On this page
Ask for the screen, not the slide
These are the questions a Doctor Ads Profit Forensics examination answers from the account itself, rewritten for the person who approves the agency’s invoices. They sit with our other Account Audits & Agency Reviews studies and assume you will never set a bid yourself.
Call the rule behind every question the screen test: a good answer points to a place in the account you can open, with a date and a number; a worrying one describes a strategy, a process or a mood. The test is not a trap: these screens are where a working agency spends its week.
Advertisers bring these questions to forums: the employee who ran the account leaves, or a hiring manager wonders what weekly work a large account needs. A thread proves the question, not the answer.
Ten questions follow, in four layers, each with good and worrying answers, the screen to ask for, a quick check and its limits.
What your bidding is told to buy
Automated bidding chases the primary actions inside the goals each campaign uses, so every later number inherits that definition. Google groups conversion actions, such as a form or a call, into goals; as a rule, an action counts for a campaign’s bidding only when it is set as primary and the campaign uses its goal.
1. What exactly does our bidding optimise towards, and can you show me where?
The answer should be a list of named conversion actions for each campaign, read from its settings rather than from memory. Google’s help says that when a goal is an account default, the “primary conversion actions within that goal” are used for bidding across all campaigns except those with campaign-specific goals. In our study of what campaigns bid on, a snapshot of 10 July 2026, the median campaign was set to bid towards 3 conversion goals. That can be deliberate, but someone should be able to name them.
A good answer sounds like: “This campaign bids on qualified leads and phone-call leads, both primary, through the account defaults. The brand campaigns use their own goals. Here are the settings.”
A worrying answer sounds like: “We optimise for conversions” or “The algorithm finds what works.” Neither names an action you can open.
Ask to see: Goals, then Conversions, then Summary, for the actions marked primary; then each campaign’s Settings and Conversion goals. Check custom goals separately: a custom goal can mix primary and secondary actions, and Google’s API documentation says a secondary action stays out of bidding unless a custom goal includes it.
Quick check: read your largest campaign’s conversion goals and find each goal’s primary actions in the Summary. Then ask which enabled primary actions no campaign goal uses. Google says these do not steer bids but “may be used to enhance predictions”, and making their goal an account default would bring them into bidding, except on campaigns with their own goals.
What it cannot prove: settings show what a campaign is told to value, not how much each goal sways a bid. If an agency script or dashboard reads include_in_conversions_metric, an API field Google calls deprecated, have it switched to primary_for_goal, as Google advises; otherwise its list of what your bidding buys may not match your settings.
2. Is each lead counted once per click, or every time?
Google’s guidance points to once per click for leads: you might want to track every sale, but “only one, unique lead per ad click”. Yet the same page makes Every the default for website actions, so a form can count every submission without anyone deciding it. In four out of five accounts that feed leads into bidding, at least one lead action in the conversions column counts every conversion instead of one per click, according to our lead-counting study.
A good answer sounds like: “Our lead forms count One. The phone-call action counts Every on purpose, because one caller can book several appointments; here is its repeat rate.”
A worrying answer sounds like: “That’s the default” or “Every gives the algorithm more data.” A default is not a decision, and a repeat enquiry with no separate value to the business should not teach bidding.
Ask to see: each lead action’s Count (Goals, Conversions, Summary, the action’s name, then Edit settings) and its Repeat rate, which Google defines as the conversions you would record under Every divided by those under One.
Quick check: for each primary lead action set to Every, read its repeat rate: would Every record more conversions than One? If yes, some clicks were counted more than once. A rate of 1 means none repeated in the period on screen, yet Every will count the next repeat. If no rate shows, ask whether there is enough data to judge.
What it cannot prove: the setting is a permission, not a measurement of duplicates. Google’s guidance is not a rule that Every is always wrong. One counts one conversion per click for each action, so a person who fills in two different forms after one click still produces two conversions.
3. Which soft events sit in the column that bidding learns from?
Soft events show interest without being a lead or a sale: page views, engagement, outbound clicks. The agency should list each one marked primary and say why it belongs in bidding. Google’s help says an action must be marked as a primary action, in addition to the account-default setting, to be used for bidding by default; a secondary action is left out of the Conversions column used for bidding but still reports under All conversions. Among accounts with meaningfully labelled actions, in about six in ten, page views, engagement, downloads, outbound clicks, directions or follow-on video views sit in the column that bidding learns from, according to our lead-counting study.
