Microsoft Advertising Evidence

A better record, kept for a quarter as long

Microsoft Advertising makes a trade that runs in the opposite direction from Google's, and understanding the trade is most of what an attorney needs from this page. Field for field, Microsoft's change record is the more useful of the two. It gives old value and new value in separate structured columns. It names the mechanism behind each change — whether an edit came from a person in the web client, from a script, from a bulk upload or from an API integration. It treats users as a tracked entity type, so permission changes fall inside the report's scope.

And it is retained six months. Google's interface change history runs two years. On a Microsoft Advertising matter the preservation clock is therefore roughly a quarter as long, which means the same delay that costs nothing on a Google dispute can cost the entire change record here. That is the whole point of this page, and if a reader takes one thing from it, that should be the thing.

The verdict is still in the account, because everything described below exists inside the advertiser's own account and exports from it. Whether the conduct the record shows matters legally is a question for counsel; I am not an attorney.

The columns the change history report gives you

The Search Campaign Change History Report exposes twenty-one structured columns. The ones that carry weight in a dispute are these.

• DateTime — date and time of the change in the campaign time zone, and the one column the report requires.
• ChangedBy — the username of the person who made the change, or Administrator where the change was system-made.
• Tool — how the change was made: API, automated rules, bulk upload, imports, Editor, mobile apps, scripts, web client, and others.
• HowChanged — added, updated or deleted.
• ItemChanged — the entity type: account, campaign, ad group, ad, keyword, call extension, location extension, sitelink extension, negative keyword list, and user.
• AttributeChanged — the specific property that changed, left blank where an entity was added or deleted outright.
• OldValue and NewValue — the before and after, in separate columns rather than in prose.

Identifiers and names sit alongside those at account, campaign and ad group level, together with the ad title, ad description, display URL and keyword. The result is a table that can be sorted, filtered and reconciled against performance data without anyone having to parse a sentence, which is the practical difference between a record that can be analyzed and one that has to be read.

Six months against two years, and everything in between

These are the two platforms' published windows side by side, read August 14, 2026 in Microsoft Learn and Google Ads Help.

Microsoft's six-month change history figure applies at every granularity — hourly, daily, weekly, monthly, yearly and summary alike. There is no longer-lived coarse version of the change record the way there is for performance data, so nothing survives the six months in reduced form (Microsoft Learn, Reporting Data Retention Time Periods, read August 14, 2026).

Several other Microsoft reports run on their own clocks and they are easy to miss. The Budget Summary Report holds twenty-four months at daily granularity. The Call Detail Report holds six months. The Asset Performance Report holds thirty days. Hourly detail for search query, product search query, goals and funnels, and segments is one month. Where a theory of the case depends on time-of-day patterns or on call volume, those are the numbers that decide whether the theory can be examined at all.

What the Tool column settles that Google's log cannot

In an agency dispute or an unauthorized-change dispute, the question is usually not only what changed but how. A budget raised by hand in the web client at eleven at night is a different fact from the same budget raised by an automated rule that had been running for months, and a different fact again from a bulk upload pushed by an integration. Microsoft records that distinction in a column. Google's change history covers changes made through automated rules, the API and Editor as well as the interface, but it does not offer a clean equivalent field, and its automated changes surface under generic system identifiers.

The second distinction is user tracking. Microsoft lists user as an entity type inside the change history report, so changes to permissions and access appear in the same record as changes to campaigns. Google's change history expressly does not track password changes, which leaves account-control theories dependent on evidence from outside the platform.

Neither of those closes the identity gap, and a report should say so before it is asked. Naming the tool does not name the human. A shared login, an agency's operational account or a script running under a service credential still collapses several people into one username, and ChangedBy reading Administrator tells you the system made the change without telling you which configuration caused it.

Two reports that keep no history at all

This is the item on this page that most often arrives too late to fix. The Negative Keyword Conflict Report and the Product Match Count Report retain no historical data. They are point-in-time views: they describe the account as it stands when the report is run, and there is no version of them for last quarter.

