AI helps you publish more. But more text doesn't make you more visible, neither to search engines nor to answer engines like ChatGPT. If you ramp up AI in content production, you first need a foundation that can carry it: your audience's real questions, claims backed by sources, and a review before publishing. Google classifies pages mass-produced without added value as spam, regardless of how they were created. For your strategy, in our assessment, that means: what counts is the quality of each individual passage, not the number of articles.
- Your own pages are the data AI systems use to build answers about your topic.
- Google penalizes volume without value, not AI as a tool.
- Our assessment: passages get cited, not whole articles. So every paragraph should stand on its own.
- Every number needs a source and a date in the same sentence.
- Revising existing articles often pays off more than writing new ones.
What is behind "Scaling Quality" at Dmexco 2026?
On September 18, 2026, the German trade magazine HORIZONT published the guest article "Scaling Intelligence beginnt mit Scaling Quality" (scaling intelligence starts with scaling quality). The author is Rasmus Giese of United Internet Media. The title sums up his thesis: if you want to scale intelligence, you first have to scale quality.
Giese speaks from the advertising market. We apply the idea to content strategy, and that is our reading: answer engines phrase their answers from what they find on websites. Your pages are therefore part of the data behind your topic. If they are thin, unsourced or contradictory, even the smartest AI won't make you visible there.
Why doesn't more AI content help automatically?
In our assessment, more AI content doesn't help automatically because answer engines select individual passages. This is how we understand the process: a language model searches many documents for the parts that fit the question and assembles an answer from them. In our guide How do I write an article that AI cites? we describe this in more detail: the source link points to the whole page, but in our observation it is the individual passage that gets processed.
So ten new articles that each answer a question halfway are worth less to you than one article that settles it. Worse: if several of your pages each half-cover the same question, they compete with each other. Scaling production without planning for this creates exactly these overlaps.
Where does Google draw the line?
Google defines scaled content abuse as generating many pages for the primary purpose of manipulating rankings rather than helping people. It covers large amounts of unoriginal content with little or no value, "no matter how it's created". As an example, Google explicitly names using generative AI to produce many pages without adding value for users.
Google's guidance on AI-generated content draws three practical consequences from this:
- Accuracy, quality and relevance should come first, especially for automatically generated content.
- This applies to metadata as well: title, meta description, structured data and image alt text.
- It can help readers to get context on how content was created, for example how automation was used.
So the line doesn't run between human and machine. It runs between pages with value and pages without it.
How do you recognize quality that AI systems cite?
Six criteria can be checked on any article, whether a human or an AI wrote it.
| Criterion | What it means | Example |
|---|---|---|
| Answer first | The key statement is in the first two to four sentences | "llms.txt is …" instead of "A lot has changed in recent years" |
| Paragraph stands alone | No back-references that depend on the paragraph before | "The Content Audit checks …" instead of "As mentioned above, it checks …" |
| Source in the sentence | Author and time period sit right next to the number | "According to Cloudflare Radar, the value in Q2 2026 was …" instead of "The share is falling sharply" |
| Verifiable claim | Every sentence could in principle be disproven | A concrete number instead of "the smartest solution on the market" |
| Text, not images | Numbers and tables are on the page as HTML text | A pricing table as a table, not as a graphic |
| One page per question | One page answers the question completely | One guide instead of three blog posts that touch on the same thing |
People benefit just as much. On our page What is good for humans, the machine understands too we explain why: humans scan headings, machines split the page into sections and judge each one separately. Tricks meant only for machines help neither.
In what order do you build a quality-first strategy?
Start before writing, not after. This order has proven itself:
- Collect real questions. Gather the questions your audience actually asks in search engines, forums and social networks. Assumed keywords are not enough.
- Check coverage. Match every question against your website: answered, partly answered or open. Watch for places where several pages compete for the same question.
- Improve what you have. For existing articles, three changes are usually enough: put an answer block at the top, phrase the headings as questions and give every number a source and a date. We plan for about an hour per article.
- Write new content only for the gaps. This is where AI can save the most time, because the question and the angle are already set.
- Review before publishing. Unreviewed AI texts carry two risks: wrong numbers or invented sources, and volume without editorial review. Facts, sources and an expert review therefore belong before the click on "Publish".
- Keep it current. Show the date of the last update visibly, in the text and in the structured data.
The most common mistake in practice is starting with step 4. Then volume appears before it is clear which questions are open.
How does flize support these steps?
flize covers the steps from the question to the review. Scouting collects questions from eight sources every week, including Google and Brave search queries, Reddit, gutefrage and the trade media of your industry. The Answer Check matches every question against your pages and names the passage down to the section. It rates coverage, not quality.
Quality is measured by the Content Audit. It gives existing editorial articles an article score from 0 to 100: structure counts for 50 per cent, substance for 30 and signals for 20. Every finding comes as a before-and-after text. In Content Production, flize flags every claim that no source confirms and checks whether the text really answers the original question. The draft stays a draft: you decide what gets published.
Does Google penalize texts written with AI?
+
No, not because of the tool. The spam policy targets many pages without value and applies equally to content created by machines and by hand. A reviewed, sourced AI text is not a violation. A thousand automatically generated pages with no value of their own can be.
How many articles should I publish per month?
+
There is no fixed number. The frequency depends on how many open questions your audience has and how many of them you can answer and review thoroughly. Only publish what has passed your review.
How long does an article have to be to get cited?
+
There is no minimum length. In our experience, a short article that answers one question completely is more likely to be cited than a long one that covers three questions halfway. If a topic gets too big, split it into two articles with one question each.
Where do I start if I have little time?
+
Start with your most important existing articles. Put a direct answer at the top and add a source and date to every number. That improves the pages that already have visitors and gives you the benchmark for everything you write from scratch.
