How do I know if my traffic drop is from AI Overviews vs a ranking loss

Run the split on a frozen query panel, not on a sitewide CTR chart. If Google Search Console shows average position holding, impressions flat or up, and clicks down on queries that now trigger an AI Overview, that is zero-click pressure — not a ranking collapse. If average position got worse and impressions fell with it, that is ranking loss, and rewriting the explainer will not put you back on page 1. Google’s own guide to debugging Search traffic drops still starts with the Performance report; this spoke adds the Overview check that chart never labels for you.

This diagnostic sits under the Answer Engine Optimization playbook. After every query has a label, the response playbook is what to do if Overviews ate your clicks. If the URL is not even eligible to be quoted, fix that with how to get cited in AI Overviews. Lane work is on /visibility.

The short answer

• Do not diagnose from property-level CTR. Diagnose from a frozen query panel plus a SERP log.
• Overview fingerprint: average position stable, impressions flat or up, clicks down, Overview present on that query.
• Ranking-loss fingerprint: average position worse, impressions often down, clicks down. Overview may or may not be there.
• Google’s impression and position docs say an AI Overview occupies a single position and every link inside it inherits that slot. “We still rank” can be that slot.
• Third-party CTR studies disagree on magnitude. Do not invent an industry CTR. Quote your panel.

Split	Position	Impressions	Clicks	SERP
Overview zero-click	Holds	Flat or up	Down	Overview present
Ranking loss	Worse	Often down	Down	Overview optional
Mixed	Worse and Overview present	Mixed	Down	Label both; do not pick one story
Demand / seasonality	Holds	Down	Down	Google Trends agrees the category cooled

What is the Overview vs ranking-loss split test?

It is a per-query comparison of four series plus one SERP fact. You are asking whether the click disappeared because Google answered on the results page, or because your URL slipped down the page.

Google’s debugging guide still offers the old reading of “impressions hold, clicks fall”: maybe the title and snippet stopped earning the click. That can still be true. In 2026 you also have to check whether an Overview is sitting on the query. Skip that check and you will spend a month rewriting titles on pages that still rank.

Input	Where it lives	What a lie looks like
Impressions	Search performance report, Queries tab	Celebrating reach while nobody leaves Google
Clicks	Same report	Treating a one-week dip as a strategy
Average position	Toggle it on above the chart	Reading property average as if it were one query
CTR	Clicks ÷ impressions	Using CTR alone when impressions jumped
Overview present / cited	Manual SERP log, plus the generative AI impression report if you have it	One logged-in screenshot from your office

Procedure for the split:

1. Freeze the query list. Do not let next week’s “top queries” replace it.
2. Pull the same date window twice (drop period vs comparable prior period) with comparison filters.
3. Record impressions, clicks, average position, and CTR per query.
4. SERP-check each query: Overview yes/no, your URL cited yes/no, blue-link position.
5. Apply the labels in the table above. Mixed is a legal label. Unknown is a legal label. “AI killed SEO” is not.

Operator rule: if you cannot name the queries, you do not have a diagnosis. You have a mood.

First 90 minutes when someone says “organic fell”:

1. Open Performance, last 16 months, Web, your country. Screenshot the chart with average position on.
2. Date compare: drop window vs equal prior window.
3. Queries tab, sort by clicks difference. Copy the top 20 losers into a sheet. Tag brand vs not.
4. SERP-check those 20 in a clean profile. Overview yes/no, cited yes/no, blue-link position.
5. Label each row. Count click-weighted labels. That sentence is the brief.

Do not spend the 90 minutes rewriting a homepage. The sheet is the deliverable.

How do I build a frozen query panel?

A panel is a named list you will not silently mutate. Thirty to eighty queries is the working size for most sites. Twenty is a week-one floor. Two hundred is a research project, not a meeting.

Build it from the drop, not from a keyword tool’s imagination.

Include	Why	Skip
Queries that lost the most clicks in the drop window	That is the pain	Queries with a handful of impressions — noise will look like a story
Queries that land on the affected URLs	Page-level and query-level must agree	Image, Video, or News mixed into a Web diagnosis
5–10 brand / navigational queries	Control group	Brand terms dumped into the “Overview stole this” bucket
5 queries that did not drop	Negative control	Queries you cannot reproduce in the target country / language
A few question-shaped variants of the same topic	Overviews fire more often on questions	Twenty doorway clones of one head term

Checklist to freeze the list:

• Saved as a sheet with a date in the filename
• Search type locked to Web
• Country locked to the market you sell in
• Brand vs non-brand tagged in a column before you look at CTR
• Owner named (one person updates the SERP log)
• No adding “interesting” queries mid-quarter without marking them as new

If Search Console hides a query (anonymized / low volume), it does not belong on the panel. You cannot SERP-check a string you cannot see.

