agentleFS
trananhtungtrananhtung/seo-audit-skills/skills/seo-traffic-diagnosis/SKILL.md

seo-traffic-diagnosis

Skill0 starsChanged 2 months agoMITView on GitHub

Diagnose why a website lost organic search traffic, rankings or visibility, and produce a recovery plan — confirming the drop is real and is Google organic, checking Manual Actions and Security Issues, pinpointing the drop date against algorithm rollout windows, scoping it to a URL, a section or the whole site, then running the matching playbook (serving and reachability failures, technical change, lost internal or external links, content change, stale content, page experience, thin or unhelpful content, indexing loss, deindexing). Built on The Art of SEO, 4th edition, chapters 9 and 10. Use whenever there is a specific decline to explain: organic traffic or rankings fell from a previously higher level, pages that were indexed have been dropped, a page disappeared from Google, traffic fell after a site launch or a Google update, someone suspects a penalty or manual action, a reconsideration request is needed, or someone asks how to recover after a core update. Also fires on the same requests in Vietnamese, such as "traffic organic giảm", "rớt top", "tụt hạng sau cập nhật Google", "mất traffic", "web bị Google phạt", "trang trước có index giờ mất index", "gỡ án phạt". Use seo-site-audit INSTEAD for a general or baseline review, a pre-launch or pre-migration check, or a site that has simply never ranked well with no decline to explain.

1 other project uses a version of this.

Adopted by

Most recently taken up by

  • Other versions1 project uses a version of this.

    Word for word · 0

    None found.

    Edited · 0

    None found.

    Wrote their own version · 1

    1. rampstackco/claude-skills-seo★ 5same name

    How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.

  • In this repositoryOne other file it gives its agents.

The file

We split this skill into the rules it gives an agent

12 rules. About 0 are common, 11 are distinctive, and 1 is very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈11

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 1

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Every step below produces a fact or a "could not determine".

    line 35
  2. Distinctive:

    Search Console notice turns a diagnosis into a story.

    line 37
  3. Distinctive:

    If the drop is not in Google organic, stop.

    line 75
  4. Distinctive:

    A notice does not automatically explain the drop: Before closing the diagnosis, check that the action's timing and scope actually match …

    line 90
  5. Distinctive:

    If timing and scope fit, the diagnosis is done.

    line 99
  6. Distinctive:

    If there is no notice, a manual action is ruled out and this is algorithmic or self-inflicted.

    line 103
  7. Distinctive:

    If the date does not match anything you shipped, check whether Google changed something.

    line 128
  8. Very specific:

    If the date falls in a known rollout, learn what that update targeted, then determine whether those factors apply to this site.

    line 149
  9. Distinctive:

    After making changes, Google must recrawl the affected pages and reprocess them through its algorithms.

    line 215
  10. Distinctive:

    Only if there is an actual manual action, and only from the Manual Actions report — Security Issues use their own review request, not this …

    line 249
  11. Distinctive:

    Write the report in the language the user is writing to you in.

    line 274
  12. Distinctive:

    Declaring failure before Google recrawled: Check the recrawl happened before concluding the fix did not work.

    line 293

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

SEO traffic drop diagnosis

A traffic drop is a diagnosis problem, not a fix-things problem. The single most expensive mistake is starting remediation before you know what happened — changing twenty things at once means you learn nothing from the result, and some of those changes will make it worse.

Work the sequence. Do not skip ahead to fixes.

The rule that matters most: never invent evidence

Every step below produces a fact or a "could not determine". Guessing a drop date, assuming an algorithm update hit, or asserting a penalty without seeing the Search Console notice turns a diagnosis into a story.

LabelMeaning
CONFIRMEDSeen directly in analytics, Search Console, a crawl, or the live page
LIKELYStrong circumstantial fit. State what would confirm it
RULED OUTChecked and eliminated. Say how
UNKNOWNNo access or no data. Say what access is needed

Say "I do not have Search Console access, so a manual action cannot be ruled out" rather than proceeding as though there is none. An undiagnosed drop honestly labelled is a working state. A confident wrong diagnosis costs months.


Step 1 — Confirm the drop is real, and is organic Google traffic

Open analytics. Segment by traffic source. Isolate Google organic and confirm that channel dropped.