A good answer sounds like: “One soft action is primary on purpose: our awareness campaign bids on engaged visits. The rest are secondary, outside any custom goal, so they only report.”
A worrying answer sounds like: “More conversions help the algorithm learn.” Volume that is not money teaches bidding to buy volume.
Ask to see: the same Summary, filtered to primary actions, with each action’s category in view.
Quick check: count the soft actions marked primary and ask which campaign needs each one. An action that only needs watching can be secondary and still report, staying out of bidding unless a campaign’s custom goal includes it.
What it cannot prove: a soft event can be the honest goal of an awareness campaign, and a real lead form can hide under a page-view label chosen years ago. The label is a clue, not a verdict.
What changed this quarter, and who changed it
Change history is the nearest thing the account keeps to a timesheet, and it is still not a timesheet.
4. What did you change this quarter, and what was each change meant to fix?
A good agency answers with dated changes, each tied to a problem and to the result it should move. Google’s change history lists changes to your account, campaigns and ad groups during the past 2 years, and filters by date range, type of change or campaign. A recurring invoice is not evidence of account work; a quiet history is a reason to ask what was done instead.
A good answer sounds like: “Early in the quarter we split brand from generic search, because brand was flattering the cost per lead, and later moved the quote form to One. Here are both rows and what qualified leads did after.”
A worrying answer sounds like: “We optimise the account daily” or “Most of our work happens outside the platform.” The second can be true, so ask for that work with dates.
Ask to see: Campaigns, then Change history, set to the quarter, with the User column in view.
Quick check: find the changes the agency named, then scan for large ones it did not mention, such as budgets, bid strategies or goals.
What it cannot prove: a row is an edit, not an hour, and a quiet log is not proof of neglect, because landing pages, feeds, offers and analysis leave no row. Google also says that sometimes not all account-level settings changes, or changes made by Google representatives during consultations, are listed.
5. How much of that came from tools rather than people?
The agency should split the quarter’s changes into interface edits by named people and changes made by software: the API that third-party tools use, scripts, rules and recommendations. In our study of what gets changed in Google Ads accounts, criteria took 91.9% of recorded events, and 63.6% came through the API, automated rules, recommendations or an unclassified client, meaning changes Google does not categorise. Criteria are targeting instructions such as keywords, negative keywords and locations, and the study cannot split their share by type.
A good answer sounds like: “Most rows are our negative-keyword script and the recommendation types we chose to auto-apply. The interface rows are our team; here are the names.”
A worrying answer sounds like: “The log shows hundreds of optimisations.” In our change-event study, a bulk load of negative keywords produced one event per row in a few seconds, so a count of events is not a count of decisions.
Ask to see: the same view, read by the User column. Google’s change history page says it shows the email address of whoever made an interface change, while API changes may show “Google Ads API” or a tool name, and automated systems may appear as “Google Ads system”.
Quick check: if almost nothing in the quarter comes from a named person, ask who reviews what the tools do.
What it cannot prove: a tool can carry a person’s decision, and an interface edit can be trivial. Our 63.6% is a pooled mix of events, not a share of hours.
6. Who works on our account each week, and through which login?
You should get names, roles and the access route: a user on your own account, or an admin of the agency’s manager account. On Google’s access levels page, only Admin users can review each user’s authentication method and last login time, and admins of a linked manager account can manage your campaigns and, with ownership, edit user access in your account. It also warns that an account with a single administrator may lose access to its tracking tags if that person becomes unavailable: the forum story of the employee who left.
A good answer sounds like: “A specialist and a team lead work through our manager account, and one of your colleagues is an Admin. Here are the Users list and the Managers page.”
A worrying answer sounds like: “A dedicated team looks after you.” A team nobody can name cannot be checked.
Ask to see: Admin, then Access and security: the Users list with access levels, and Managers along the top of the page, including the Owner column.
Quick check: confirm that at least one Admin works for your company, then ask that person to read each user’s last login time.
What it cannot prove: a login is not work, and the page shows neither seniority nor hours. Ask whether each named person appears in your Users list or works only through the manager account; either way, their interface edits show their email address in change history.
What the agency steers by
Scores, waste and brand searches are where a report can look healthy while the money does something else.