Impression floor (operator rule, not Google policy): if a query had fewer than ~50 impressions in the longer of the two windows, do not let it vote. Small sites should lengthen the window rather than lower the floor to 5 and then argue from noise.

Site traffic shape	Panel construction	Risk if you ignore it
Fat head, long tail	30–40 head/torso queries + brand control	Tail noise looks like a strategy
Thin, few queries	15–25 queries, 90-day windows	7-day exports that thrash labels
Many locales	One panel per paying country	US Overviews explaining EU clicks
Heavy brand	Oversample non-brand informational	Brand CTR hides the glossary bleed

Failure mode: rebuilding the panel every Monday from “Queries, sorted by clicks.” You will never see the same 40 strings twice, so you will never know what changed.

How do I pull Search Console without mixing series?

Open the Search performance report. The default chart is a blender. Filters are the test.

Google’s debugging doc tells you to stretch the range to 16 months, then compare the drop to a similar period — previous period or year over year, same number of days. Do that before you blame Overviews for January.

Filter	Setting for this split	What happens if you skip it
Search type	Web	Image/Video/News will invent a “ranking” story
Date	Compare drop window vs prior window, equal length	Monday vs Saturday, or holiday vs ordinary week
Country	The market that pays you	US Overviews + a global property average
Device	Desktop and mobile as two exports	One device hides the other
Query	Panel list, or regex of the panel	Sitewide CTR dressed up as a query test
Page	Optional second pass on the bleeding URLs	Query says Overview; page says a different URL cannibalized

Pull procedure:

1. Date filter → Compare. Equal-length windows. Prefer weekly granularity so a Tuesday does not fight a Saturday, as Google notes in the advanced comparison help.
2. Turn on Clicks, Impressions, CTR, and Average position.
3. Queries tab. Export. Keep only panel rows.
4. Repeat the export filtered to brand queries, then to non-brand.
5. Repeat desktop vs mobile if the drop is large enough to matter.
6. Pages tab, same dates: confirm the click loss sits on the URLs you think it does.
7. If the property has the Search generative AI performance report, export impressions by page for the same dates. It is impression-led. It does not replace query-level clicks.

The UI export is not the whole web. Search Console’s Queries table is a sample of what it will show you, and low-volume strings get anonymized. If a panel query is missing from the export, it is not a ranking mystery — it did not clear Google’s reporting floor for that window.

Pull constraint	What to do	What not to do
Anonymized / missing query	Drop it from the panel or lengthen the window	Invent the string from a rank tracker and pretend GSC agreed
UI row cap	Filter to the panel with query regex, or use the Search Analytics API / bulk export	Diagnose from the first 1,000 rows sorted by clicks this week
Preliminary dotted-line days	Exclude the newest incomplete day	Compare a complete June to an incomplete last Tuesday
Property vs page aggregation	Know which grouping you exported	Mix a property impression with a page CTR
Generative-AI report present	Use it as a page-level impression cross-check	Treat it as query-level clicks or average position

Do not calculate a blended “AI CTR” by dividing Web clicks by generative-AI impressions. Those denominators are different objects. Google rolled the dedicated report out to a subset of properties in June 2026; if you do not see it, do not invent a workaround metric.

What does average position stable actually mean?

Stable is not “our brand is fine.” It is a small movement in Search Console’s average of the topmost impression-bearing slot, for that query, across the window.

Google’s position documentation is the constraint: position is recorded only when there is an impression; the number in the report is an average; an AI Overview occupies one position and every supporting link inherits it. You can look “#1” because the Overview is #1 and you were cited inside it.