Rule out first, because these are common and look identical on a total-traffic chart:

  • A tracking break — tag removed, consent banner change, GA4 migration artifact, a filter or a bot-exclusion setting changed.
  • A drop in a different channel (paid paused, email campaign ended, a referral source dried up) dragging the total down.
  • Seasonality — compare year over year, not just to last month.
  • A branded-traffic change, which is usually a marketing or PR event, not an SEO event.
  • Impressions vs clicks in Search Console. Flat impressions with falling clicks suggests a SERP or CTR problem rather than a ranking problem, and falling impressions suggests the reverse. These are hypotheses, not conclusions — Google notes that position losses can cut clicks without moving impressions much, so the two are not cleanly separable. Compare matching segments (same queries, pages, device, country, search type) before concluding anything, and investigate demand, ranking and presentation separately.

If the drop is not in Google organic, stop. It is not this problem.

Step 2 — Check Manual Actions and Security Issues

These are two different reports with two different review processes. Check both, plus the notification bell, and keep them separate.

Security Issues (hacked content, malware, deceptive pages). Remediate the compromise, then request review from the Security Issues report. A reconsideration request is the wrong channel and will not clear it.

Manual Actions. Go to references/penalties-and-manual-actions.md, work the remedy for the specific action named, then file a reconsideration request from the Manual Actions report.

A notice does not automatically explain the drop. Before closing the diagnosis, check that the action's timing and scope actually match what you observed. Two common mismatches:

  • The action predates the drop by months, so something else caused this decline.
  • The action's scope is narrower than the drop — a structured data manual action removes rich-result eligibility but does not demote ordinary web rankings, so it cannot explain a site-wide collapse in blue-link traffic.

If timing and scope fit, the diagnosis is done; continue to remediation. If they do not, fix the flagged issue anyway and keep going through steps 3–6 for the real cause.

If there is no notice, a manual action is ruled out and this is algorithmic or self-inflicted. If you have no Search Console access, mark it UNKNOWN, put getting access at the top of the action list, and continue.

Step 3 — Pinpoint the date

Find the exact day the decline started. Not the week — the day. Use the analytics day view and the Search Console Performance day view together.

The shape tells you a lot:

PatternUsually means
Cliff on one day, flat afterAlgorithm update, manual action, or a site change shipped that day
Cliff, then partial recovery over daysA technical error that got fixed, or a temporary deindexing
Steady decline over weeks or monthsContent decay, competitors improving, or slow loss of links
Sawtooth / repeated dipsServer errors, crawl budget problems, or intermittent rendering failure
Drop confined to one page typeA template change

Now overlay your own change log. Ask the dev team what shipped that week — this is the single highest-yield question in the whole process and it is routinely skipped. Most drops with a clean one-day cliff are self-inflicted.

Step 4 — Check the date against known algorithm updates

If the date does not match anything you shipped, check whether Google changed something. Trackers named in the book:

  • MozCast
  • Moz's Google Algorithm Update History
  • Semrush Sensor
  • Rank Ranger's Rank Risk Index
  • AccuRanker's "Google Grump" rating
  • Algoroo
  • Advanced Web Ranking's Google Algorithm Changes
  • cognitiveSEO Signals' Google Algorithm Updates

Plus industry reporting: Search Engine Roundtable (Barry Schwartz), Search Engine Land, Search Engine Journal, Moz, and SEO practitioners on X — an unconfirmed update usually surfaces there first.

Compare your date against each update's full rollout window, not its announcement date — core updates typically take one to three weeks to roll out, so a decline that begins mid-rollout still fits. Timing corroborates a cause; it never establishes one on its own.

If the date falls in a known rollout, learn what that update targeted, then determine whether those factors apply to this site. references/algorithm-updates.md covers the major named updates and what each one actually went after.

If there is no notice and no matching update, recovery is harder because you do not know the reason. Google makes small changes daily that it does not announce, and even these can move traffic materially. The book's guidance for this case is direct: focus on best practices, or bring in an experienced SEO practitioner. Say that plainly rather than inventing a cause.

Step 5 — Scope the drop

Which is it?

  • A single URL
  • A section or page type
  • The whole site

Get this from the analytics landing-page report and the Search Console Pages report, comparing the periods either side of the date from step 3.

Scope narrows the cause dramatically. If traffic fell only in one area, the cause is very likely in that area. A site-wide drop with one section untouched points somewhere different from a uniform site-wide decline.

Also separate:

  • Ranking loss (positions fell, impressions fell) from
  • Indexing loss (pages left the index — Search Console Pages report shows the reason) from
  • CTR loss (positions and impressions held, clicks fell — a SERP layout or title/description problem).

Step 6 — Run the matching playbook

references/diagnostic-playbooks.md holds the full set.

Always run playbook §0 (Serving and reachability) first, whatever the scope. DNS failures, 5xx spikes, a 503 left in place, a robots.txt returning 5xx, a WAF or CDN rule blocking Googlebot, an X-Robots-Tag: noindex in the response headers, a broken redirect chain, or Google selecting a different canonical will each produce a drop that looks exactly like an algorithmic hit. They are fast to check and they invalidate everything downstream if missed.

