How SEO Master audits and fixes your sites

Page scrape + detection of Elementor / Yoast / RankMath / SEOPress

Pull from Google Search Console: last 28 days of queries, impressions, clicks, average position

Pull from PageSpeed Insights + CrUX field data (real Chrome user data, not lab numbers)

GPT-4o + DeepSeek-V3 analyze in parallel; semantic voting via cosine embedding → consensus, or a Gemini judge if they diverge

Semantic assembly: each page is split into narrative sections (heading + paragraphs + media + CTAs together), not into isolated widgets

Classifies 10 types: hero / services / proof / pricing / FAQ / about / CTA / feature / content / footer-cta — each type has its own May 2026 Google AI criteria

Recommendations preserve each section's existing angle and tone — an "angle preserved" badge guarantees we never rewrite your message

Design-safe length budgets: titles ±15%, paragraphs ±20%, CTAs ±10% — your Elementor design stays intact, no button ever wrapping onto two lines

Per-section verdict shown at the top of the editor: Compliant / Already optimized / Needs reinforcing / Needs reworking / Not analyzed — you know exactly where to put your energy

For every GSC query, we measure the gap between your real CTR and the theoretical CTR for your position (AWR 2024 desktop/mobile benchmarks)

If actualCTR < 0.6 × expectedCTR → flag: Google sees you, shows you, but nobody clicks → gradual demote

AIO ceiling heuristic: question-style queries in the top 5 with collapsed CTR = likely AI Overview → pivot strategy to "extractable passage" rather than rewriting the title

Output: a prioritized list of queries to recover, with estimated lift in clicks/month (computeExpectedLift on the server, no LLM hallucination)

SEO Master Connector plugin installed on your WP (5 min)

HMAC anti-replay auth (timestamp + nonce + signature) — no password

Title / meta / H1 / schema / alt pushed straight into the right meta keys (Yoast / RankMath / SEOPress) OR into `_elementor_data` if Elementor

Elementor CSS cache flushed automatically after every push

Snapshot before push (page_snapshots table) → 1-click undo / rollback if needed

Daily cron re-pulls GSC for the targetedQueries of each applied recommendation

Measures Δ clicks, Δ position, Δ impressions vs the baseline captured at push time

Stored in recommendation_outcomes — you see the real data, not a theoretical estimate

Weekly regression watchdog: if an applied recommendation vanishes from the rendered HTML (theme update / stale cache), status flips to `regressed` + notification