7. When you quote optimisation score, what result did it predict?
A good agency treats optimisation score as a list of Google’s suggestions, not as a result, and can show what the accepted suggestions did. In our score-rank study, rankings of campaigns by score inside each account had median correlations of 0.05 with click-through rate and -0.05 with conversion rate. Both are close to zero: the score order said almost nothing about either rate.
A good answer sounds like: “The score rose because we accepted the keyword suggestions and declined the budget ones. Here is what those keywords did to the cost per qualified lead.”
A worrying answer sounds like: “We keep the score high, so the account is well managed.” A score is not an outcome.
Ask to see: the recommendations applied this quarter, found in change history, each next to the metric it was meant to move.
Quick check: sort your campaigns by optimisation score, then by conversion rate. If the orders look unrelated, the score is not telling you which campaign converts better. Cost per lead is a separate question about money.
What it cannot prove: our medians describe our study, not your account. They do not show that recommendations damage results, and they say nothing about profit.
8. Which search terms took money and produced nothing?
The agency should open the search terms report, sort the terms with no conversions by cost and say what it did about the top of that list. Waste concentrates: in our audit checklist ordered by the strength of evidence, the worst tenth of zero-conversion search terms held a median 48.8% of the spend in that zero-conversion pool.
A good answer sounds like: “These are the costliest terms with no conversion this quarter. We excluded the irrelevant ones and kept those that convert later in the CRM.”
A worrying answer sounds like: “Broad match needs room to learn.” It can be true, and it does not say which terms cost money.
Ask to see: Campaigns, then Insights and reports, then Search terms, filtered to zero conversions and sorted by cost.
Quick check: read the top of that list aloud. Any term your business would never sell to is a follow-up.
What it cannot prove: a zero in one window is not a permanent verdict, because some terms convert later or offline. The 48.8% describes the zero-conversion pool, not the whole budget, and it is not a forecast of what you would recover.
9. Does “performance is up” include brand searches we would get anyway?
The agency should report brand and non-brand search separately, because growth from people already searching for your name says less about the agency’s work than growth from searches without it.
A good answer sounds like: “Non-brand leads rose and brand was flat. Here are the two campaign groups side by side, same date range, same conversion definition.”
A worrying answer sounds like: “Overall conversions are up.” Ask which part of the account produced the increase.
Ask to see: the Campaigns table split into brand and non-brand campaigns, for this quarter and the last, with conversions defined as in question 1.
Quick check: ask which campaigns carry your brand name as a keyword, then compare their share of conversions this quarter with the last.
What it cannot prove: a brand split does not measure how many of those customers would have come anyway; only a controlled test can. It does show whether the headline rests on the easiest demand in the account.
What the quarterly report proves
The report is where the first nine answers either show up or disappear.
10. What was the result, how is it defined, and what is the next decision?
Every number in the report should carry a definition, a comparison and the decision it informs. Conversions in the report mean whatever question 1 found the account counting. The report’s first line should say which actions that includes.
A good answer sounds like: “Qualified leads from non-brand search rose this quarter. Conversions means our primary lead actions, counted One, with the CRM figure for the same period beside it. Next, we test a new target on the generic campaign.”
A worrying answer sounds like: a deck of impressions, clicks, tasks and scores, with no line that says what the business got.
Ask to see: the definition behind each conversion number, the CRM figure for the same period, and one decision per number.
Quick check: for each number on the summary page, ask: compared with what, defined how, and what changes because of it? A number without all three is decoration.
What it cannot prove: a clear report can still rest on a broken definition, which is why questions 1 to 3 come first. Nor is a long report misconduct; careful agencies write them too.
What to check in your account
You can run these checks yourself before the review.
- Open Goals, then Conversions, then Summary. Note every action marked primary, its category and its Count (One or Every), and read its Repeat rate in the conversion actions table.
- Open Campaigns, then Campaigns, and select the Settings tab. For each large campaign, expand Conversion goals, note whether it uses the account-default goals or its own, and flag any custom goal.
- List the enabled primary actions that no campaign goal uses. Decide what should happen to each one before anyone adds a goal.
- Open Campaigns, then Change history. Set the quarter as the date range and read the User column: named people, API or tool names, and system entries.
- Open Admin, then Access and security. Check that someone at your company holds Admin access and ask that person to review each user’s last login time, then select Managers along the top to see linked manager accounts and the Owner column.
