Operational playbook for diagnosing the on-page content layer of a page or page set: title and meta description optimization, heading structure, search-intent match, answer placement and AI-search visibility per Google's 2026 guidance, E-E-A-T proof (author byline, experience signals, named specifics), and information-gain vs the top 10 SERP results. Encodes the per-check rubric, the severity matrix, and the unverified-check escalation rule. Use when auditing the on-page content surface of one or more URLs. Output: prioritized content findings with cited evidence and fix direction.

Apache-2.0 Stack-agnostic. Requires webfetch (target page + top-10 SERP pages) and websearch (SERP for the target keyword + competitor pages).

Instructions

Overview

A page can pass every structural check, load fast, validate every schema, and still rank nowhere — because the content doesn't match search intent, the headings bury the answer, the byline says "Admin," or the top 10 already covers everything this page says. This skill diagnoses the on-page layer that the structural audit can't see: how the content is shaped against the SERP it wants to win.

The audit cross-references the target page against the actual SERP — top 10 results, "People Also Ask"-style questions, AI Overview presence where it can be observed — and returns findings where the page falls short of what's already winning. Information gain (what does this page add that the top 10 doesn't) is the priority test: Google's May 2026 AI optimization guide asks for non-commodity, first-hand content, and third-party readings of the March and May 2026 core updates saw original sources gain over aggregators.

Mental model

On-page content audit is intent-versus-execution:

What is the target SERP rewarding? (intent, format, depth, structure)
What does the target page actually deliver? (against the same dimensions)
Where is the gap? (intent mismatch, structural deficit, missing experience proof, no novel signal)
What's the cheapest fix per unit of ranking lift?

The audit is not "compare to a checklist." It's "compare to the SERP." Generic on-page advice (keyword in title, keyword in H1) is necessary but not sufficient in 2026. Sufficient is matching the format, depth, and experience signal of the pages that already rank.

Instructions

Fetch protocol (parallel, ONE block)

Read the target URL with knowledge(command="read", source="<url>") — title, meta, headings, body, byline, JSON-LD.
Web search the target keyword — capture the ranking competitors and related questions.
Read 3–5 top-ranking pages with knowledge — for direct comparison.
Web search "[target keyword] site:reddit.com" — capture community phrasing and discussion.

Run all four in one block. Don't serialize. The web search tool is not Google Search: AI Overview / AI Mode presence, featured-snippet holders, and People Also Ask are Unverified unless the user supplies a screenshot or Search Console data.

Check rubric

#	Check	How to verify	Severity if failed
1	Intent match	Read the target keyword aloud — what does the searcher want? Compare to page intent.	Critical if format mismatches dominant SERP intent
2	Title tag	50–60 chars, primary keyword front-loaded, compelling (clear value, not generic)	Moderate per failed dimension
3	Meta description	150–160 chars, keyword present, answers "why click this result"	Moderate per failed dimension
4	H1	Present, unique, semantically related to title (not identical), describes the page's promise	Critical if missing or duplicate
5	H2 structure	Headings describe what each section answers; question phrasing where it reads naturally	Minor if headings are vague labels that hide the content
6	Answer placement	The direct answer to the section's question appears early, before supporting detail	Moderate per key section that buries its answer
7	Self-contained sections	Each major point reads on its own without reconstructing it from several paragraphs	Minor; don't recommend splitting content into fragments — Google says chunking isn't needed
8	Data density	At least one concrete number, percentage, or statistic per major section	Minor per data-free section; Moderate if entire piece lacks specifics
9	FAQ section	Present when users genuinely ask follow-up questions (informational and commercial intent)	Minor if absent and intent is suitable; FAQ rich results ended May 7, 2026, so judge it on reader value only
10	Author byline	Real name + role + credentials + outbound link (LinkedIn / publication / org bio). "Admin" or anonymous = Critical	Critical if anonymous or generic-name attribution
11	Experience proof	Original screenshots / dashboards / photos from the field / named tools+versions / time-boxed case-study language	Critical if pure explainer with no experience signal on YMYL or expertise-required topic
12	Information gain vs top 10	Identify the specific thing this page adds (original data, contrarian take, novel synthesis, practitioner detail) that the top 10 doesn't	Critical if the page is reconstructable from the top 10 (information-gain-zero)
13	Internal linking	5–10 contextual links per piece; descriptive anchors (not exact-match); pillar/cluster pattern present where applicable	Moderate if under 3 internal links or all anchor text is generic
14	Schema readiness for content	Article schema present with author/datePublished/dateModified; markup matches visible text	Moderate per invalid or contradicting markup; Google states AI features need no special schema
15	Primary-source posture	Page presents primary data/experience/original synthesis OR cites primary sources where it relies on them	Critical if the page is an intermediary roundup of other rankers without added signal