Google’s traffic-drop guide already splits ranking movement into two sizes you can steal as language, not as a law:

Google’s example	What they describe	What you should not do
Small drop, e.g. position 2 → 4	Noticeable click loss without a huge impression change	Radical redesign of a page that still sits in the top results
Large drop, e.g. position 4 → 29, or out of the top 10 across many terms	You left the candidate set	Call it an Overview problem and write more blog posts

Operator thresholds for this split (working rules, not Google policy):

Call it	Position movement on that query	Plus
Stable	Δ average position within about 1.0, still on page 1	Impressions not collapsing
Small rank slip	Worse by ~1–3, still top 10	Treat as ranking first if impressions also fell
Ranking loss	Persistent move out of the top 10, or a jump like 4 → 20+	Check indexing, updates, and competing URLs
Unreadable	Low impression count, so the average jumps around	Do not label; lengthen the window

Checklist before you say “position held”:

• You are reading the query row, not the property chart
• Device is split or locked
• You know whether the impression is a blue link, an Overview supporting link, or both
• You did not compare a 7-day window to a 28-day window
• A second URL from your site is not stealing the topmost slot (cannibalization looks like “we still rank”)

A stable average position with dying clicks is the Overview fingerprint only after the SERP log says the Overview is there. Position math without the screenshot is incomplete.

How do I log whether an Overview is present?

Search Console will not stamp “Overview” on the Queries table. The dedicated generative-AI report, when you have it, is impression-led and page-grouped. The SERP log is still the query-level fact.

Log like a lab notebook, not like a mood board.

Field	How to capture it	Common self-own
Query string	Copied from the panel, exact	Close-variant you invented
Date / time	ISO date, timezone you will reuse	“Sometime last week”
Country / language	Matches the GSC filter	Office VPN vs customer country
Device class	Phone or desktop, named	Responsive desktop window pretending to be mobile
Overview present	Yes / no / intermittent	One yes treated as ninety days of yes
Your URL cited	Yes / no / mention with no link	Brand name in the paragraph with no supporting link
Other cited domains	Top 3	“Competitors” with no names
Blue-link position	What you can see logged-out	Personalized rank while logged into Chrome
Screenshot	Stored next to the row	Slack image nobody can find in a month

Procedure:

1. Use a logged-out or clean profile. Target the same country as the GSC filter.
2. Search the exact panel query. Do not “help” it with extra words.
3. Record Overview yes/no before you scroll for your brand.
4. If an Overview is present, record whether your URL is a supporting link.
5. Record the first blue-link position you can actually see for your domain.
6. Repeat on a weekly cadence for the same 20 money queries; monthly is the floor for the rest of the panel.
7. Mark intermittent. Overviews often do not trigger. A week with no Overview is a serving decision, not proof you “recovered rank.”

Rank trackers can backfill history if they stored Overview presence. They do not replace a log you can defend in a meeting. If the tool cannot tell cited vs merely ranked, it is a packing list, not the split test.

How do AI Overviews and AI Mode differ in this split?

They both sit inside the Web search type in the classic Performance report, and they are not the same SERP object. If you treat every generative impression as an Overview, you will mis-label AI Mode follow-ups as “we still rank #1.”

Google’s impression and position page splits the counting:

Feature	Click	Impression	Position
AI Overviews	External link in the Overview	Standard rules; the link must be scrolled or expanded into view	The Overview is one slot; every supporting link inherits it
AI Mode	External link in AI Mode	Standard rules; a follow-up question is a new query and can add impressions	Follows ordinary Search position methods, including carousels/images inside the experience

Diagnostic implication	Overview	AI Mode
“Average position 1.2, clicks down”	Often the shared Overview slot, not a restored blue-link #1	Could be a real top slot inside the Mode experience — still not classic ten blue links
Impressions jumped	Citation expansion or more Overview serving	Follow-ups multiplying impressions on related strings
Panel query to log	The string in GSC	The string in GSC plus whether the user was in Mode; you usually cannot see Mode from a clean Overview SERP
Dedicated GSC report	Impression-led, with AI Mode, if rolled out to the property	Same report; still not query-level CTR

Checklist so you do not blend them:

• SERP log has two columns: Overview yes/no, AI Mode yes/no (Mode is a different UI)
• Follow-up-looking query strings are tagged, not merged into the parent query
• You did not divide Web clicks by generative-AI impressions
• A Mode-heavy query with falling position is still ranking loss until proven otherwise
• A classic results page with an Overview module uses the Overview row of the split test, not the Mode row

If you cannot tell Mode from Overview in the screenshot, label the SERP “generative — unspecified” and do not call it a clean Overview fingerprint. Unspecified is honest. A forced Overview label is how the sheet starts lying.

