Skip to main content
Ask Pi to run the batch through keyword-evaluation (a 5-dimension ACCEPT / REVIEW / REJECT gate) and serp-cliff-analysis, which reads the live SERP to find the cliff — the position where result quality drops off. A low cliff (weak pages ranking by accident) is your insertion point; a high cliff (strong, exact-match, high-authority top results) means you’re locked out. Both skills are read-only, so they report without changing keyword state; you tag the ACCEPTs next in 2.2.3. This is why you pick keywords on the cliff, not on volume.
Pick keywords on the SERP cliff, not on volume. Drop keywords you can’t win, even high-volume ones. See Part 1 §1.4 SERP cliff, and the one-intent / no-self-competition rule in §1.5 Intent & cannibalization.
keyword-evaluation grades each keyword against five dimensions (2-D label fit, SERP cliff, intent winnability, ICP/offering specificity, and evidence/source) and it never relies on keyword difficulty. Every non-ACCEPT verdict comes with an Exploration Prescription: a named tactic and the literal next tool call to run, not vague feedback. It also surfaces weak-competitor openings as “leads worth chasing.” The skill is read-only, so it reports without mutating keyword state; tagging and filing happen afterward in 2.2.3.

​
How the verdicts map to action

VerdictWhat it meansWhat you do
ACCEPTWinnable, relevant, right intentKeep, then tag it into the cycle
REVIEWPromising but flagged: soft SERP unclear, borderline intent, or needs human triageRead the Exploration Prescription, then refine or decide
REJECTLocked-out SERP, wrong intent, no demand, or a content-idea queryDrop, or reshape per the prescription
The SERP-cliff read layers on top. A low cliff (weak pages ranking by accident, no exact-title match, low domain authority) is your insertion point. A high cliff (positions 1-7 all strong, exact-match, high-authority) means you’re locked out, so defer or find a softer variant.

​
Steps

  1. Select the candidate batch in the Keywords dashboard (from 2.2.1).
  2. Ask Pi to run it through the keyword evals + SERP cliff.
  3. Read the report: ACCEPT / REVIEW / REJECT per keyword, plus cliff position and “are you already ranking?” for each.
  4. Apply the verdict rubric below, including the drop/keep/optimize decision when you already rank.
  5. Hand the ACCEPTs to tagging in 2.2.3.

​
Real prompts

The blunt version. Run everything through the gate: 
look at ALL the keywords right now and run it through the keyword evals.
Pi runs keyword-evaluation and returns a per-keyword ACCEPT / REVIEW / REJECT report with an Exploration Prescription on every non-ACCEPT and “leads worth chasing” where a competitor is winning without deserving it. The full winnability + cannibalization pass with SERP cliff: 
use the keyword explorer and serp cliff to check through all the drafted keywords. for each:
- get US volume
- run serp cliff analysis
- check if it already has a landing page (no duplicates / cannibalization)
- check if we already rank for it and at what position
- flag any keywords in our list too similar to each other (pick the stronger, drop the other)
The keep/drop/optimize rubric when you may already rank (paste positions back in): 
before we commit these, check the SERP for each and whether [my company] already ranks:
- already #1/#2 → drop (don't build a competing page)
- not ranked yet → keep
- striking distance (pos 8–20, high impressions) → keep to optimize the existing page
🎬 Video planned: select a batch, run evaluation, read the SERP-cliff signal, make the keep/cut decision. See the shot-list.
Previous: 2.2.1 Keyword exploration · Next: 2.2.3 Organise, tag & file keywords · Theory: Part 1 §1.4 SERP cliff · §1.5 Intent & cannibalization · The rigorous cycle version: 3.4 Organise keywords for a cycle