> ## Documentation Index
> Fetch the complete documentation index at: https://docs.synscribe.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Keyword evaluation vs SERP-cliff analysis: what's the difference?

> Keyword evaluation is the full batch quality gate (a 5-dimension rubric, cannibalization gate, and ACCEPT/REVIEW/REJECT verdicts); SERP-cliff analysis is just the winnability lens inside it, run as a quick check on a small keyword cluster.

`keyword-evaluation` is the **full batch quality gate**: ask Pi to run it and it grades every
candidate against a 5-dimension rubric (SERP cliff is one of the five), runs a cannibalization gate,
and hands back an ACCEPT / REVIEW / REJECT verdict per keyword with a concrete next step on every
non-ACCEPT. `serp-cliff-analysis` is **just the winnability lens** — one of those five dimensions —
run as a fast tactical check on a small cluster of 2-5 overlapping keywords when all you need to
know is "can I win this one right now?". Both replace keyword difficulty with the cliff. Neither is
the *guardrail* that gates content before publish — these two gate **keywords before you build**. See
[Evaluate + SERP-cliff a batch](/platform/evaluate-serp-cliff) for how to run them, and
[Refiner vs guardrail](/platform/refiner-vs-evaluator) for the full three-way distinction.

## The two skills, side by side

|                       | `keyword-evaluation` (the gate)                                                                          | `serp-cliff-analysis` (the lens)                                         |
| --------------------- | -------------------------------------------------------------------------------------------------------- | ------------------------------------------------------------------------ |
| Scope                 | A whole batch of candidates                                                                              | A small cluster of 2-5 overlapping keywords                              |
| What it produces      | ACCEPT / REVIEW / REJECT per keyword + an Exploration Prescription on every non-ACCEPT                   | A cliff score (1-8+) and a keep / target / defer / park call per keyword |
| Dimensions            | 5 (commercial-intent fit, **SERP cliff**, intent winnability, ICP/offering specificity, evidence/source) | 1 — the SERP cliff                                                       |
| Cannibalization       | Mandatory gate (SERP-overlap test against existing assets)                                               | Lighter sibling-asset check before a keep/drop call                      |
| Persistence           | Writes a report + followups file to Pi's `/memory/`                                                      | Reports in chat; no persisted artefact                                   |
| State                 | Read-only (never mutates keyword state)                                                                  | Read-only                                                                |
| When you reach for it | Gating a drafted batch before it enters the plan                                                         | Deciding between a handful of variants, or during cannibalization triage |

## How they nest

SERP-cliff analysis isn't a competitor to keyword evaluation — it's the D2 dimension *inside* it,
pulled out so you can run it on its own. Every survivor in a keyword-evaluation run gets a mandatory
SERP read (the cliff); the standalone `serp-cliff-analysis` skill is that same read, scoped to a
few keywords when you don't need the full rubric, the cannibalization matrix, or the saved report.

The cliff is the position in the top 7 results where quality drops off: the first result with no
exact keyword in its title tag *and* a low-authority domain. A low cliff (1-3) is a wide-open
insertion point; a high cliff (8+) means the SERP is locked. This is why you never score on keyword
difficulty — the cliff is the substitute. See [Part 1 §1.4 SERP cliff](/theory/serp-cliff).

## Which one to reach for

* **"Gate this whole batch before I plan a cycle."** -> `keyword-evaluation`. You get verdicts,
  prescriptions, the cannibalization matrix, and a saved audit trail.
* **"I've got three near-identical variants — which do I keep?"** -> `serp-cliff-analysis`. Fast,
  focused, no report overhead.
* **"Can this content pass compliance before I publish?"** -> neither — that's the *guardrail*
  (guardrail). Keyword eval and SERP cliff gate keywords *before* you build a page;
  the guardrail gates finished content *before* it ships. See
  [Refiner vs guardrail](/platform/refiner-vs-evaluator).

## The three-way distinction, in one line

> **Keyword evaluation** = the whole gate for keywords (verdicts, prescriptions, cannibalization,
> audit trail). **SERP-cliff analysis** = the winnability lens inside it, run standalone on a small
> cluster. **The guardrail** = a separate publish-time gate for *content*, not keywords. Don't let
> the shared word "evaluate" blur them.

***

**How to run them:** [2.2.2 Evaluate + SERP-cliff a batch](/platform/evaluate-serp-cliff) · **Theory:** [Part 1 §1.4 SERP cliff](/theory/serp-cliff) · [§1.5 Intent & cannibalization](/theory/intent-cannibalization) · **The content-side gate:** [2.7.4 Refiner vs guardrail](/platform/refiner-vs-evaluator) · **Previous:** [2.2.4 Monthly keyword report](/platform/monthly-keyword-report)