That matters because negative keyword conflicts are exactly the kind of finding a mismanagement theory rests on — a negative keyword blocking the advertiser's own best-performing term, sitting in a shared list nobody reviewed. If that conflict was corrected before anyone ran the report, the state that produced the harm is unrecoverable from the platform. The same is true of product match counts in a shopping dispute.

So on any Microsoft matter where campaign configuration is in issue, a contemporaneous snapshot of both reports is worth taking on day one, before anyone touches the account to investigate. It costs minutes. Reconstructing the same state later from a change history that only reaches back six months is, in most cases, not possible at all — and where it is possible it is inference rather than record, which is a materially weaker thing to put in a report.

Invalid clicks: three buckets, one of them billed

Microsoft classifies clicks in three tiers — standard quality, low quality and invalid — and states that advertisers are billed for and pay only for standard-quality clicks. Invalid clicks are described as originating from spiders, robots, questionable sources or test servers. Where Microsoft bills for clicks it later determines were low quality or invalid, it states that the bill is adjusted, and that automatic account credits are issued where invalid clicks are suspected to come from search engine robots, automated click tools or fraudulent means (Microsoft Advertising Help, Monitoring clicks, read August 14, 2026).

Microsoft's published detection methods are categories rather than algorithms: analyzing search patterns, recognizing different types of click activity to distinguish malicious from accidental and valid clicks, and using campaign performance data to identify invalid activity. No thresholds are published, and no click-level ledger is exposed to the advertiser.

The consequence is the same as on the Google side and it should be stated the same way. A credit is documented as an automatic consequence of filtration. It is not an adjudication of anyone's intent, it does not identify who sent the traffic, and it does not describe what the filtered traffic was. An account credit is evidence that the platform's systems classified some traffic as invalid, and that is the extent of what it supports on its own.

What is not documented, and what I decline to assume

Three gaps are worth naming, because writing around a gap is safer than papering over it.

First, the six-month change history figure is solidly documented for the reporting API. The corresponding Microsoft Advertising help article describing the window as displayed in the web interface did not resolve when I went to read it, so I do not publish an interface number and I do not assume parity with the API. If the interface holds something different, that has to be verified in the live product before it goes in a report.

Second, Google restricts its search terms report by a privacy volume threshold. I have not located anything in Microsoft's primary documentation establishing a comparable restriction on its search query reporting, and I will not assert parity in either direction. Assuming Microsoft behaves like Google is one of the more common ways a paid-search opinion gets into trouble.

Third, Google Ads holds current accreditation from the Media Rating Council for clicks and invalid clicks, with an accreditation letter dated March 31, 2026. I have located no equivalent letter for Microsoft's advertiser-facing metrics. That absence is not a finding about Microsoft's measurement quality; it means the question should be checked against the Media Rating Council's published accreditation list rather than answered from either platform's own description of itself.

What I ask for first on a Microsoft matter

The order is set by the clocks, not by importance. The change history report goes first, pulled across the full six months and with every column included rather than a narrowed selection — a column layout chosen for readability is how a production ends up missing the field that decides the case. Then hourly search query detail, which is one month wide. Then the Asset Performance Report at thirty days. Then contemporaneous snapshots of the two reports that keep no history.

After that, the records with longer memories: performance at daily granularity across thirty-six months, budget summary across twenty-four, and billing statements across thirty-six. Alongside all of it, the advertiser's own server and CDN logs, which sit outside the platform entirely and are the only traffic record the advertiser controls end to end.

There is one thing this sequence will not do, and it should be said plainly. Nothing in it recovers a Microsoft change record older than six months. If the conduct in dispute happened a year ago and nobody exported the change history at the time, the log of who changed what is not somewhere else in the account — it is not anywhere. What remains is performance data, billing, and whatever the advertiser or its agency kept in their own files, which is a reconstruction rather than a record, and I would describe it in those terms.

Frequently Asked Questions

Published 2026-08-14