• Shopware Gold Partner & Shopify Partner

Chapter 7

Content that machines cite

Table of contents

A good paragraph on a product page should be understandable even if you read only that paragraph. With AI systems, this becomes more important. In search and source processing, they work with individual text passages and must be able to recognize which section answers a specific question. If you spread the actual statement across several paragraphs, you make it unnecessarily hard for machines.

Are longer texts cited more often?

A stubborn assumption is that longer texts are cited more often. The data clearly contradicts it. An Ahrefs evaluation of more than 174,000 pages from 560,000 AI Overviews (December 2025) found a correlation of 0.04 between word count and citation position – practically zero.

  • 0.04: correlation between word count and citation position.

  • 53.4% of all cited pages have fewer than 1,000 words.

  • 16% of citations go to pages with more than 2,000 words.

That does not mean short texts are automatically better. Other evaluations do find higher citation rates for very extensive content – but because it covers more sub-questions, not because it is longer. Writing longer for the sake of writing longer achieves nothing. For stores with limited content capacity, that is good news.

What counts instead?

The most important structural tip is simpler than any word count: answer the question first. Explanation, reasoning and details come after.

A section about the suitability of a running jacket should not start with three paragraphs about material development before the answer eventually arrives. Write first which conditions the jacket is made for – then explain why.

For a Shopware product page, this means: the first paragraph after the heading answers the central purchase question directly – what is the product suitable for, and for whom is it not – instead of starting with a brand story or a general category introduction.

What do headings look like that already carry an answer?

A subheading like "Details" or "Description" tells neither a person nor a machine much about what follows. "What water depth is this watch suitable for?", on the other hand, clearly assigns the following section to a specific question. The clearer this structure, the easier it is to match an individual passage to a user's question.

  • Instead of "Details": What water depth is this watch suitable for?

  • Instead of "Description": What this running jacket is suitable for – and what it is not

  • Instead of "Our espresso machines": Which espresso machine for which need?

  • Instead of "Our customers are very satisfied": Reduces drying time by 30 percent compared with the previous model

Which content is particularly interesting for purchase decisions?

A striking finding comes from an Ahrefs evaluation of 26,283 sources cited by ChatGPT: "best X" lists accounted for 43.8 percent of all cited sources – almost twice as much as the next most common page type. That does not mean a store should now mass-produce "The ten best…" articles. But it shows how well content can work that already structures a selection decision.

For a store, the more interesting question is therefore: where can we provide genuine decision support? "Which espresso machine for which need?" is more valuable for this than "Our espresso machines". That is exactly what good buying advice needs: naming differences, comparing products honestly and saying which variant suits whom.

Incidentally: a considerable share of this heavily cited comparison content does not come from stores themselves but from third parties – which makes this finding relevant for the external authority from Chapter 4 as well.

Why does specificity beat claims?

"Our customers are very satisfied" gives a machine nothing it can extract. "Reduces drying time by 30 percent compared with the previous model" does. The value lies less in the number itself than in the precision a number forces onto the text. Only if the number can be proven, of course. An invented number is not a GEO lever.

Does this contradict the schema recommendation from Chapter 5?

The Ahrefs evaluation of schema markup from Chapter 2 finds no measurable influence on citations. That does not contradict the recommendation from Chapter 5. Schema makes content machine-readable – it does not make it worth citing. Both are needed, but one does not replace the other.

Our assessment

The effort for good structure is lower than for more text. An existing product text where the answer is moved to the front and the subheadings are made more specific can be revised in twenty minutes. An additional 2,000-word guide costs a day – with no guarantee of achieving more through its length alone.

If you have content capacity, invest it in comparative advisory content. If you do not, revise the structure of what is already there. Both beat "write more".

What you can check in ten minutes today

  1. Open your most visited product page and read only the first paragraph, isolated from everything else. Does it already answer whom the product is suitable for?

  2. Check your subheadings: can you tell from them alone which question the section answers?

  3. Look for phrases like "high quality" or "particularly practical" – can each of them be replaced by a specific, verifiable statement?

Frequently asked questions

Length plays practically no role. An Ahrefs evaluation found a correlation of 0.04 between word count and citation position. More than half of the cited pages have fewer than 1,000 words. What matters is how many sub-questions a text answers clearly.

No. Written clearly for people is also easy for machines to evaluate: answer first, meaningful subheadings, specific and verifiable statements. Google explicitly rejects the idea of a special AI style, as Chapter 13 shows.