or narrative. The structured layer gets you into the answer. The narrative layer wins the click and the sale. A narrative without facts is invisible to AI; a spec sheet without a story doesn't sell.
A description structure that works for both
Across categories, the same three-part structure keeps performing — on brand sites and on marketplaces:
- The need or product type first. Open with the problem it solves or what the product is, not with the brand name. That's the sentence AI is most likely to quote.
- The key facts early. Ingredients or materials, the main result with a number, who it's for — within the first few sentences. Specific data placed early measurably increases how confidently an assistant uses your page.
- Then the story. Origin, process, brand voice, sensory detail. This is where tone matters, and where you keep the page from sounding like every competitor.
Around the description, the structured layer does the rest: complete attributes, five bullets that each carry one new fact, and a short FAQ that answers the questions buyers actually ask.
Before and after
Example · generic moisturising shampoo
Before — “Discover the secret to beautiful hair. Our luxurious formula pampers your hair with nature's finest ingredients for a salon-worthy finish every day.”
After — “A moisturising shampoo for dry, frizz-prone hair. Its formula with coconut oil and glycerin reduces frizz for up to 48 hours and is safe for colour-treated hair. Gentle enough for daily use, it leaves hair soft without weighing it down — the everyday step in a routine built around hydration.”
The second version still sounds like a brand. But every sentence answers a question someone could ask an AI assistant.
Rules I apply to every description
Product description · do and don't
Do
- Put facts in text, not only in images
- Quantify claims you can support
- Say who it's for — and who it isn't for
- Use one spelling of brand and product names everywhere
- Write a different description for every variant
Don't
- Open with the brand name or a slogan
- Invent facts to sound more convincing
- Copy the same text across variants
- Hide key information in PDFs or tabs AI can't read
- Let attributes contradict the text
The one rule above all others: change the form, never the facts. Optimisation reorganises and clarifies what's true about the product. It never adds claims the product can't back up — especially in regulated categories like cosmetics, supplements or medicines.
If you want your product pages reviewed this way, it's part of my e-commerce SEO and AI shopping assistant work. For Amazon specifically, see Amazon content for SEO and AI search.
Based on product-content audits I have run for brands in haircare, spirits, eyewear and consumer health; the example is generic.
FAQ
Frequently asked questions
Both. The structured layer — attributes, specs, bullets, FAQ — is what AI assistants extract and trust. The narrative layer persuades the customer. A narrative with no facts is invisible to AI; a spec sheet with no story rarely sells.
Treat it as if it can't. Facts that appear only in images, infographics or A+ graphics are effectively invisible to AI shopping assistants. Every important fact should also be in the page text.
Long enough to hold the need, the key facts and a short story — usually a few short paragraphs. On marketplaces a compact description of a few hundred characters with facts in the first sentences works well; on your own site you can go deeper.
Yes. Cloned descriptions across variants give AI no way to tell them apart, so it can't recommend the right one. Each variant needs at least the facts that make it different.
No. Structured data helps machines read attributes, but AI assistants still quote the visible text. Both need to say the same thing.
No. Optimisation changes the form, never the facts. It reorganises and clarifies what's true about the product and never adds claims the product can't back up.
