Google Ads Change History: Diagnose Drops Fast (2026)

Monday morning. CPA is up 50%, conversions are down by half, and the client wants an answer today. The instinct is to start rewriting ads and moving bids. The correct first move is to open Google Ads change history and find out what was actually done to the account in the past two weeks. Eight times out of ten the cause is not “the algorithm” — it is one specific edit made by one specific person or automation.

This is a working protocol: where the report lives, how to read it, what it never records (this part matters most), how to separate cause from coincidence in fifteen minutes, and what to do when the log comes back empty.

What Google Ads change history is and why it is the first screen you open

Change history is a log of edits made to your account, campaigns and ad groups: who, when, what — with the previous and new value recorded. Google keeps two years of it. Anything older is gone.

The value is not the list itself. The value is that it turns “I think the algorithm broke” into a testable claim: “on 17 August at 14:22, user X raised tCPA from $40 to $65 — and here is the chart where impression share dropped 18 points a day later.” It is the only report in the interface that answers what changed rather than what we spent.

House rule that saves hours: until you have checked change history, you are not allowed to propose external causes. Internal causes take minutes to rule out. External ones take days.

Where to find it and how the report is built

In the 2026 interface:

Left navigation → Campaigns → Change history for the account-wide view.
To scope it, tick the campaigns or ad groups you care about on the Campaigns / Ad groups page and click Change history in the action bar. The report opens pre-filtered.
At manager (MCC) level the report is available per client account. There is no single cross-client change log in the UI — if you manage many accounts, that view has to be built via the API.

Four filters do the real work:

Date range. Never “last 7 days”. Use a window that starts 3–5 days before the drop began. Edits almost never show their effect on the same day.
Change type. Budgets, bids and strategies, status, keywords, ads and assets, targeting, campaign settings, conversions. Start with three: budgets, bids/strategies, status.
User. This is where you see who edited — including Google’s own automation (system entries such as ads-dartsearch-budget) and third-party tools connected through the API.
Campaign / ad group. Noise control on large accounts.

Each row carries “changed from” and “changed to” columns — that is the payload. One caveat straight from Google: the page limits the detail shown when a single change touches a very large number of entities. A 20,000-keyword bulk upload collapses into one unexplained line. Practical consequence: break big edits into batches, or in a month you will not be able to reconstruct what you did. Batch mechanics are covered in our guide to bulk operations in Google Ads Editor.

What the log records — and what it silently ignores

This is the most underrated section. Half of all wrong conclusions come from someone finding an empty log and deciding “so nobody changed anything.”

Recorded	Not recorded (or only partially)
Daily budget changes, with old and new values	Website changes: price, form, page speed, layout
Bid strategy switches and tCPA / tROAS target edits	GA4 and tag edits: broken events, redefined goals, consent changes
Pausing and enabling campaigns, ad groups, keywords, ads	CRM-side and offline-import failures
Keyword, negative keyword and negative list changes	Some account-level settings and edits made by Google reps during consultations
Ad, asset and extension edits	Merchant Center feed changes: price, availability, product disapprovals
Targeting changes: geo, languages, devices, audiences	External pressure: a competitor raising bids, seasonality, holidays, failed payments
Auto-applied recommendations, shown as system entries	Policy actions — disapprovals live in their own report
Actions taken by third-party tools via the API	Detail of very large bulk edits (roughly 2,000+ entities at once)

The takeaway: change history covers the inside of the account. If it is clean, move to the outside — tag, feed, site, competitors, billing. In that order, not the reverse.

The 15-minute diagnostic protocol

Run these in order. Each step costs more than the one before it, so the cheap checks go first.

Step 1 (2 min). Pin the inflection date

Open the daily chart for the affected campaign across 30 days and find the first day the line deviates — not the day you noticed. Write that date down. The whole investigation now lives inside “date minus 5 days to date plus 2 days”.

Step 2 (3 min). Sweep the three high-impact change types

Change history → your date window → filter by budget, bids and strategies, status. These three account for most sharp drops. Look at increases as well as decreases: raising a daily budget by 60% destabilises Smart Bidding just as effectively as cutting it.

Step 3 (3 min). Ask “who”

Clear the type filter and filter by user instead. Three questions: are there edits from people you did not expect; is a third-party tool writing through the API; are there system entries from Google. An unfamiliar account in that column is no longer a performance question — it is an access question, and it is handled separately.