Then pick by scope and symptom:

ScopePlaybook
Any drop, always first§0 Serving and reachability
One URL or one sectionTechnical change · lost internal links · lost external links · content change · stale content
Site-wideTechnical change at scale · internal link structure change · lost external links · site-wide content change · stale content · algorithm update · E-E-A-T and YMYL
Pages not indexednoindex · insufficient content quality · insufficient link authority · manual action · Remove Outdated Content tool
Pages not crawledrobots.txt · meta robots · no inbound links · form-gated · session IDs · insufficient link authority
Clicks down, positions flatSERP layout change · Google rewriting your titles and descriptions
Page experienceCore Web Vitals · mobile friendliness · interstitials · HTTP pages

Work the playbook's checks in order. Each ends in CONFIRMED, LIKELY, RULED OUT, or UNKNOWN — all four labels, including in the final report. Do not stop at the first plausible cause — drops frequently have two causes, and fixing one while missing the other looks like the fix failed.

Step 7 — Remediate, then wait properly

State the expected timeline explicitly, because this is where most recovery efforts are abandoned too early.

After making changes, Google must recrawl the affected pages and reprocess them through its algorithms. Depending on site size, the number of pages involved, and how often those pages are crawled, this takes one week to several months.

So the plan says:

  1. The change, and which URLs it touches.
  2. How to verify Google has seen it — URL Inspection for a handful of pages; the Pages report for a section; log data if available. Do not evaluate before confirming the recrawl.
  3. What should move, in which report, and when.
  4. What to do if it does not — go back to step 5 and look for the factor you missed. Say this in advance, so a non-recovery is treated as more diagnosis rather than as a failure.

Change one thing at a time where the timeline permits. Recovery work that changes ten variables at once teaches you nothing.

Recovering from a core update

Core updates are not penalties, and there is nothing to appeal. A core update changed how Google weighs things; the site did not necessarily break a rule. Recovery comes from being genuinely better on what the update rewards — which generally means content quality, expertise, and the proxy signals for E-E-A-T.

[UPDATE] On recovery timing. The book's framing is that recovery lands at the next core update. Google's current guidance is looser than that: changes may register anywhere from days to months, and improvements can surface between announced updates as well as at one. Say "days to months, and possibly not until a later update" rather than promising either a quick recovery or a fixed wait for the next update.

Filing a reconsideration request (manual actions only)

Only if there is an actual manual action, and only from the Manual Actions report — Security Issues use their own review request, not this one.

references/penalties-and-manual-actions.md holds the full guidance on what to write and what to avoid. It is the single authority; do not restate it here. The two things worth carrying in your head: keep it short and factual, and what clears an action is evidence of genuine remediation, not a persuasive letter.


Report format

  1. Verdict — one paragraph: what happened, when, how confident, and on what evidence.
  2. The evidence — the drop chart description, the date, scope, and each check with its CONFIRMED / RULED OUT / UNKNOWN label.
  3. Causes — primary and, where present, contributing. If undetermined, say so and list what would determine it.
  4. Recovery plan — ordered, with owners and effort.
  5. Timeline and verification — per step 7, with the honest wait.
  6. Prevention — what process change stops a recurrence. An SEO review gate before deploys, a monitored crawl, alerting on index count and CWV. Most self-inflicted drops are process failures, and the diagnosis is incomplete without naming the process.

Write the report in the language the user is writing to you in.

Traps

  • Blaming an algorithm update on loose date proximity. Core updates roll out over one to three weeks, so compare against the full announced rollout window rather than a single date — a drop three days into a rollout is entirely consistent with it. But the reverse trap is worse: a date falling inside a rollout window is corroborating, not proof. A deploy that shipped the same day is usually the better explanation. Check your own change log before you blame Google.
  • Assuming a penalty with no Search Console notice. Penalties are announced. Algorithmic demotion is not a penalty, and calling it one sends the client down the reconsideration path, which does nothing.
  • Disavowing links reflexively. Google largely just ignores the value of bad links, so disavowing is usually unnecessary — and a disavow file containing perfectly good links will hurt the site.
  • Fixing before diagnosing. Twenty simultaneous changes destroy the ability to attribute the recovery, or the further decline.
  • Declaring failure before Google recrawled. Check the recrawl happened before concluding the fix did not work.
  • Reporting on total traffic instead of the Google organic segment. Half the "SEO emergencies" that arrive are a paid campaign ending.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.