How do I label each query in the panel?

Labels are for tickets. If two smart people cannot apply the same label from the same row, the row is incomplete.

Worked example (illustration, not a client result and not an industry CTR):

Query	Δ position	Impressions	Clicks	Overview	Cited	Label
“what is [category] audit”	2.1 → 2.4	1,200 → 1,480	84 → 39	Yes	No	Overview zero-click
“best [tool] for [job]”	3.0 → 3.2	900 → 1,100	70 → 44	Yes	Yes	Overview zero-click (cited, clicks still down)
“[brand] pricing”	1.4 → 1.3	400 → 410	210 → 205	No	—	Control / healthy
“[service] near me”	4.1 → 17.8	800 → 190	62 → 7	No	—	Ranking loss
“how to [job]”	2.8 → 9.5	1,000 → 640	90 → 22	Yes	No	Mixed

Classification procedure:

1. Sort the sheet by clicks lost, not by CTR change.
2. Apply Overview zero-click only when position held and the SERP log shows an Overview.
3. Apply ranking loss when position broke, even if an Overview also exists — then add Mixed if both are true.
4. If impressions and clicks both fell and position held, check Trends and seasonality before you say Overview.
5. If one page’s queries split 50/50, the URL has two jobs. Do not pick a single narrative for the domain.
6. Count labels. The majority label on click-weighted queries is the story you take upstairs.

Click-weight or you will let twenty glossary queries outvote the one pricing query that pays rent. Illustration only (not a client result):

Slice	Query count	Clicks lost in the window	What “majority” would claim
Overview zero-click	18	420	“It’s all Overviews” if you count queries
Ranking loss	4	610	The actual click hole if you count clicks
Mixed	3	90	Noise unless you isolate those URLs
Brand control	6	15	Not a story; it is the control

Take the 610 upstairs, not the 18. Then file ranking tickets on those four strings and Overview tickets on the eighteen. Both can be true. Query-count majority is how a glossary cluster hijacks the narrative.

Label	First ticket	Wrong ticket
Overview zero-click	Change what that page is for — see the traffic-drop response spoke	Core-update panic, mass title rewrites
Ranking loss	Indexing, competing URL, helpful-content, ranking updates	“Get us in the Overview” as the only fix
Mixed	Two tickets, two owners	Averaging them into one OKR
Demand	Inventory, offers, seasonality	Publishing 40 posts into a cooling query
Technical	Crawl, noindex, site move	Content calendar
Noise	Longer window	A strategy off three impressions

Being cited and still losing clicks is allowed. Citation is not a promise that informational CTR returns. Log it as Overview zero-click with a citation note, then decide whether the page’s job is brand and quotation rather than sessions.

Which confounders fake an Overview drop?

Several failures share “clicks down.” Only one of them is an Overview. Google’s debugging guide already lists the non-Overview causes; run that list before you brief leadership on generative search.

Confounder	What GSC / Search typically shows	How you rule it in
Seasonality / cooling demand	Impressions and clicks down, position often stable	16-month chart; Google Trends on the same strings
Algorithmic update	Position worse across many terms	Date lines up with a posted ranking update; small vs large drop per Google’s examples
Technical / indexing	Impressions collapse on a URL set	Page indexing spike, noindex, robots, site move
Spam / manual action	Broad ranking loss	Manual Actions report; spam policies
Security interstitial	Clicks die harder than rank	Security Issues report
Cannibalization	Property “still ranks”; the wrong URL is now topmost	Pages tab: click loss on URL A, gain on URL B for the same query
Site move	Temporary rank swing, then a stall if redirects are wrong	URL changes in the drop window
Reporting glitch	Shape that matches nothing operational	Search Console data-anomaly notes linked from the debugging guide
Title / snippet miss	Impressions hold, clicks fall, no Overview	SERP log is empty of Overviews; then titles are on the table
Device or country mix shift	Property CTR moves; query-level does not	Split device and country before you tell a story

Checklist when someone says “it must be AI”:

• 16-month Performance chart reviewed for last year’s rhyme
• Drop date compared to a known ranking update
• Page indexing / crawl stats checked for a spike
• Manual Actions and Security Issues opened once
• Brand series isolated
• At least 20 SERP logs in the selling country
• Pages tab checked for cannibalization

If impressions and clicks both fell, you do not get to skip this list. Overviews can coexist with a demand drop. The panel still has to say so.

