Content analyzer
What is passage extractability (the 8 flags)?
Passage extractability is Seeklyn's 8-flag rubric for measuring whether an AI engine can lift a passage verbatim as a self-contained answer. The 8 flags split into 2 strong-evidence structural flags (ANAPHORA_OPEN, NO_ANSWER_FIRST), 5 moderate-evidence quality flags, and 1 weak-evidence signal (KEYWORD_STUFFING). Clearing all 8 puts a passage at or near 100/100 on Seeklyn's citability scale.
Each passage card in a Seeklyn content report shows fired flag chips with evidence grades and one-line fixes — run your page free at /dashboard/content.
What is extractability?
Extractability is a passage's fitness to be lifted verbatim by an AI answer engine. A high-extractability passage answers its question in sentence 1, contains no pronoun openers requiring prior context, states at least 1 concrete fact, and names the entities it is about. A passage failing any of these dimensions is harder to quote cleanly — the extracted text either loses meaning out of context or contains nothing specific enough to cite.
Seeklyn operationalizes extractability as a lint pass over every passage, firing 1 or more of 8 coded flags where the rubric is violated. Each flag carries an evidence grade (strong, moderate, or weak) reflecting how reliably that signal predicts reduced citation, per the research literature (Princeton GEO, KDD 2024). The 8 flags span 3 dimensions: structural independence, content quality, and signal integrity.
What are the 8 extractability flags?
The 8 flags are defined in Seeklyn's frozen rubric (`flags.go`) and applied in the order shown below. The 2 strong-evidence flags directly break self-containment and carry the highest citability penalty. The 5 moderate-evidence flags each reduce citability without voiding self-containment status. The 1 weak-evidence flag (KEYWORD_STUFFING) is an inverted signal — high repetition correlates negatively with citation — and is shown as a warning rather than a structural failure.
| Flag code | What triggers it | Evidence grade | One-line fix |
|---|---|---|---|
| ANAPHORA_OPEN | A sentence opens with a pronoun or anaphora ("This", "It", "They", "Such", "These", "Those", "As mentioned") that requires prior context | Strong | Restate the subject so the sentence stands alone |
| NO_ANSWER_FIRST | The first sentence is not a direct declarative answer — it is a question, opens with a hedge or pronoun, or contains no fact token (digit, percent, unit) | Strong | Lead with a direct declarative answer capsule containing a concrete number or entity |
| HEDGE | The sentence contains epistemic qualifiers: "may", "might", "could", "possibly", "arguably", or "perhaps" | Moderate | State the claim declaratively where the evidence supports it |
| RUN_ON | A sentence exceeds 30 words | Moderate | Split into shorter, single-claim sentences |
| LOW_FACT_DENSITY | Fewer than 1 fact token (digit, date, unit, percent) per 100 words in the passage | Moderate | Add concrete figures, dates, or named entities |
| NO_ATTRIBUTION | The passage contains a numeric claim but no attributable source (named study, year, "according to", etc.) | Moderate | Attribute the figure (e.g. "according to Ahrefs, 2026" or "per Princeton GEO, KDD 2024") |
| ENTITY_SPARSE | A passage of 50+ words names fewer than 2 distinct entities (people, products, places, or organizations) | Moderate | Name the specific entities this passage is about — brands, tools, standards, people |
| KEYWORD_STUFFING | Keyword density is unnaturally high: most-repeated token > 8% of passage, overall repetition > 50%, or explicit target keyword > 6% of passage | Weak | Reduce repetition; density correlates negatively with citation |
How do the 2 strong-evidence flags differ from the other 6?
ANAPHORA_OPEN and NO_ANSWER_FIRST are the 2 structural flags that break a passage's ability to stand alone. Seeklyn's `isSelfContained` function returns false if either fires. Clearing 1 of these 2 flags is the highest-ROI edit on a failing passage.
6 remaining flags reduce the 0–100 citability score without voiding self-containment. A passage with HEDGE and NO_ATTRIBUTION might score 55/100 but still be self-contained. Strong-evidence grading on the 2 structural flags reflects the research consensus (Princeton GEO, KDD 2024): passages opening with a pronoun or burying the answer lose citation consideration at the retrieval-reranking step.
What does evidence grade mean for each flag?
Evidence grade reflects how reliably Seeklyn's engineering team and the underlying AEO research (Princeton GEO, KDD 2024) can trace a flag to reduced citation probability. Strong means the causal mechanism is well-understood and directly implemented: an anaphoric opener provably breaks standalone extraction; a missing answer-first sentence provably reduces retrieval relevance. Moderate means the signal is well-motivated and directionally correct but harder to isolate in controlled experiments. Weak means the signal is real (keyword stuffing does correlate negatively with citation) but the threshold is approximate and content-dependent.
Seeklyn surfaces these grades in the flag chips on every passage card — not to excuse weak-evidence flags, but to help you prioritize. A strong-evidence flag at 1 sentence costs much less to fix than 5 moderate-evidence flags across the passage, and fixing it clears the self-contained gate entirely.
Frequently asked questions
Do all 8 flags fire on every passage?
No — flags only fire when the specific condition is met. A well-structured 150-word passage with a direct answer-first opener, 2 named entities, a cited statistic, and no hedging will fire zero flags and score near 100/100. Most real-world passages fire 1–3 flags.
Which flag should I fix first?
Always fix ANAPHORA_OPEN or NO_ANSWER_FIRST first — these are the 2 strong-evidence structural flags that break self-containment. Clearing either one raises citability more than fixing multiple moderate-evidence flags and may unlock the self-contained badge in a single edit.
Can a passage have ANAPHORA_OPEN and NO_ANSWER_FIRST simultaneously?
Yes — both fire independently on different sentences. ANAPHORA_OPEN fires on any sentence that opens with a pronoun; NO_ANSWER_FIRST fires specifically when the first sentence is not a direct declarative answer with a fact token. If the first sentence is "This section explains…" it triggers both: anaphoric opener ("This") and missing fact token.
Why is KEYWORD_STUFFING evidence-graded weak if high density is bad?
Because the negative correlation between keyword density and citation is real but the threshold is fuzzy — a passage on "BM25 retrieval" naturally repeats those terms, and the model cannot always distinguish topic focus from manipulation. Seeklyn flags it as an inverted signal to watch, not a structural failure, which is why it carries a weak evidence grade rather than strong.