Step 4 (3 min). Check conversions and campaign settings

Filter to “Conversions” and “Campaign settings”. Look for a changed primary/secondary conversion action, a changed attribution or conversion window, or a conversion action switched off. This is the nastiest category: the metric collapses while traffic is unchanged, because what dropped is not sales but counted sales. The mechanics are unpacked in our pieces on primary and secondary conversions and conversion windows and conversion lag.

Step 5 (2 min). Overlay the edit on the metric

Go back to the chart and mark the date of the edit you found. You are checking that the metric moved after the change, not before it. If the decline started two days earlier, you have found a colleague reacting to the problem, not causing it.

Step 6 (2 min). Rule out the outside world

If the log is clean, spend two minutes on three things: impression share (did a competitor outbid you), ad approval status, and whether conversions are still arriving in GA4. The first one is covered in our breakdown of impression share and auction insights.

Seven common scenarios and their fingerprints

Symptom	What to look for in change history	What is actually happening
Traffic drops to near zero overnight	Status change on campaign/ad group, geo or language edit, budget cut	Someone paused it “for a test” and never restored it
CPA up, volume up	tCPA raised, daily budget raised, caps removed	The strategy is executing a new target — expected behaviour, not a fault
CPA up, volume down	tCPA or tROAS lowered, audiences narrowed	The algorithm re-entered learning and lost the cheap auctions
Conversions hit zero, clicks unchanged	Conversions section: action disabled, primary action switched	A measurement failure, not a traffic failure
Sudden flood of irrelevant search terms	Negative list removed, match type changed, brand list edited	Reach expanded with no negative-keyword safety net
Small daily edits from “the system”	Filter by user: auto-applied recommendations	Auto-apply is on and is rewriting parts of the account
Everything moved right after “an optimisation”	A cluster of edits on one day: bids + structure + ads	Too many variables changed at once — cause is now unrecoverable

That last row is the most common and the most expensive. If strategy, budget and half the ads changed on the same day, change history will show you everything and tell you nothing. This is why “one meaningful change at a time, then an observation window” is not pedantry — it is the precondition for the report being useful at all.

Cause or coincidence: three tests

Direction of time. The cause precedes the effect by at least a day. Account for conversion lag: with a 5–7 day sales cycle, yesterday’s edit cannot have destroyed yesterday’s conversions.
Locality. The edit hit one campaign but everything dropped? Then the edit is not the cause — look for something shared: tag, site, billing, season.
Control group. If untouched campaigns behave the same way, you are looking at an external factor. That is the same logic behind proper testing — see experiments in Experiment Center.

One adjustment for the Smart Bidding learning period: some edits reset learning, and the following 5–7 days are unstable by design. If you are inside that window, do not diagnose a drop — wait for learning to finish, or you will be treating a symptom that does not exist.

Reverting: when “put it back” helps and when it hurts

You can select rows in change history and undo a subset of edits. The temptation is strong, but a revert is itself a change, and it restarts learning too.

Revert immediately when:

the edit was plainly wrong — a negative list deleted, the wrong geo selected, a needed location excluded;
less than 24–48 hours have passed and the algorithm has not rebuilt around it;
it concerns status or targeting rather than strategy targets.

Do not rush to revert when:

tCPA / tROAS was changed — going back is a second learning reset in a row, and the combined cost usually exceeds the original mistake;
more than a week has passed and the strategy has already accumulated data on the new target;
you are not yet certain this edit is the cause. Confirm first, act second.

If a target genuinely needs to come back, move it in 10–15% steps rather than one jump, and model the outcome first with bid, budget and target simulators, which exist precisely to show what a different target would have produced.

Building a process where diagnosis takes minutes

One meaningful change at a time. A strategy switch and a structural rebuild on the same day guarantees indistinguishable causes a week later.
Batch large edits. Stay under the threshold where Google collapses the detail (roughly 2,000 entities per operation), or the log will read “bulk change” and nothing more.
Individual logins, not a shared one. The user column is worthless if the entire team edits under one company email.
Annotate outside Google. The platform records account edits; it does not record site releases, feed updates or tag deployments. A three-column sheet — date, what changed, who — resolves half of all future investigations.
Alert before you investigate. Automated rules emailing you on a spend or CPA deviation catch problems earlier than any manual check, and custom columns put the deviation metrics on one screen.

A weekly ten-minute pass over change history fits naturally into the broader Google Ads account audit checklist — far cheaper than a quarterly archaeology session.

