# Product page SEO: writing for Google and AI

> A product page that ranks answers shoppers' questions instead of telling a pretty story. Our two-layer method, measured on a fashion store: +78% visits from organic search.

- Original page: https://dotsland.com/en/product-page-seo-ai/
- Language: en
- Published: 2026-10-31T09:00:00+01:00
- Updated: 2026-10-30T19:35:53+01:00
- Section: AI
- Publisher: Dotsland, https://dotsland.com/en/

---

## In brief

- A product page has two readers: the shopper who is hesitating, and the engines, Google and the AI assistants alike, that compare. Pretty sentences serve neither.
- On an online fashion store we follow, visits from organic search rose by 78% from July to September 2026 compared with 2025, after its product descriptions were rewritten at the end of June 2026.
- Our method has two layers: a short description in bullet points with the facts (sizes, shapes, colour, material, size equivalents), then a long description written to answer search intent, folded into an accordion.
- For Google, optimising for AI is still SEO. There is no magic markup and no miracle file: what works is pages that are useful, precise and up to date.
- It is never a one-off job. You analyse, optimise and start again, in regular rounds.

Product page SEO that works in 2026 answers the precise questions shoppers type into Google or ask an AI, instead of telling a pretty story. On an online fashion store we follow, visits from organic search rose by 78% from July to September 2026 compared with 2025, after its product descriptions were rewritten at the end of June 2026. Here is what was changed, how to do the same on your catalogue, and what not to expect.

> **How we checked**  
> The store’s figures were taken on 30 October 2026 from its Google Analytics 4 property (“Organic Search” sessions, month by month, from October 2023 to October 2026) and from its Search Console (clicks, year-on-year comparisons, searches with and without the brand name). Google’s recommendations come from its official documentation, read the same day. Sources are listed at the end of the article.

## Why doesn’t a beautiful product page rank?

It is the mistake we see most often among the online retailers we work with: descriptions written to look pretty, not to answer. To borrow the example from the French fashion federation’s magazine *Flair*, “Our timeless take on the summer silhouette” tells nothing to the shopper hesitating between two sizes, and nothing to the engine looking for a midi-length linen dress for an outdoor wedding.

The difficulty is real: shoppers want to read little, and the page has to carry a lot of information. Both extremes miss. On one side, the short, elegant, empty sentence. On the other, Amazon’s pages, packed with keywords: there it makes sense, because Amazon is a search engine in its own right and its listings have to rank in its own search. But copying that model on a brand’s own store produces pages nobody reads.

Then there is the copy problem. The same brand often has different descriptions on its own site and on the marketplaces where it sells. When the text is identical everywhere, it is duplicate content, and the brand’s own store is rarely the one that benefits.

Finally, a product page is first a technical matter: the description has to be visible. A few years ago, some sites hid a block of keywords written in the same colour as the page background. Google explicitly lists this technique as spam (“white text on a white background”). Text folded into a tab or an accordion, on the other hand, is visible and read by crawlers, and Google lists it among the uses that comply with its rules: that is what we use.

## What does GEO change for a product page?

GEO, short for *Generative Engine Optimization*, is the work of getting cited in answers written by an AI: Google’s AI Overviews and AI Mode, ChatGPT, Gemini. Three things change for a product page.

1. **Questions are longer.** The shopper no longer types “long boho dress”; she asks for a flowing midi dress for an outdoor wedding, within a budget, in a natural fabric, delivered before a given date.
2. **The AI breaks the question down.** Google explains it itself: its AI features run several related searches to build an answer (the technique is called *query fan-out*). Each criterion in the question becomes a search, and your page has to be able to answer each one.
3. **Being cited becomes visibility in its own right.** On the store we follow, Search Console counted 86,900 impressions of its pages in Google’s AI answers between 26 June and 27 October 2026.

Does that mean writing “for AI”? No. Google says it plainly in its guide to generative AI: optimising for its AI answers is SEO. It adds that there is no need to rewrite content specifically for AI, that there is no ideal length, and that no special markup is required. The French women’s ready-to-wear federation sums up the same point in its magazine *Flair* (our translation): “precision becomes the new sensoriality”. An exact sentence serves the shopper and the engine; vagueness serves nobody.

On this store, pages are also described in an llms.txt file, since moved to its second version. Google writes that its search does not use it: the file is for other assistants. We covered it in [what llms.txt v2 changes](https://dotsland.com/en/llms-txt-v2-what-changes/). It is a complement, not the core of the work.

## How do you structure product page SEO in two layers?

Our method separates what must be seen straight away from what can be unfolded.