What if Overviews and ranking loss happen on the same URL?

That is the mixed row, and it is common. A glossary URL can slip from position 3 to 11 and pick up an Overview on the way. Blaming only Overviews hides the rank slip. Blaming only “the algorithm” hides the zero-click.

Split time, not opinions.

Question	If yes	If no
Did average position break before the Overview log showed yes?	Ranking-loss ticket first	Overview ticket can lead
Did impressions fall with position?	You left the SERP, not just the click	Zero-click is still in play
Is a different URL on your site now the topmost impression?	Cannibalization ticket	Stay on this URL
Are brand queries on this URL still healthy?	The URL still has a job	The URL may be losing the entity, not just a feature
Is the Overview citing you while blue-link position dies?	Two KPIs: citation log + rank recovery	You are losing both surfaces

Procedure for a mixed URL:

1. Plot that URL’s queries as two lines: average position, and clicks. Do not use CTR as the only line.
2. Annotate the first week your SERP log shows a durable Overview (3 of 4 weekly checks, not one sighting).
3. If position broke first, run indexing and competing-result work as if Overviews did not exist.
4. If position held through the Overview’s arrival, do not file a core-update ticket for that URL.
5. Keep mixed queries in their own slice of the exec chart so nobody averages them into a slogan.

Failure mode: one “organic is down” epic that assigns the same rewrite to ranking-loss URLs and Overview URLs. You will ship the wrong HTML on half the list.

How do I use brand queries as a control?

Brand is the experiment’s untreated arm. If people who already want you still find you, the site is not generally “deindexed.” If brand dies with non-brand, you have a bigger problem than an Overview on a what-is query.

Control slice	Healthy pattern during an Overview wave	Pattern that is not Overview-only
Exact brand	Clicks and position hold	Brand clicks fall with position — ranking, entity, or SERP hijack
Brand + service	Mild CTR pressure, position holds	Position collapse on “[brand] pricing”
Non-brand informational	Impressions up or flat, clicks down, Overview yes	Same click loss with no Overview and worse position
Non-brand commercial	Less Overview coverage on many sites; confirm on yours	Treating every commercial dip as zero-click

Procedure:

1. Tag every panel query brand / non-brand before you look at CTR. Do it in a column, not in your head.
2. Export two comparison views in Search Console: query contains brand name, query does not.
3. If brand clicks hold and non-brand informational matches the Overview fingerprint, you have a feature problem on those queries.
4. If brand and non-brand both lost position, stop the Overview-only narrative until ranking work is done.
5. Keep 5–10 brand queries on the weekly SERP log even when they look boring. Boring is the point.

Operator rule: never average brand and non-brand CTR into one “site CTR” and then blame Overviews. That average is how a healthy brand hides a bleeding glossary.

What usually fails first when teams run this split?

The diagnostic fails in predictable ways. None of them require a new tool. All of them produce a confident wrong story.

Failure	What it looks like in the room	Fix
Sitewide CTR as the test	One chart, one percentage, one villain	Panel, click-weighted labels
Mutating query list	“Top queries this week” every Monday	Frozen sheet, dated filename
No SERP log	“We assume Overviews”	20 screenshots in the selling country
Property average position	“We still rank #3 overall”	Query-level position + Pages tab
Logged-in SERPs	Founder’s Chrome, personalization on	Clean profile, matching country
Unequal date windows	14 days vs 30 days, or holiday vs ordinary	Equal length; 16-month context
Invented industry CTR	A LinkedIn screenshot in the deck	Your panel delta; no fake benchmark
Mixing search types	Web diagnosis polluted with Images	Search type = Web
Skipping indexing	Content sprint on a noindex cluster	Page indexing + URL Inspection
One-week Overview sighting	Strategy off a single SERP	Intermittent vs durable (repeat checks)

Checklist before you publish the diagnosis internally:

• Click-weighted label counts exist (not just query counts)
• Brand split exists
• Mixed is allowed
• At least one negative-control query still looks healthy — or you can explain why nothing does
• Nobody used a percentage you cannot source from this property

I have been SEO certified since 2021. The certification does not make Overviews kinder. It makes this split boring if you actually export the Queries table.

What do I do after every query has a label?