Severity matrix

Critical — directly suppresses ranking under 2026 systems: anonymous byline on YMYL, information-gain-zero, intent format mismatch with SERP, intermediary-only positioning, missing H1.
Moderate — degrades extraction or comprehension: buried answers, weak title/meta, invalid or contradicting schema.
Minor — polish: data-free secondary sections, internal-link count slightly low, descriptive anchor tweaks.

Information gain is the highest-leverage content check — when in doubt, weight it Critical.

How to score information gain

Before scoring, fetch and skim 3–5 top-ranking pages. Identify everything the SERP already says. Then score the target on:

Original data, survey, internal benchmark, measurement, or proprietary statistic.
First-hand case study with named specifics (time-boxed, named tools, named costs, named errors).
Contrarian take supported by evidence (not just opinion).
Synthesis across sources nobody else combined.
Practitioner detail the ranking pages skip (edge cases, real failure modes, gotchas, actual cost).

If you cannot name what the target page adds that the top 10 doesn't, that is the finding: "Information-gain-zero — page reconstructable from current top 10. Highest-priority fix."

Headings, answer placement, and AI features

Google's May 2026 AI optimization guide says AI Overviews and AI Mode need no special phrasing, chunking, llms.txt, or schema, and that a page per query variation counts as scaled content abuse. Audit structure for readers, not for a machine:

Headings that say what the section answers ("How to set up FAQ schema" beats "Schema") help readers and passage extraction alike; don't flag a page for lacking question-form H2s.
If an AI Overview cites competitors but not this page, look first at information gain, answer clarity, and indexability — not at formatting tricks.
Measure AI visibility with Search Console's Generative AI performance report (impressions only, from June 2026) and third-party trackers (Otterly.ai, Semrush AI Toolkit, Ahrefs Brand Radar, Profound) as sampled estimates. Recommend monitoring setup as part of the hand-off.

Unverified checks — escalation rule

Top-10 fetch blocked or rate-limited → "Unverified: cannot fetch SERP competitors; information-gain scoring depends on this. Recommend manual SERP review or share top-3 page snapshots."
AI Overview check impossible (search tool isn't Google) → "Unverified: AIO presence unknown; share a screenshot of the live SERP or Search Console's Generative AI report."
Author identity unclear from the page → "Unverified: author byline ambiguous; confirm author name + credentials before scoring E-E-A-T."

Never infer information gain from one source; never grade an anonymous byline before checking if there's an author bio page linked.

Output

Use the canonical output structure from the agent that runs this skill. Every finding includes: severity, evidence (URL section, line, the missing element, the SERP comparison point), why it matters, fix direction. Save as seo-audit-[slug]-[YYYY-MM-DD].md in working directory.

Checklist

All four fetch operations ran in parallel
SERP fetched and top 3–5 competitors read before scoring information gain
Every check in the rubric was evaluated or flagged as Unverified
Intent classified before format/structure assessment
Information-gain explicitly articulated (what does this page add) — flagged Critical if absent
E-E-A-T scored with named author + credentials + experience signals
AIO presence checked from user-supplied evidence or flagged Unverified; AI-visibility monitoring recommended in the hand-off
Schema checked for validity and match with visible text (Article / Person / Organization); retired FAQ/HowTo rich results not counted as lost visibility
No finding asks for chunking, llms.txt, or AI-specific phrasing — Google says none is needed
Each finding cites concrete evidence (URL section, line, missing element, SERP-comparison reference)
Report saved to disk with timestamped filename

Composition / References

Within-domain pairings:

Pairs with the sibling SEO skill that audits the structural / technical layer (crawlability, schema validity, CWV, indexability).
Pairs with the sibling SEO skill that audits link profile (referring-domain quality, anchor distribution).

External authoritative sources:

View on GitHub