**First layer: the short description, in bullet points.** This is where we put the facts that answer precise questions: sizes, shapes, colour, material and size equivalents. The shopper sees them without scrolling, and the engine finds them without searching.

**Second layer: the long description, in an accordion.** This is the written part, where we work on SEO and GEO: uses, occasions, the questions shoppers ask. Folded, it does not turn the page into four or five screens of text.

Google Merchant Center, which feeds Shopping ads, points the same way. It accepts up to 5,000 characters of description, but recommends putting the most important information in the first 160 to 500. Even when there is room, the essentials come first.

| Layer | What it contains | Who it serves first |
| --- | --- | --- |
| **Short description, visible** | Sizes, shapes, colour, material, size equivalents, as bullet points | The shopper who wants a precise answer, and the engine that compares |
| **Long description, in an accordion** | Uses, occasions, answers to common questions, written out | Search intent, in SEO as in GEO |

## How do you find your customers’ search intent?

We combine several sources: Google Trends to see what is rising and in which season, AnswerThePublic for the questions phrased around a word, DataForSEO or Semrush for volumes, rankings and the keywords to climb on. Any tool that gives rankings and keywords will do; what matters is combining them.

The most valuable source is still your own Search Console, because it shows the questions that actually led to your pages. In the performance report, add a “Query” filter, choose “Custom (regex)”, and paste the expression below. It keeps only searches that start with a question word, in French, English, Spanish, Italian, Portuguese, German and Polish.

> **The expression we use**  
> `^(qui|que|quoi|quel|quelle|quels|quelles|où|quand|pourquoi|comment|combien|est|sont|était|étaient|peut|peux|pouvez|doit|dois|doivent|faut|si|who|what|where|when|why|how|which|was|were|did|do|does|is|are|can|could|should|would|will|if|quién|quiénes|qué|cuál|cuáles|dónde|cuándo|por qué|cómo|cuánto|cuánta|cuántos|cuántas|es|son|era|eran|puede|puedo|pueden|debe|debo|deben|hay|chi|che|cosa|quale|quali|dove|quando|perché|come|quanto|quanta|quanti|quante|è|sono|erano|può|posso|possono|deve|devo|devono|c'è|ci sono|quem|o que|qual|quais|onde|por que|porque|como|quantos|quantas|é|são|eram|pode|podem|devem|há|wer|was|wo|wohin|woher|wann|warum|wieso|weshalb|wie|welche|welcher|welches|welchen|wem|wen|ist|sind|war|waren|kann|können|könnte|könnten|soll|sollen|muss|müssen|darf|dürfen|hat|haben|wenn|ob|kto|co|gdzie|dokąd|skąd|kiedy|dlaczego|czemu|jak|jaki|jaka|jakie|jacy|który|która|które|którzy|ile|czy|jest|są|był|była|było|byli|może|mogę|mogą|powinien|powinna|powinno|muszę|musi|muszą)([ ?'’-]|$)`

Two notes before you use it. The expression does not end with the usual word boundary `\b`, which does not recognise accented letters in Search Console: on the store we follow, “où” followed by `\b` found no questions at all, while the same start followed by a space returned sixteen. And some short words, such as “si”, also bring up searches that are not questions: read the list, do not rely on the total alone.

On the store we follow, this filter filled the table to its limit of a thousand rows between 26 June and 27 October 2026. The questions that brought the most clicks nearly all said the same thing, in several languages: “where is this dress from”, “where to buy”, “onde comprar”, “wo kann ich es kaufen”, “dove acquistare”, and in French “où trouver cette robe”. People had seen a dress and were looking for where to buy it. A page that names the product, its material and its cut clearly helps answer that; a poetic page, much less.

For the writing itself, you should use AI tools today: they make the work much faster. On one condition, which we apply: the AI writes from the product’s real facts and from written rules (what must never be invented, the words the brand refuses), and a person proofreads. Google accepts AI-assisted content when it is useful; its spam policies target pages generated in bulk to manipulate rankings.

## What results to expect, and how soon?

On the fashion store we follow, the work was done in two stages. In spring 2026, the technical part: category titles and descriptions, category URLs, image alt texts, product names. At the end of June 2026, the rewrite of every product description, a little over two hundred pages in six languages, to answer search intent rather than to produce pretty copy.

| July to September | Visits from organic search | Change |
| --- | --- | --- |
| **2024** | 19,340 |   |
| **2025** | 20,197 | +4% on 2024 |
| **2026, after the rewrite** | 35,887 | +78% on 2025 |

The contrast with the start of the year is stark: from January to April 2026, the same visits were down 51% on 2025. The recovery began in May, after the technical work, then accelerated after the product pages were rewritten. In Search Console, on searches that do not contain the brand name (people looking for a product, not for the store), clicks rose by 49% from July to September 2026 and by 85% from 1 to 27 October, compared with the same periods of 2025.

