psyke-audit（心理审计）

Psyke SEO & GEO Audit

Branded audit product for Psyke (psyke.co). Produces a scored, dark-mode slide deck covering technical SEO, GEO readiness, competitor benchmarking, and actionable recommendations.

Output

A single index.html file deployed to Vercel containing:

• SEO Health score (out of 100)
• GEO Readiness score (out of 100)
• Scored breakdown by category
• Issue slides with evidence and fixes
• Competitor benchmark (when Ahrefs data available)
• AI visibility test results
• Prioritised action plan
• Psyke CTA slide

References

Read these before building an audit:

• references/scoring.md — scoring methodology (SEO + GEO rubrics)
• references/psyke-theme.css — complete CSS theme (paste into <style>)
• references/slide-structure.md — fixed slide order and design rules
• references/geo-framework.md — GEO assessment process

Process

Phase 1: Data Collection

1. Technical Crawl (always)

- Fetch homepage, robots.txt, sitemap.xml
- Fetch 4-6 key inner pages (services, about, contact, sector pages)
- Use browser to evaluate each page for: title, meta description, canonical, H1s, H2s, H3s, schemas, OG tags, image alt text, viewport meta
- Check for soft 404s on common URL patterns
- Note the CMS/platform if identifiable

2. Ahrefs Data (when MCP connected)

Pull via Ahrefs MCP or have Psyke provide:
- Domain Rating
- Organic keyword count + top keywords
- Monthly organic traffic
- Referring domains count
- Google AI Overview citation count
- ChatGPT citation count
- Branded vs non-branded split (estimate)
- Same metrics for 2-3 competitors

3. AI Visibility Testing (always)

Run 5-8 queries through ChatGPT and/or Perplexity:
- "[client industry] + Australia" query
- "best [service/product]" query
- "how does [client service] work" query
- "[client] vs [competitor]" query
- Location-specific query
Record: was client mentioned? which competitors? what was cited?

Phase 2: Scoring

Apply the rubrics from references/scoring.md:

1. Score each SEO category, noting specific evidence for each deduction
2. Score each GEO category, noting specific evidence
3. Calculate totals — do not round

Phase 3: Build the Deck

Setup:

cd ~/Developer
mkdir [client]-seo-audit
cd [client]-seo-audit
git init
echo "# [Client] SEO & GEO Audit" > README.md

Build index.html:

1. Start with the viewport-base styles from the frontend-slides skill
2. Paste the full Psyke theme CSS from references/psyke-theme.css
3. Follow the slide structure from references/slide-structure.md exactly
4. Include the SlidePresentation JavaScript class (keyboard nav, touch, progress bar, nav dots, intersection observer)

Font include:

<link href="https://fonts.googleapis.com/css2?family=Space+Grotesk:wght@400;500;600;700&display=swap" rel="stylesheet">

Key HTML patterns:

Title slide:

<section class="slide slide-title">
    <div class="glow-orb glow-purple" style="top: -10%; right: -5%;"></div>
    <div class="glow-orb glow-lime" style="bottom: 10%; left: -3%;"></div>
    <div class="content-wrap">
        <p class="psyke-logo reveal">PSYKE</p>
        <h1 class="reveal reveal-delay-1"><span class="accent">[Client Name]</span></h1>
        <p class="subtitle reveal reveal-delay-2">SEO & GEO Audit — [Month Year]</p>
        <p class="label-muted reveal reveal-delay-3">[client-url.com.au]</p>
    </div>
</section>

Dual score slide:

<section class="slide">
    <div class="slide-split">
        <div class="split-left">
            <div class="dual-score reveal">
                <div class="score-block">
                    <div class="score-ring" style="--score-pct: 0.XX;">
                        <svg viewBox="0 0 100 100">
                            <circle class="ring-bg" cx="50" cy="50" r="45"/>
                            <circle class="ring-fill" cx="50" cy="50" r="45"/>
                        </svg>
                        <div class="score-number">XX</div>
                    </div>
                    <p class="score-label-big" style="color: var(--psyke-purple);">SEO Health</p>
                </div>
                <div class="score-block">
                    <div class="score-ring" style="--score-pct: 0.XX;">
                        <svg viewBox="0 0 100 100">
                            <circle class="ring-bg" cx="50" cy="50" r="45"/>
                            <circle class="ring-fill ring-fill-lime" cx="50" cy="50" r="45"/>
                        </svg>
                        <div class="score-number">XX</div>
                    </div>
                    <p class="score-label-big" style="color: var(--psyke-lime);">GEO Ready</p>
                </div>
            </div>
        </div>
        <div class="split-right">
            <!-- Breakdown grid -->
        </div>
    </div>
</section>

Competitor benchmark:

<div class="competitor-grid">
    <div class="competitor-row header">
        <div class="competitor-cell"></div>
        <div class="competitor-cell">DR</div>
        <div class="competitor-cell">Keywords</div>
        <div class="competitor-cell">Traffic</div>
        <div class="competitor-cell">AI Citations</div>
    </div>
    <div class="competitor-row client">
        <div class="competitor-cell">[Client]</div>
        <div class="competitor-cell">XX</div>
        <!-- etc -->
    </div>
    <!-- competitor rows -->
</div>

Phase 4: Deploy

git add -A
git commit -m "[Client] SEO & GEO Audit - SEO: XX/100, GEO: XX/100"
gh repo create beebeebot/[client]-seo-audit --public --source=. --push
npx vercel --yes --prod

Phase 5: Deliver

Send the Vercel URL to the requester with:

• Both scores and breakdown summary
• The presentation link
• 2-3 sentence summary of key findings

Writing Style

• Direct and specific. No filler.
• Evidence before opinion. Show the data, then make the observation.
• One sharp insight per issue, not three paragraphs restating the same point.
• Recommendations must be specific: "Add FAQ schema to /services/ page" not "consider adding structured data".

AI Writing Patterns to Avoid (mandatory)

These are the most common tells. Fix every instance before deploying:

• Em dashes (—): Replace with periods, commas, or restructure. Use sparingly if at all.
• AI vocabulary: Don't use "crucial", "landscape", "delve", "enhance", "foster", "showcase", "underscore", "pivotal", "vibrant", "testament", "garner", "interplay", "tapestry". Use plain words.
• Rule of three: Don't force ideas into groups of three. "Buying guides, how-tos, and comparison content" → "buying guides and comparisons".
• Negative parallelisms: Don't write "It's not just X, it's Y". Just say what it is.
• Promotional language: No "breathtaking", "groundbreaking", "stunning", "nestled", "in the heart of".
• -ing superficial analysis: Don't tack on "ensuring...", "highlighting...", "showcasing..." to add fake depth.
• Vague attributions: No "experts believe", "industry reports suggest". Cite specifics or don't claim it.
• Filler phrases: No "it's important to note that", "in order to", "at the end of the day".
• Sycophantic tone: No "Great question!", "Excellent point!", "That's a really good observation".
• Generic positive conclusions: No "the future looks bright" or "exciting times ahead".

What NOT to Do

• Don't improvise the slide order — follow the structure
• Don't use light-mode slides — everything is dark Psyke theme
• Don't skip the GEO assessment, even without Ahrefs — the AI query testing and citability audit always apply
• Don't round scores for optics — 43 is 43
• Don't write generic recommendations — every fix should reference a specific URL or element
• Don't include the Psyke logo on every slide — title and closing only