What is keyword stuffing (KEYWORD_STUFFING) and why does it reduce AI citations?

Keyword stuffing is the unnatural repetition of a word or phrase at a density that exceeds what fluent prose requires. In AEO, KEYWORD_STUFFING is an inverted signal: passages with high keyword density are cited less often, not more often. The inversion distinguishes KEYWORD_STUFFING from the other 7 flags in Seeklyn's rubric — where clearing a flag always raises the citability score, clearing KEYWORD_STUFFING removes a penalty caused by over-optimization rather than under-optimization.

The evidence grade for KEYWORD_STUFFING is weak — the only flag in the rubric at that grade. The inverted correlation (high density = less cited) is an observed pattern across content audits, but the causal mechanism is not peer-reviewed in the same way as structural flags like ANAPHORA_OPEN. The flag exists because the signal is consistent enough to warrant flagging, not because the relationship is settled science.

KEYWORD_STUFFING fires on passages of 20 or more words via 3 triggers: (a) any content word ≥4 chars appears in more than 8% of tokens — "content" appearing 9 times in 100 tokens is 9%, which fires; (b) unique-token ratio falls below 50% — fewer than half the tokens are distinct; or (c) a user-supplied target keyword appears in more than 6% of tokens. Passages under 20 words are exempt.

Triggers (a) and (b) fire on the passage's own content with no target keyword required — the flag can fire even when no keyword was set. Trigger (c) requires a user-supplied keyword and fires only when that keyword exceeds 6% of tokens. Any 1 trigger is sufficient to fire KEYWORD_STUFFING.

AI engines use the statistical patterns of fluent prose as a quality signal. Unnatural repetition is associated with low-quality, non-authoritative content. Engines trained on human writing recognize high-repetition passages as outliers and rank them lower as citation candidates.

KEYWORD_STUFFING carries a medium-severity weight in Seeklyn's 8-flag rubric — even though its evidence grade is weak, the observed pattern is consistent enough that the flag has meaningful impact on the citability score. Clearing it by diversifying language and naming specific entities produces a real score improvement.

Reduce the target keyword to below 6% of tokens — in a 100-word passage, that means the keyword appears 5 times at most, and ideally fewer. Use synonyms and related terms to maintain topical relevance without mechanically repeating the same word. Replace generic repeated nouns with specific named entities: instead of repeating "content" seven times, name "Seeklyn's passage analyzer," "Google AI Overviews," and "Perplexity" — each substitution adds entity variety while eliminating one repetition.

Check the unique-token ratio: if fewer than half the words in a passage are distinct, the prose is too repetitive regardless of whether a specific keyword is causing it. Vary sentence structure and vocabulary. A passage that clears KEYWORD_STUFFING — fewer than 8% density for any content word, more than 50% unique tokens, and target keyword under 6% — reads like natural informational writing, which is exactly the register AI engines favor.