These figures show a correlation, not proof. Seasonality, cookie consent, Google updates and the cut in paid ads in spring 2026 all play a part too. That is why we compare year on year, which neutralises seasonality, and isolate non-brand searches, which the cut in paid ads could not inflate.

What we always tell our clients: there is no single truth, and this is not a one-off job. Google does not reward change for its own sake; it values content that is useful and up to date. So do not expect immediate results: correct and optimise regularly, and give it time.

## Four things to do on your product pages tomorrow morning

1. **Pick the pages.** Start with your best sellers and the pages losing rankings: your SEO tracking tool or Search Console will show you which. A short list beats the whole catalogue at once.
2. **Write the short description as bullet points.** On those pages, put sizes, shapes, colour, material and size equivalents at the top, visible straight away.
3. **Rewrite the long description for search intent.** Find out what your customers type (Google Trends, AnswerThePublic, DataForSEO or Semrush, and the expression above in Search Console), answer it in a written section folded into an accordion. Nothing hidden, no keyword stuffing.
4. **Measure, then start again.** Note the date of the change, track non-brand clicks in Search Console, and do not judge before several months have passed. Then launch the next round: on this store, roughly one per quarter.

## The bigger picture

Product pages will never be enough on their own. An online store’s success always comes from several levers lining up: pages that answer, blog articles that cover the wider questions, and links earned from other sites. None of these levers makes up for the absence of the others.

This work is not heavy when it is regular. You analyse, optimise, act, then analyse again. A rank tracking tool, such as [SEO Hub](https://seohub.datafirefly.com/), which we use, flags the pages to rework and the rankings lost; the method does the rest. If you want to know which pages to start with, [send us your store’s address](https://dotsland.com/en/contact-us/): it is the work of our [search and acquisition expertise](https://dotsland.com/en/expertise/acquisition-search-advertising-social/) and our [e-commerce expertise](https://dotsland.com/en/expertise/ecommerce-and-marketplaces/).

## Sources

1. Google Search Central, “Google’s Guide to Optimizing for Generative AI Features on Google Search”. [Read](https://developers.google.com/search/docs/fundamentals/ai-optimization-guide)
2. Google Search Central, “AI Features and Your Website”. [Read](https://developers.google.com/search/docs/appearance/ai-features)
3. Google Search Central, “Creating Helpful, Reliable, People-First Content”. [Read](https://developers.google.com/search/docs/fundamentals/creating-helpful-content)
4. Google Merchant Center Help, “Description \[description\] and structured description \[structured_description\]”. [Read](https://support.google.com/merchants/answer/6324468)
5. Fédération Française du Prêt à Porter Féminin, *Flair* no. 3, 2026, “GEO : quand l’IA fait le shopping à votre place”, pp. 26-27 (in French).
6. Google Search Central, “Spam Policies for Google Web Search”, sections “Hidden text and link abuse” and “Scaled content abuse”. [Read](https://developers.google.com/search/docs/essentials/spam-policies)
7. Aggarwal et al., “GEO: Generative Engine Optimization”, KDD 2024. [Read](https://arxiv.org/abs/2311.09735)

## FAQ

### How long should a product description be?

There is no ideal length; Google says so itself. What matters is the order: the facts first, as visible bullet points (sizes, shapes, colour, material, size equivalents), then a written section, folded into an accordion, that answers search intent. For the Merchant Center feed, Google accepts 5,000 characters and recommends putting the essentials in the first 160 to 500.

### Should product pages be written differently for AI?

No. For Google, optimising for its AI answers is SEO: pages that are useful, precise, up to date and easy to read. What changes is the questions, which are longer and more precise. A page that answers those questions clearly serves both classic search and AI answers.

### Does Google take into account text folded into a tab or an accordion?

Folded text stays visible to the shopper who opens it and readable by crawlers: it is the method we use for the long description. Google even lists it among the uses that comply with its rules. What to avoid is the old hidden-text technique, keywords written in the background colour, which Google classes as spam.

### Can Dotsland help me rework my product pages?

Yes. We start from your data: the pages that sell, those losing rankings, and the questions that already lead to your site in Search Console. Then we rewrite in rounds, and we measure. [Contact us](https://dotsland.com/en/contact-us/) with your store’s address.