What a PPC Expert Report Contains

The test the whole document has to pass

A paid-search report has one structural obligation, and every section below serves it: somebody else, handed the same files, should be able to reproduce every number in it without asking a question.

That sounds modest and it is not. It rules out summarizing an export instead of citing it, a figure whose date range and filters are not stated, and a conclusion resting on the author's judgment rather than on something in the record. It also forces the document into an order that is not the order of the argument: the record first, the arithmetic second, the observations third, and the things the analysis cannot show stated rather than left for cross-examination to find.

What follows is what goes into each part and, at the end, what a report that reaches past its evidence looks like. Bill Hartzer is not an attorney; nothing here describes what a report must contain as a matter of procedure, only what makes the substance hold.

The assignment, and what was not asked

The first section states the questions counsel posed, in the form they were posed, and nothing else. Its usefulness comes from what it excludes.

Paid-search matters generate scope creep faster than most, because an account holds so much that is adjacent and interesting. A report that answers the questions counsel asked and then adds observations nobody requested has widened its own attack surface for no benefit. If something outside the assignment looks material, the move is to raise it with counsel and let the assignment change on the record.

This is also where the disclaimers that matter belong, stated once: that the analysis addresses what the account record shows and not whether any conduct breached an agreement or supports a claim; that no lost-profits figure is offered; and that the examination is bounded by the period, platforms and accounts named.

The materials relied on, itemized to the file

A list of exhibit names is not a materials section. What belongs here is an inventory precise enough that a reader can tell where every number came from: each file by name and format, the range it covers, the account it came from, who produced it, the date it was received or exported, and a hash for anything relied on numerically.

Two distinctions do real work. The first separates records exported directly from an account under observation from records received in production. The second separates a live export from a screenshot; a screenshot preserves a rendering at one moment without the metadata, the full range or the ability to re-segment, and where one is relied on, the capture time and time zone belong next to it.

Two export traps deserve naming here rather than being discovered later. Search Ads 360 displays times in the browser's time zone but converts all timestamps to GMT on download, so an export and a screenshot of the same screen show two different times for the same event. And a download only captures the columns currently visible in the table, so a narrow export silently omits fields that existed in the record. Both are documented in Google's Download a change history report, read August 14, 2026. A thin production is not necessarily incomplete data; it may be a complete record exported badly, and the difference is worth establishing before anyone argues about it.

How the data was obtained and verified

This is where a report earns or loses its credibility, because it is where discrepancies get explained before anyone else raises them. Paid-search data does not reconcile cleanly, and a report presenting it as though it does either has not looked or is hiding it.

Three known mismatches should be stated as expected behavior, with sources. The interface change history and the API change feed are not the same record: Google's developer documentation warns that a change event may not include every row from the Change History in the web client, and Google Ads Editor changes surface in the interface while documented as unsupported in the API feed. Search terms rows will not sum to campaign totals, because low-volume queries are withheld while their clicks and cost remain in the aggregates — expected, not evidence of tampering. And Google Ads will not reconcile to Analytics: one counts clicks and the other sessions, one dates a conversion to the click and the other to the conversion, and they attribute across different channels.

Where the record was verified against something outside the platform — invoices, payment records, the client's own order data, tag manager version history — the method and the residual difference belong here too, stated as a figure rather than an adjective. The change-feed caveat is from Google's Change Event documentation, read August 14, 2026.

Findings, in the order the record produces them

The order below is not rhetorical. It is the order in which each layer becomes checkable, so a reader can stop at any point and know what has been established.

1. Account identity, ownership and access. Which accounts existed, who created them, which manager account held ownership and when, the link and unlink history, and every user with access, at what level, granted and revoked on what dates.
2. Structure and settings, with their change history. Campaign types, networks, location targeting and the presence-versus-interest option, ad schedule, device modifiers, negative keyword coverage, and bidding strategies and their targets — each with the dates it was set and changed and the user recorded against it.
3. Spend. Cost by day and campaign across the period, with budget changes overlaid. Where daily overspend is alleged, the platform's documented behavior — a campaign may spend up to twice its average daily budget on a given day, with the month capped at 30.4 times that budget — has to be applied before any figure is stated.
4. Conversion measurement integrity. Which conversion actions existed, their creation dates, whether tags were detected, whether each was included in the reported conversions column, whether an automated bidding strategy was optimizing toward an action that recorded nothing, and whether any definition changed mid-period.
5. Reconciliation. Platform cost against invoiced cost, and platform-reported conversions against the client's own records, month by month.

Each finding is written the same way: what was observed, where in the record it sits, the date range it covers, and what it is inconsistent with. No adjectives, and no inference the reader has not been given the material to draw for himself.

Show the arithmetic

Every number should carry the recipe that produced it: source file, date range, filters, columns, and any rows excluded, with the direction of the exclusion stated. Where a calculation ran in a script, the script is part of the work product and should be re-runnable against the same inputs.

