traffic-decline-diagnosis
Skill7 starsChanged 3 months agoMITView on GitHub
Diagnose why a site or page lost search traffic by separating ranking loss from demand loss from CTR loss. Use when asked why traffic dropped, why clicks are down, what a Google algorithm update did to the site, whether a decline is the site's fault, or which pages are declining.
In this repository7 other files it gives its agents.
- AGENTS.mdAGENTS.md
- .cursor/rules/gsc-mcp.mdcCursor rule
- ctr-gapsSkill
- gsc-recipesSkill
- index-healthSkill
- monthly-reportSkill
- nonbrand-performanceSkill
The file
We split this skill into the rules it gives an agent
6 rules. About 0 are common, 2 are distinctive, and 4 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈2
Distinctive · estimate
This author's own choices. Worth a look.
4
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 26
Call list_sites. Match the user's domain to a site_url.
- Distinctive: line 46
Ask the user which question they are asking if it is not obvious.
- Distinctive: line 71
If a user reports a drop that this tool does not find, incomplete-day trimming is the first thing to check.
- Very specific: line 130
Never present unclear as "no reason found".
- Very specific: line 132
Do not attribute impressions_drop to the site's own work without a year-over-year check.
- Very specific: line 140
Use data_state: all (the default) to match the Search Console UI.
≈ “Common” is an estimate: at least 1 in 5 of the 372,429 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.
Files
Traffic Decline Diagnosis
"Traffic is down" is not a diagnosis. Three completely different things produce the same drop, and they need opposite responses:
| Cause | What actually happened | What to do |
position_drop | You lost ranking | Look at who overtook you |
impressions_drop | Fewer people searched, or the SERP changed | Often not your fault. Do not rewrite the page |
ctr_drop | Same ranking, same searches, fewer clicks | This one points at the title |
Getting this wrong is expensive in both directions: rewriting titles when search demand fell seasonally, or ignoring a competitor who took your position because "traffic is just down".
find_declining_pages does the attribution. It is the only reason to use this
tool over a plain compare_periods call.
Step 1 — Confirm the property
Call list_sites. Match the user's domain to a site_url.
Step 2 — Choose the two periods
Period A is the baseline (earlier), period B is current (later). Every delta is
B minus A. The periods must be the same length, and A must end before B starts —
reversed or overlapping ranges return invalid_input.
Two ways to pick them, and they answer different questions:
"What changed recently?" — consecutive equal windows:
period_a: 2026-06-15 → 2026-07-12 (28 days)
period_b: 2026-07-13 → 2026-08-09 (28 days)"How does this compare to last year?" — same window, previous year. Use this when the business is seasonal, because a consecutive comparison will attribute seasonality to your work.
Ask the user which question they are asking if it is not obvious. A drop measured against the wrong baseline is worse than no measurement.
find_declining_pages
site_url: <resolved>
period_a_start / period_a_end: <baseline>
period_b_start / period_b_end: <current>Step 3 — The incomplete-data trap this tool closes for you
Counting the last few days of the current period against a complete baseline is the classic way to invent a decline that does not exist. Google's data is not final for the most recent days.
This tool reads metadata.firstIncompleteDate and trims the end of both periods
by the same number of days, keeping them equal length. Read these fields and
report them:
| Field | Meaning |
assumptions.excluded_incomplete_days | How many days were trimmed off each period |
assumptions originally-requested ends | Present when trimming happened — the dates you asked for vs. the dates used |
If a user reports a drop that this tool does not find, incomplete-day trimming is the first thing to check. Their number probably included unfinalized days.
Step 4 — Read the attribution, then trust it selectively
attribution | Confidence | What to do next |
position_drop | High | Check the SERP for this query manually. Who is above you now? |
impressions_drop | High | Check whether this is seasonal or a SERP layout change. Compare year-over-year before concluding anything |
ctr_drop | High | Title and meta description. Also check whether an AI Overview or featured snippet appeared |
unclear | None — by design | The page is missing a position on one side: it appeared or vanished. This means "we do not know why", not "no cause" |
unclear is the honest answer, not a gap in the tool. A page with no position in
period A is new; a page with none in period B may be deindexed, redirected, or
deleted. Those need inspect_url or a look at the site, not a guess.
When several pages share one attribution, treat it as one event, not N problems.
Twelve pages all showing position_drop in the same window is a site-level event —
an algorithm update, a technical regression, a migration — not twelve page problems.
Say that. Fixing twelve pages one at a time is the wrong response to one cause.
Step 5 — Thresholds
| Parameter | Default | Move it when |
min_click_drop | 10 (inclusive) | Small site → lower to 3–5, or nothing reports. Large site → raise to 50 to cut noise |
position_drop_threshold | 1 | Rarely. Under 1 position of movement is noise |
impressions_drop_pct_threshold | -10 | Volatile niche → -20 so normal fluctuation is not called a decline |
Rows are ranked by clicks_delta ascending — biggest absolute loss first. The
applied values come back in assumptions; quote them.
Step 6 — Report
Lead with the split, because the split is the diagnosis:
Total decline: −340 clicks across 14 pages
| Cause | Pages | Clicks lost | Reading |
position_drop | 8 | −210 | Competitors gained. Likely one event, see below |
impressions_drop | 3 | −80 | Demand fell — not a site problem |
ctr_drop | 2 | −40 | Title issue |
unclear | 1 | −10 | Page appeared or vanished — needs a look |
Then the worst pages within each cause, with their actual numbers:
| Page | Clicks A → B | Position A → B | Attribution |
| /example-page/ | 120 → 40 | 4.2 → 9.8 | position_drop |
Close with one sentence naming the most likely single cause, and say explicitly which part of the decline is not the site's fault. Clients rarely hear that, and it is usually the most valuable thing in the report.
Accuracy rules
- Never report a decline without stating
excluded_incomplete_days. It is the difference between a real drop and an artefact. - Never present
unclearas "no reason found". It means the page appeared or vanished — that is a specific finding requiring a specific follow-up. - Do not attribute
impressions_dropto the site's own work without a year-over-year check. Seasonality is the most common explanation and the easiest to mistake for a failure. - One attribution shared by many pages is one event. Report it as one.
- This tool makes three API calls (a freshness probe plus one page-level scan per
period, each up to 25,000 rows). Read
scan_cappedbefore calling the list complete. - The tool classifies but does not advise. Every recommendation is yours to defend.
- Use
data_state: all(the default) to match the Search Console UI.
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
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.