- Open Campaigns, then Insights and reports, then Search terms. Filter to zero conversions, sort by cost and read the top of the list.
- Take last quarter’s report and mark every number that has no definition, no comparison or no decision attached.
When this does not apply
The agency’s work is mostly outside the account. If you pay for creative, landing pages or product feeds, change history will be quiet by design, because the log records edits inside Google Ads, not work done elsewhere. Ask for a dated record of that work and judge it by what it moved in the results.
Your campaigns bid on value across several goals on purpose. With value-based bidding and a different value attached to each goal, a long goal list is the strategy rather than a mistake. Then ask where each value comes from and who keeps it current when prices or margins change.
Your account is short of conversions. Moving lead actions from Every to One can lower the reported count from that day, and Google’s counting help says count changes apply only to future conversion reporting. If valid leads are already scarce, ask how the agency will grow genuine leads and manage the drop in counted conversions; a cautious answer to question 2 is not a worrying one here.
When to call an independent check
Count the worrying answers, not the slides. One worrying answer is a follow-up for the next meeting; a good agency can have one weak corner. Two or more are a pattern. A pattern in questions 4 to 6 is about attention; a pattern in 1 to 3 is about the numbers themselves, because every number in the report is built on those definitions.
Give the agency the fair version of the test first. Send the questions in advance and accept “let me check and come back” once. An agency that comes back with screens has passed; one that comes back with a new deck has answered a different question. Keep each answer with the screen it pointed to, and ask the same questions next quarter.
A pattern is not a verdict on the people. It means the agency’s explanation needs checking; if the gaps sit in questions 1 to 3, so do the report’s numbers. If two or more answers come back vague, an independent one-week check of the account can investigate those gaps; that is what Profit Forensics does.
Researcher’s take
I have sat on both sides of this table, as the agency answering and as the auditor asking. The agencies I trust are not the ones with the best slides. They open the account when I ask a question, and say “let me check” when they do not know. That reflex is most of the screen test. I would not advise firing anyone over one vague answer, and I would not advise renewing an agency that cannot show what the bidding is told to buy. Ask for the screen, and judge the agency by how quickly it opens.
Igor Ivitskiy, Doctor Ads Profit Forensics
FAQ
What should I ask our paid search agency before the quarterly review?
Ask what bidding buys, what changed and who changed it, what the agency steers by and what the report proves. Start with lead counting: in four out of five accounts that feed leads into bidding, at least one lead action in the conversions column counts every conversion instead of one per click.
How do I know if my Google Ads agency is actually working on our account?
Ask for dated changes from change history, which covers the past 2 years, plus dated work outside the account and the results both should move. A busy log does not settle it, and nor does a quiet one.
What does a good answer sound like when we ask what bidding optimises to?
It names each campaign’s conversion actions, says which are primary and whether the campaign uses account defaults, its own goals or a custom goal, then opens the settings. Expect a list: in our published snapshot, the median campaign was set to bid towards 3 conversion goals.
Who exactly logs in to our account each week?
Only an Admin can review each user’s last login time, according to Google’s access table. Edits by people working through the agency’s manager account show in change history. A login shows access, not work.
How we know
Google’s documentation, all checked 19 October 2026. Google Ads Help: About change history; About conversion goals; About account-default conversion goals; About conversion counting options; About access levels in your Google Ads account; About ownership of client accounts; About the search terms report. Google Ads API documentation: Conversion goals; ChangeClientType. Menu paths follow Google’s help pages on the same date; your interface can differ slightly.
Figures from our published studies: 3 goals for the median campaign from our conversion goals study (snapshot of 10 July 2026); four out of five and about six in ten from our lead-counting study (July 2026 settings snapshot); 91.9% and 63.6% from our change-event study (12 June to 9 July 2026); 0.05 and -0.05 from our score-rank study; 48.8% from our evidence-ordered audit checklist. All describe a managed portfolio, not a random sample of advertisers, and settings or events, not hours or profit.
Opinion, not measurement: the screen test, the sample answers, the brand split and the rule of two worrying answers are Igor Ivitskiy’s practitioner positions. Forum questions are paraphrased without names.
No new portfolio data in this article: it combines Google’s documentation (checked 19 October 2026) with what our audits look for. Where a figure comes from our published studies, the study is linked.
The first questions mirror where a Profit Forensics examination starts: the conversion source, the campaign goals and the change history.