Visibility on the web is shifting. Users increasingly put their question to a chat interface and receive a summarised answer with a few references — instead of a list of ten blue links. Whoever is named as a source in these answers gains reach. Whoever writes well but structures poorly falls through.

The technical term for this is LLM readability: the readability of content for a large language model, i.e. for the systems behind chatbots and AI search functions. This does not mean simple text, but clearly structured text — one from which a single, self-contained statement can be extracted without having to read the rest of the page.

Why conventional text flow does not work here

Journalistic and promotional texts often work with a tension arc: first context, then example, the core statement comes in the fourth sentence or in the next paragraph. This works for human readers. A language model that extracts passages and condenses them into an answer works differently — it needs sections that stand on their own.

Concretely, this means: a paragraph that begins with „Genau deshalb ist es sinnvoll…“ loses its meaning as soon as it is viewed in isolation. References via pronouns, ironic turns of phrase, deferred punchlines and long preambles make reuse more difficult. A section whose first sentence contains the answer to a clearly formulated question, by contrast, is quotable.

What changes concretely at paragraph level

The change is less a question of keywords than of order and cut:

  • Core statement up front. The first sentence of a paragraph answers the question, the rest substantiates and differentiates.
  • One topic per paragraph. Where two thoughts have been pushed together, two paragraphs emerge.
  • Self-supporting formulations. Instead of „das“ and „dieser Ansatz“, the subject is written out once again.
  • Headings as questions or clear assertions. A subheading that names the content of the following section is an anchor — one that merely arouses curiosity is not.
  • Lists and tables for enumerable items. Criteria, steps and comparisons are more unambiguous in structured form than in running text.

The result reads more soberly. But it also reads faster — which is rather an advantage for impatient human readers. The often-feared conflict between writing „for machines“ and „for people“ is smaller for factual specialist content than it seems.

Where the CMS has to play along

Structure does not arise solely in the minds of the editorial team, but in the tools available to them. An editorial system that essentially offers one large free-text field produces unstructured content — regardless of how well the author writes.

In TYPO3 this can be shaped. Instead of a single rich-text field per page, dedicated content elements for question-and-answer blocks, definitions, step-by-step instructions and fact boxes suggest themselves — each with clearly separated fields for heading, core statement and explanation. The advantage is twofold: the markup in the delivered HTML becomes correct and consistent, and the structure is no longer a matter of negotiation for each individual article.

This includes clean semantics in the markup, i.e. the technically correct markup of meaning: a correct heading hierarchy without jumps, real lists instead of paragraphs begun with hyphens, headings not as bold-formatted text. Such errors are invisible to the eye and disruptive for automated processing.

Editorial consequences

Anyone who takes this path changes workflows, not just texts. Experience shows that three points are decisive:

  1. Collect questions before text is created. Which formulations do customers really use when they describe a problem? These questions become headings.
  2. Rebuild existing content instead of reinventing it. Many good pages contain the right statements in the wrong place. A reorganisation is cheaper than a new text.
  3. Include structure in approval. A short check — is the core statement up front, does each paragraph stand on its own, does the heading logic fit — prevents the discipline from disappearing after three months.

Realistic expectations

Two limitations belong to this. First, it cannot be verified by which criteria individual AI systems select sources; providers do not disclose this and change it. Structural work increases the chance, it guarantees nothing. Second, form does not replace substance: a clearly structured text without its own information does not become a source, but only a better-sorted repetition.

It therefore makes sense to treat LLM readability as a craft and not as a channel. It is about the same work that also makes accessibility, translatability and reuse of content easier.

What this means for web and software projects

Content structure thus becomes a requirement that belongs in the conception phase — on a level with performance and accessibility, not in the rework. For relaunch projects this means: define content types, fields and heading logic early, instead of later breaking up a free-text field after the fact.

Anyone who sets up their CMS so that structure is the standard path and not the exception will not have to have this discussion again the next time search interfaces change.

Sources