Writing to be cited.

Why the way you write matters more

AI assistants have become a real source of traffic, even if it remains modest. According to Similarweb, AI platforms generated more than 1.13 billion visits to the world’s 1,000 most visited websites in June 2025, 357% more than a year earlier. Google Search sent 191 billion over the same period. The order of magnitude calls for measurement, the trend for attention.

The most important change lies elsewhere. A classic results page shows ten links. A generated answer cites only a few, and readers often stop there. Being chosen as a source becomes what matters, and that choice depends largely on the quality of the writing.

Starting with the foundations

Before any question of writing, content must be accessible. Google reminds us in its May 2026 guide: a page must be indexed and eligible for a snippet to appear in its generative features. Important content must be in the text of the page, and not only in an image or behind a complex interaction. A slow or poorly structured website starts at a disadvantage.

Internal linking also counts. Logical links between content on the same subject help engines understand the breadth of an expertise, and help readers go deeper.

The same guide dismisses several practices often presented as essential. There is no need to create an llms.txt file, to cut texts into small blocks for AIs or to add special markup. Structured data remains useful for rich results, but no markup is required to appear in AI Overviews or AI Mode.

Answer first, develop next

People now ask complete questions, phrased as they would ask a colleague. Effective content starts from these questions. The point is not to turn every subheading into a question, but to know precisely which question each section answers.

The answer-first technique is simple: under the subheading, one or two sentences that answer directly, then the development, the nuances and the examples. It serves the reader in a hurry as well as the systems looking for a clear phrasing to reuse. It is the principle of the inverted pyramid, well known to journalists.

Form follows substance. Short paragraphs that each develop one idea, lists when comparing or listing steps, tables when putting data side by side. Not as a matter of principle, but because these formats make information easier to read and to reuse without error.

Proving what you claim

Citable content makes verifiable claims. Precise figures with their source and date, explicit definitions, references to official texts. In 2023, researchers from Princeton and IIT Delhi tested several optimisation methods on generative engines. Adding quotations, statistics and sources was among the most effective approaches, with visibility gains of up to around 40% in their experiments. It is an experimental setting, not a guarantee, but the direction is consistent with what the engines themselves recommend.

Credibility also comes from the author. Google’s E-E-A-T principles (experience, expertise, authoritativeness, trustworthiness) guide the assessment of content quality. An article signed by an identifiable person, with their role and field, inspires more trust than an anonymous text.

Bringing what others do not have

Language models have read billions of pages that often repeat the same thing. An article that rephrases the top ten results brings nothing new to a generated answer. Google puts it clearly: it values non-commodity content.

In practice, this can be first-hand experience, data from your own activity, an anonymised case, a comparison you carried out yourself or a reasoned position. It is the material no assistant can produce on its own, and the main reason it will cite you.

Keeping it up to date, honestly

Outdated information weakens a whole article. Regularly updating reference content, refreshing figures and examples, maintains its value. The update date must then reflect a real change. Changing a date without touching the substance does not fool readers or engines for long, and erodes the very trust you are trying to build.

Frequently asked questions

How do you make content citable by an AI?

By answering a precise question clearly at the start of the section, backing up claims with verifiable figures and sources, and bringing your own information that cannot be found elsewhere.

Is structured data needed to appear in Google’s AI answers?

No. Google states that no specific markup is required for AI Overviews and AI Mode. Structured data remains useful for rich results.

Should you change the date of your articles regularly?

Only when the content really changes. Refreshing figures and examples is useful; changing the date without changing the substance brings nothing and can harm trust.