The premise of the question comes from an era that ended. There was a period when repeating a phrase at a target density, stuffing variations into the text, and writing to a word count produced results, because the systems were matching strings rather than understanding meaning. Those techniques now produce material that reads badly and is identified as thin.
What replaced them is closer to ordinary good writing with a few structural habits. Answer the question in the first two sentences rather than building to it. Use headings that describe what follows. Keep paragraphs short. Cover the subject thoroughly enough that somebody leaves with the answer. All of those serve readers, and they happen to be what the systems reward.
The one genuine adjustment is vocabulary, and it is not a compromise. If customers say pipe leaking under sink and your industry says supply line failure, the page should contain both, with the customer phrasing in the heading. That is not writing for a machine. It is writing in the language of the person you are trying to reach, which is what you should have been doing anyway.
Specificity is the other overlap, and it is the largest one. Numbers, examples, and detail that could only come from somebody who does the work are what distinguish a page from the generic version of the same subject. They are also what makes writing worth reading, which is why the two goals converge rather than competing.
The tension people feel is usually between writing for customers and writing for peers, rather than between customers and search engines. Material aimed at impressing others in your field uses terminology customers do not search, addresses concerns they do not have, and demonstrates depth on subjects they did not ask about. That is the actual failure, and it gets misdiagnosed.
Titles and descriptions are where a small amount of deliberate optimisation still pays, since those determine whether anybody clicks. Writing them to state what the reader will learn, in their words, within the length that displays, is the highest return editing available and it takes ten minutes.
Be careful about the advice that survives from the previous era, since a great deal of it is still circulating. Keyword density, exact match phrasing repeated unnaturally, and minimum word counts all belong to systems that no longer exist, and following them now actively harms a page.
Then judge it by whether a customer would find it useful if they arrived by any route. A page that would serve somebody who received the link directly is a page that will serve somebody who found it through a search, and the reverse is not reliably true.
Check how a page reads aloud before publishing, since anything written to accommodate a phrase rather than a reader is audible immediately. That test catches the residue of optimisation thinking faster than any analysis.
Update rather than republish when a page ages, since the existing address has accumulated whatever visibility it has and a new one starts from nothing.
Check what actually brings people to each page after a few months, since the queries are frequently different from what you targeted and the page can be improved to match what it is already attracting.
Write the title and description for the reader rather than for a phrase, since those appear in the results and determine whether anybody clicks at all.