/seo-traffic-drop

name: seo-traffic-drop
description: "Diagnose a traffic or ranking loss before fixing anything. Input: Search Console exports from a healthy period and a damaged one, plus what changed on the site. Output: whether the drop is real or a reporting artifact, the exact date and shape of the loss (visibility, ranking, click-through, demand), the pages and queries carrying it, a differential diagnosis with the test that confirms each cause, and the handoff to the skill that fixes it. Bundled zero-dependency script compares two exports and dates the drop. Use when traffic fell, rankings dropped, impressions or clicks collapsed, a core update landed, pages got deindexed, a migration or redesign went wrong, or a client asks what happened."
license: MIT
metadata:
  author: "Sorank (https://sorank.com)"
  version: "1.0.0"

Traffic drop: find what changed before changing anything

A drop is a diagnostic problem, not a content problem. The failure mode of this work is not missing the cause, it is acting on the first plausible one: a core update landed near the date, so the team rewrites 40 articles, while the actual cause was a `noindex` shipped in a template on the same week. The rewrite then hides the evidence.

So the order is fixed. Confirm the loss is real, date it, shape it, name the candidate causes, confirm one with a test that could have failed, and only then fix. Every step below produces evidence someone else can check.

Company knowledge first (Obsidian)

If the working environment contains an Obsidian vault or any local knowledge base (a folder of .md notes, often with a .obsidian directory), read the relevant notes before acting: the SEO action log (what was shipped and when), past incidents, the site's seasonality, migrations, and known technical debt. The action log is the highest-value note in a drop investigation, because it answers "what changed" without waiting for the client. At the end of the session, append the diagnosis, the evidence and the remediation to that log so the next incident starts with a history. Vault structure and protocols: the obsidian-brain skill.

When to use this skill

Use it for:

• "Our traffic dropped", "we lost rankings", "clicks fell off a cliff"
• Impressions or clicks collapsing in Search Console, on the whole site or one section
• Pages disappearing from the index, or a sudden coverage change
• A suspected core update, a manual action, or a penalty
• Traffic loss after a migration, a redesign, a replatform or a big deploy
• Sessions falling in GA4 while Search Console looks fine, or the reverse
• AI referral traffic falling, or the brand losing its citations in AI answers

Hand off once the cause is named:

| Confirmed cause | Skill that fixes it | |---|---| | Indexation, robots, rendering, speed, migration, hack | seo-technical | | Client-rendered site built by an AI generator | seo-ai-site-builders | | Thin, stale or outranked content | seo-content-blog and the page-type content skills | | Blocks missing from a page that competitors have | seo-page-sections | | Cannibalization, orphaned pages, weak internal links | seo-internal-linking | | Lost or toxic backlinks | seo-backlinks | | Local pack or Business Profile loss | seo-local | | Lost AI citations, AI Overview displacement | geo-visibility | | No measurement to detect the next one | geo-tracking |

Step 0: Freeze the scene

Before any analysis, stop changes to the site. Every deploy during an investigation destroys evidence and adds a variable, and a drop that gets "fixed" mid-diagnosis by three simultaneous changes teaches nobody anything.

Collect, in this order:

1. Search Console exports: the performance report for the damaged period and for an equal-length healthy period, same property, same filters. Export Queries, Pages, and Dates. Equal length matters: comparing 28 days against 90 measures the calendar. 2. The deployment and release log, or a git log, covering the six weeks before the drop. 3. Anything non-technical that changed: a rebrand, a price change, a product discontinued, a campaign that ended, a season that turned, a competitor that launched. 4. Search Console Manual Actions and Security Issues. Two clicks, binary answer, and it reframes everything if positive.

If the client cannot say what shipped and when, that gap is itself a finding worth reporting.

Step 1: Is the loss real

Three things masquerade as traffic drops. Rule them out before spending a day on causes.

**Reporting artifacts.** Two documented events sit inside the window most investigations look at:

• The num=100 removal of September 2025. Rank trackers stopped loading 100-result pages, so many properties saw desktop impressions collapse and average position improve overnight, with clicks flat (https://searchengineland.com/google-num100-impact-data-462231).
• The Search Console impression logging error, disclosed on the Data Anomalies page: from 13 May 2025 until the fix rolled out on 27 April 2026, impressions were logged inaccurately, which also distorted CTR and average position. Clicks were not affected, and Google did not reconstruct the historical data. Any before-and-after comparison spanning 27 April 2026 therefore compares inflated impressions against corrected ones (https://support.google.com/webmasters/answer/6211453 and https://www.seroundtable.com/google-search-console-fix-data-logging-issue-41260.html).

The operating rule that survives both: **clicks are the metric that pays**. An impressions drop with flat clicks is a measurement story until proven otherwise. Check the Data Anomalies page for the period under investigation before writing any conclusion.

**Measurement breakage.** If GA4 fell but Search Console did not, the problem is downstream of Google: a tag removed in a redesign, a consent banner change, a bot filter, a new cookie mode, a GA4 property or filter change. Compare Search Console clicks against GA4 organic sessions over the same days. They never match exact