Amazon indexes a listing from several fields, and it does not care how many times a word appears. Once a term is indexed, saying it again buys nothing and costs the space a sentence could have used to sell.
So the question is not how many keywords to add. It is which field each one belongs in.
The title. The two or three terms a shopper would actually type, in a sentence a person can read. Brand, product, deciding attribute.
The bullets. Secondary terms, carried inside sentences that are doing conversion work anyway. A bullet written to answer an objection will contain the vocabulary naturally.
The attributes. Size, material, colour, compatibility, intended use. These are structured fields and they decide which filters and browse nodes the listing appears under, which is a kind of relevance the words cannot buy.
The backend search terms. Synonyms, spelling variants, and the words shoppers use that you would never print on a listing. This is the one field where the rules are strict and widely misunderstood; see
how paid search works on Amazon.
A term indexed from the title does not get more indexed by appearing in every bullet. The listing is not scored on density.
What repetition does cost is readable space and, on the title specifically, click-through rate. A title stuffed to the character limit converts worse than a shorter one that reads like a product.
The exception is where the phrasing genuinely differs. Two ways of saying a thing are two terms, and both can be worth indexing.
The best source is your own
search terms report, because it is what shoppers typed before they bought from you. Terms converting there and missing from the listing are the ones to add first.
The second source is the reviews and questions, yours and your competitors'. People describe products in the words they searched with, and those words are frequently not the ones the brand uses.
Keyword tools are the third source, not the first. They are good for discovering variants and bad at knowing which of them converts for your product.
Competitor brand names, which is against policy and gets listings suppressed rather than ranked.
Terms that describe a product you are not selling, which buys traffic that bounces and pulls conversion down.
Subjective claims that trigger compliance review, and anything approaching a medical claim, which is a faster route to a suppressed listing than any ranking gain justifies.
Check the listing is indexed for the term, then watch impressions rather than rank. Impressions are the first thing to move; position and sales follow if the listing converts.
Give it two weeks, and change one field at a time, so the result is attributable. Rewriting the title, the bullets and the backend at once produces a number that cannot be read.
Advertising is how a new term gets tested before it goes anywhere near the listing. A term that converts in an exact match campaign has earned its place; one that spends without converting has not, whatever its search volume.
That loop runs in both directions, which is why the two jobs are best done by the same people. It is the subject of
the strategies built on those search terms, and the reason a keyword list is a working document rather than a deliverable.