The density rule, and why it survives
“One to two percent keyword density” appears in a surprising number of SEO guides. The figure comes from an era when engines really did count words, and it survives because it has two convenient properties: it can be taught, and it can be checked. Neither makes it right.
Google has said repeatedly that there is no keyword density target. Writing towards a percentage reliably produces sentences like “As a roofer in Lüneburg, your Lüneburg roofer offers all roofer services”. Every reader notices, and readers leaving again is a signal Google very much does see.
One side effect costs more than the rule itself: it points attention at a number while the decision is made somewhere else entirely.
What counts instead: speaking the same language
Businesses describe their work in trade language and customers search in everyday language. A roofer writes “roof renovation and pitched-roof sealing”, the customer types “roof leaking what to do” or “roofer Lüneburg cost”. Both mean the same thing, and only one of them gets typed into a search box.
That is the actual work: find out which words people use, then use them naturally. Not more often, but at all.
The four places where a word really works
If you only touch four places, make it these, in this order:
| Place | Why it counts |
|---|---|
| Page title | It appears in the search result and is the strongest single signal of what the page is about |
| The H1, the visible heading | Tells reader and machine in the same moment whether they are in the right place |
| First paragraph | Search engines and assistants draw on it as the answer more often than on anything below |
| The subheading above the answer | Makes the section that actually answers the question quotable on its own |
Everything else is polish. Rewriting the URL of an existing page just to fit a word in rarely pays and costs you the address you have built up.
German works differently: inflection and compounds
This is the part most guides skip, because they are translated from English. German inflects and compounds: Dachdecker, Dachdeckers, Dachdeckern, Dachdeckermeister, Dachdeckerbetrieb, Dachdeckerinnung. Search engines map these forms onto each other, so they do not need the exact spelling.
In practice that means you are allowed to write normally. Forcing the nominative in everywhere (“We are your Dachdecker Lüneburg for Dachdecker services”) makes the text worse for nothing in return. And compounds are often the more natural form in German: “Dachsanierung” reads better than “Sanierung von Dächern” and covers the same search.
One exception is worth naming: with place names the separated form is the usual one and gets searched that way, so “Dachdecker in Lüneburg” rather than “Lüneburger Dachdecker”. Using both is fine, one in the title and one in the body.
How to find out which words those are
No paid tool needed. Four sources, all free:
Google's own suggestions. Type your service and read what Google completes. Those are real inputs from real people. Same for “related searches” at the foot of the results page and the question boxes.
Search Console. The Performance report shows which queries already bring your pages impressions. Very often they are not the ones you wrote for, and those are exactly the ones to serve properly.
Your customers. How do callers describe the problem before you translate it into trade language? Those sentences are the search queries, word for word.
Forums and question sites. Someone asking about their problem in a forum phrases it exactly the way they would type it into a search.
The test for too much
Read the paragraph out loud. If you would never say the sentence that way, the word is in there too often. That test beats any counting tool, because it measures the thing that actually matters.
A second check if you are unsure: if the term is in every single subheading, it is too much. Subheadings are there to structure the text, not to repeat one term six times.
What changed with AI answers
An assistant quotes a paragraph, not a page. That moves the decisive question from “how often does the word appear” to “does one paragraph answer the question completely”. A paragraph that stands on its own gets picked up. One that refers back to three screens above it does not.
It is also why the density rule carries even less weight now than before: it describes a property of a whole page, while the selection happens at paragraph level. More on that in our article on visibility in ChatGPT and Perplexity.
Frequently asked questions
How often should a keyword appear in the text?
There is no number to write towards. Google has said repeatedly that there is no keyword density target. If the term is in the title, the H1 and the first paragraph, and the text then reads normally, it is covered.
Does repeating a word too often hurt?
Past a point yes, though less as a penalty than as an effect: the text reads mechanical and people leave. Google sees them leaving. It does not see a percentage.
Does the keyword have to appear exactly as typed?
In German it does not. Search engines handle inflected forms and compounds, so Dachdecker, Dachdeckers and Dachdeckerbetrieb belong together as far as they are concerned. Write the form a person would say.
Should keywords go in the alt text?
Alt text describes the image for someone who cannot see it. If the term fits naturally, fine. As a place to park search terms it fails, and it is unfair to people using a screen reader.
Does any of this still count for AI answers?
The question shifts there. An assistant quotes a paragraph, not a page. What decides it is whether one paragraph answers the question completely, rather than how often a word appears.