You stop arguing about the climate and you file tickets that match the labels. This spoke does not rebuild the response plan; that lives in If AI Overviews ate your clicks. Eligibility and quotation live in get cited in AI Overviews.

Majority (click-weighted)	Next motion	Not the next motion
Overview zero-click	Change page jobs on those URLs; citation + brand scoreboard	Restoring 2022 informational CTR as a moral right
Ranking loss	Indexing, competing results, helpful-content, updates	An AEO plugin
Mixed	Split the backlog by query	One rewrite brief for the whole domain
Demand	Offers and seasonality	Forty thin posts
Technical	Crawl / index / move	A new editorial calendar

Procedure once labels exist:

1. Count click-weighted labels. Write the majority in one sentence.
2. Hand Overview-majority URLs to the response playbook. Hand rank-majority URLs to technical and content quality work.
3. Protect brand and money URLs from being “optimized” into glossaries while you diagnose.
4. Keep the panel running. A label is a hypothesis you re-check weekly, not a tattoo.
5. If you cannot complete the split after two weeks of Search Console access, that is when a visibility audit is the right hammer — not after six weeks of vibes.

Do not open twenty content tickets on the day you finish the sheet. The sheet is the product of this post. The HTML changes are a different post.

How long should the panel run before I trust it?

A one-week screenshot is a sighting. A diagnosis needs a window that can survive a Tuesday.

Google already tells you to read 16 months for seasonality and to wait weeks after content changes before you re-judge rank. Use that patience on the Overview side too. Overviews do not always fire. Intermittent serving will fake a recovery if you sample once.

Horizon	What you can claim	What you cannot
7 days	“We spotted Overviews on these queries”	Ranking loss vs zero-click
14–28 days	First labels, still provisional	A budget freeze
90 days on a frozen list	Durable fingerprint vs rank trend	That next month will copy this month
Two quarters	Kill / keep / reinvent decisions	That a cited URL restored old CTR

Cadence after the panel exists:

Cadence	Artifact	Done looks like
Weekly	20 money queries: GSC export + SERP log	Overview yes/no and cited yes/no filled
Weekly	Brand vs non-brand clicks and position	Two lines, not one blended CTR
Monthly	Full panel relabel	Mixed and unknown shrink or are explained
Monthly	16-month chart glance	Seasonality not forgotten
Quarterly	Click-weighted label mix	Strategy matches the mix, or you change strategy

If you only have a week, skip the perfect 80-query panel. Run twenty click-leaders, the 16-month chart, brand split, indexing check, and twenty SERPs. That is enough to stop a homepage panic. It is not enough to freeze a year of content strategy.

Trust the split when labels stay put week to week on the same strings. If labels thrash, you have intermittent Overviews, a short window, or a query list you keep editing. Fix the method before you fix the site.

FAQ

How do I know if my traffic drop is from AI Overviews vs a ranking loss?

Run a frozen query panel in Search Console. Overview zero-click is average position stable, impressions flat or up, clicks down, and an Overview present on that query. Ranking loss is average position worse, usually with impressions down. Do not read it from sitewide CTR.

How do I measure whether this Overview vs ranking-loss split is working?

The split is working when the same queries keep the same labels week to week, mixed and unknown shrink, and tickets match the labels. Count click-weighted labels, not a vibe. If the panel mutates or the SERP log dies, you are back to anecdotes.

What usually fails first when teams try this?

They diagnose from property-level CTR, rebuild the query list every week, and skip the SERP log. Logged-in personalized results and unequal date ranges are close behind. Fix the method before you rewrite pages.

How long does this take to show results?

You can spot Overviews in a day. Durable labels take two to four weeks on a frozen list; a 90-day panel is what you should trust for strategy. Google already asks you to wait weeks after ranking-side changes before you re-judge position.

What should I skip if I only have a week?

Skip the 80-query research panel, the industry-CTR scavenger hunt, and any homepage redesign. Do twenty click-leading queries, a 16-month chart, brand vs non-brand, an indexing check, and twenty SERP logs in the selling country.

When is this not worth doing yet?

Skip it if you do not have Search Console, the site never had query-level traffic, or the “drop” is a single day. Fix crawl and indexing first if Page indexing is on fire. A brand-new property needs a baseline before a split test.

CTA

If organic fell and nobody can say whether the Overview or the rank moved, freeze the query panel before you freeze the budget.

Lane: /visibility · Book a visibility audit.