Comparisons belong in tables rather than sentences, because prose hides a comparison's basis. A table forces both figures, the periods they cover and the source of each into one frame, which makes a mismatched basis visible immediately.

Time zones deserve their own line, because they are the quietest source of error here. Platform reporting runs in the account's time zone, exports may convert to GMT, and server logs usually run in UTC. A finding that a change happened before or after some event can invert on a time zone nobody stated.

The limitations section, which a weak report omits

A paid-search report without a limitations section is missing the part that makes the rest credible. These are not caveats added for safety; each is a specific, checkable property of the data, and an opposing expert will raise every one left out.

• Reported conversions are not a count of events. Google reports observed and modeled conversions blended in the same column, and the standard interface offers no way to separate them. A report cannot say how many of a stated conversion figure were observed. Google's About modeled online conversions documentation, read August 14, 2026, is the citation.
• The search terms report is incomplete. Low-volume queries are withheld, so any figure for irrelevant or off-target spend derived from it is a floor and must be stated as one.
• Change history identifies an account, not a person. It records the email attached to a change, and shared logins, agency service accounts and API credentials collapse several people into one identifier. It does not record password changes, does not always list account-level settings changes or changes made by platform representatives during consultations, and truncates detail on very large bulk edits.
• Absence of change events is not absence of work. Analysis, reporting, creative development and a deliberate decision to leave a working account alone all produce no change events.
• Invalid-click figures are counts, not ledgers. The platform discloses how many clicks it filtered, never which ones or why.
• Share-of-voice metrics disclose nothing about a competitor's economics. No bid, budget, cost or click volume, and nothing at all below a 10% impression share.
• Location reporting shows where the platform determined a user was. It is inferred from signals, and the platform states that complete accuracy is not achieved in every situation, so a geographic finding carries an error term.
• Retention limits what was examinable. The report states which windows applied on the date of examination, and what could therefore not be reviewed.

What overreach looks like

Reports here fail in the same five ways, each visible from the table of contents.

1. An opinion that a manager fell below professional standards. There is no published standard of care for paid-search management — no licensing body, no accrediting body, no association with practice standards and disciplinary authority. An opinion framed that way rests on a document that cannot be produced.
2. A vendor percentage converted into a loss figure. A third-party detection tool classifies traffic under its own definitions, after the click lands, on the advertiser's side. The platform filters under different definitions, before billing. Subtracting one from the other produces a number with no defined meaning.
3. A before-and-after difference presented as damages. The delta has to survive seasonality, competitive entry, the advertiser's own pricing and site changes, and platform product changes — and a check that the measurement itself did not change, since a redefined conversion action or an attribution change makes the two periods different instruments.
4. A rendering of an ad nobody preserved. Responsive search ads assemble headlines and descriptions per auction. The record holds the asset set and asset-level serving data; a claim to show the exact creative one person saw, absent a screenshot, is reconstruction presented as observation.
5. A lost-profits number from a paid-search expert. Different discipline. The paid-search contribution is the predicate — what ran, what it cost, which spend is unrecoverable, and whether the compared periods are comparable — and the quantification belongs to a forensic accountant or economist.

A sixth is subtler: a report that treats a platform's own commercial metric as a quality measure. Optimization score measures how many of Google's recommendations an account has adopted, and Google states that partners need not adopt all recommendations or reach a perfect score. Offering it as evidence of account quality in a dispute about overspending argues the opposite of what it is offered for.

The material that surrounds the analysis

Several things sit outside the findings and earn a place anyway.

A record of what was requested and not received. Often more consequential than the list of what arrived. Noting that the manager account link history, the tag container version history and the platform billing export were requested on stated dates and not produced tells a reader which findings could not be made, and why.

A contemporaneous engagement log. Dates of access, what was exported and when, communications that changed the scope, and the order in which materials arrived. Reconstructing it afterward is unconvincing; keeping it as the work proceeds costs nothing.

A statement of how the data was handled. Where exports were stored, how they were hashed, whether anything was transformed before analysis, and whether any automated tooling was used and to do what.

Nothing that inflates the author. Credentials belong in the qualifications section as facts. A platform vendor's product certification is a timed exam on that vendor's own product, valid for a limited period, and treating it as a professional qualification overstates it in a document whose whole value is not overstating things.

After the report leaves my hands

A finding has to survive being challenged, which is the practical reason for every choice above: a number tied to a named file, a stated date range and a repeatable method is defensible in a way a summary is not.

One property of paid-search evidence is worth stating and then handing on. An export is not self-proving simply because a platform generated the underlying data. A file downloaded from an advertising account is a document a party produced, and who produced it, from which account, on what date and by what method becomes a question in its own right. That is why the materials section carries production dates and hashes, and why a screenshot is a weaker artifact than an export of the same record.

What happens to those questions procedurally is deliberately not covered here. This site describes what the paid-search record contains, how it is obtained and verified, and what it does not show.

Frequently Asked Questions

Published 2026-08-14