When the log is clean

Measurement. Fire a test conversion and confirm it lands, allowing for lag. A broken tag looks exactly like lost sales.
The site. Speed, uptime, form changes, price changes. Plenty of “Google Ads drops” are landing page redesigns nobody mentioned.
Competition. Impression share lost to rank versus lost to budget. A rising first number means you were outbid.
Policy and status. Disapproved ads, limited serving, disapproved products in the feed.
Billing. A declined card stops delivery instantly and leaves no trace in change history.
Seasonality. Compare with the same period last year, not with last week.

If delivery keeps stopping because of payment or account-level issues, that is an infrastructure problem rather than a reporting one — agency Google Ads accounts with managed billing remove the single-card dependency. The rest of the stack is listed under PPC Rebels services.

Common mistakes

Only looking at the last 7 days. A week-old edit surfaces as a “sudden” drop today.
Ignoring system entries. Auto-applied recommendations are real changes, sometimes large ones.
Treating an empty log as proof. No edits in Google Ads does not mean no edits in your funnel.
Reverting everything. A cascade of undos damages learning more than the original mistake did.
Losing the reasoning. “tCPA: 40 → 65” means nothing in a month if nobody recorded why.
Shared access. Kill the user column and you kill half the report’s value.

FAQ

How far back does Google Ads change history go?

Two years. Older data is not retained and cannot be retrieved through the interface or the API, so long-horizon analysis has to be exported and stored on your side in advance.

Where is change history in the 2026 interface?

Left navigation → Campaigns → Change history. Alternatively, select specific campaigns or ad groups with the checkboxes and click “Change history” in the action bar to open the report already filtered to those objects.

Can I see who made a change?

Yes — there is a user column. It is only meaningful with individual access, though. If everyone works under one shared login, every edit shows the same address.

Why are some of my changes missing from the report?

Three usual reasons: the edit touched too many entities and the detail was collapsed; it was an account-level setting that is not logged; or it happened outside Google Ads — in GA4, the tag, the Merchant Center feed, or on the website.

Are auto-applied recommendations logged?

Yes, they appear as changes from system users. If you keep finding edits nobody on the team made, review your auto-apply settings — several of those recommendations move bids and structure noticeably.

Can I undo a change directly from the report?

Some changes can be reverted by selecting their rows. Remember that a revert is a change too: it restarts strategy learning, so it makes sense within a day or two of an obvious mistake, not two weeks later.

How do I tell a real cause from a coincidence?

Three checks: the edit precedes the drop by at least a day; the campaigns that dropped are the ones the edit touched; untouched campaigns behave differently. Fail any of the three and you are looking at a coincidence.

The log is empty but performance collapsed. What now?

Work the external contour in this order: conversion measurement (tag, GA4, offline import), site and landing page, impression share and competition, policy status, billing method, seasonality.

Does opening change history affect campaign performance?

No — the report is read-only. What affects performance are the changes it records: target, budget and structural edits can reset the Smart Bidding learning period and make several days of data unrepresentative.

How do I export change history for a client report?

Use the download button in the interface to export CSV/Excel with your current filters applied. For recurring reporting, pull it through the API into your own warehouse — that removes both the two-year limit and the loss of detail on bulk edits.

Should I check change history when nothing is wrong?

Yes, five to ten minutes weekly. That is how you catch unwanted edits and automation drift before they turn into a drop rather than after.

My agency makes the edits and results are getting worse. What do I ask for?

Give the agency individual access, agree on “one meaningful change at a time”, and reconcile change history against the agreed plan weekly. It converts an argument about quality into a conversation about specific rows in a report.

The short version

Change history is the cheapest diagnostic in Google Ads and the only report that answers “what changed”. It covers the inside of the account — budgets, bids, status, targeting, conversions — and is blind to your site, tag, feed and competitors, which need a separate pass. Individual logins, batched bulk edits and one change at a time turn it from an archive into a working instrument, and turn a half-day investigation into a fifteen-minute one.

The natural next step is predicting the effect of an edit before you make it — see bid, budget and target simulators. If the problem turns out to be who you are reaching rather than how you are bidding, start with custom segments and intent-based audiences. More practical breakdowns are in the PPC Rebels blog.

Google Ads Change History in 2026: Diagnose a Performance Drop in 15 Minutes

What Google Ads change history is and why it is the first screen you open

Where to find it and how the report is built

What the log records — and what it silently ignores