Turn a structured account review into a manageable consulting decision
Google Ads audits for freelancers: review findings without losing the client's context
Understand the checks, inspect the relevant targets and choose a justified next step. Use ppctor's shared audit evidence while preserving commercial context, unavailable sections and human approval for live changes.
Define the freelance audit assignment
A freelance Google Ads audit is a structured review of a client's account against defined checks and the client's actual objective. It should establish useful next decisions, not produce a long list of changes solely to demonstrate activity.
The assignment can differ between onboarding and an established engagement. During onboarding, you may need to understand the purpose of inherited settings. During a periodic review, you may be checking whether the agreed configuration still fits a changed offer.
Define the deliverable before running the review. Is the client expecting a configuration assessment, a performance investigation or a proposal for supported edits? These can inform each other, but they are not the same service outcome.
For an independent consultant, the practical challenge is managing review time without losing detail. A shared audit engine can collect defined evidence consistently. Your interpretation still needs the client's constraints, intended outcomes and decision authority.
Be clear about what the review does not test. Reading settings and provider reports does not independently test every landing page, sales process or measurement implementation. Those checks may require a separate appropriate investigation.
A good audit can conclude that a setting is intentional, that a decision needs more context or that a supported change is justified. It does not need to turn every recommendation into an executed operation.
Gather the client's operating context
Ask what the client sells, where it can serve customers and which recorded outcome matters. An account can be configured in a way that looks unusual but fits a particular service arrangement or commercial constraint.
Establish the chosen conversion definition. A form submission, call and purchase are different outcomes. A client wanting qualified opportunities may need business information that a generic conversion total does not provide.
Identify relevant external ownership. Tracking templates may support another measurement system, landing-page changes may belong to a developer and product-feed corrections may need a catalogue owner. The freelancer should not remove settings merely because their purpose is unfamiliar.
Ask about deliberate exclusions and planned pauses. An inherited account may contain product selections, location restrictions or campaign statuses that reflect current stock and service availability. Audit evidence can identify them without establishing whether they are mistakes.
Confirm the scope of your authority. Access for review is not necessarily authority to change every resource. ppctor's conversational execution also uses its own connected-owner and proposal-context controls.
Keep these facts beside the audit record. The person reviewing a later finding should be able to understand the client requirement without reconstructing an onboarding call or assuming the account's current setting expresses it perfectly.
Understand the audit's structure
ppctor's audit combines defined checks into sections. Account settings, conversion configuration, keywords, ads, assets and performance views can appear where applicable. Each finding has its own scope and interpretation.
A check can inspect a provider field, a collection of entities or a heuristic recommendation. These are different kinds of evidence. Read the finding's description and affected targets rather than infer its meaning from the colour alone.
Applicability matters. Performance Max uses asset groups rather than the ad groups examined by an empty-ad-group check. A Search keyword diagnostic should not become a claim about a campaign type that has no such keywords.
The Slack summary and the full audit use the same underlying run. They are different presentations of evidence, not separate independent opinions. A brief reply cannot carry every section's detail and caveat.
Performance sections require their own returned data. A configuration check can succeed while a performance read fails. The result should preserve that distinction rather than portray the whole account as either entirely measured or entirely unavailable.
For a freelancer, this structure helps organise the investigation. Group related questions by the decision they support, while retaining each check's actual target and evidence type.
Read pass, issue and review accurately
A pass means that the returned evidence met the specific check's expectation. It does not certify the account's overall profitability, complete tracking health or suitability for every client objective.
An issue means that the check identified its defined condition. Read what was counted or compared. An issue count often counts findings, not every affected keyword, ad or campaign beneath those findings.
A review recommendation needs human interpretation. A heuristic can draw attention to a setting worth considering without proving that the current choice caused poor performance. Commercial facts can justify retaining it.
Informational and unavailable cards are separate again. The current headless summary excludes informational cards from its issue, review and pass totals. Do not treat those totals as proof that every possible account concern was assessed.
Understand the unit behind every number. A finding about disapproved ads can count ads across groups, while another finding lists campaigns lacking a particular configuration. Adding those counts together would not produce a useful unique-entity total.
When explaining the review, state the specific observation and next decision. “These enabled groups contain no ads” is more precise than “the account is broken.” The narrower statement also makes the proposed response easier to review.
Keep unread checks out of clean-account claims
Some provider reads can fail while the rest of the audit remains useful. A returned empty fallback must not be mistaken for evidence that the account has no affected entities. ppctor marks dependent checks as not checked when their data was unavailable.
Those cards are neutral and excluded from the normal summary buckets. They do not become a pass because a query returned nothing usable. Read the full result for the unavailable sections before describing the scope of the review.
A temporary response problem and a rejected provider request can have different recovery paths. The card's explanation helps distinguish a retryable interruption from a fault that has been logged. Avoid repeating an unchanged unsupported request.
Keep successfully measured findings available for interpretation. One failed section does not erase the rest of the run. Equally, success elsewhere does not establish that the unread section is healthy.
A client note should name the unresolved area and its next step. “The asset section could not be checked” is a limitation, not an asset defect. Do not ask the client to fix an account setting merely because the audit could not read it.
Revisit the unread portion when the required condition is resolved. Do not make an incomplete audit look final by omitting its unavailable cards from the handoff.
Separate configuration from historical performance
Current settings and historical performance answer different questions. A campaign paused now can have spending in the reviewed period. A newly changed conversion role does not establish how every earlier outcome was counted.
The shared headless audit uses its defined last-thirty-days data context. If you ask a different performance question in conversation, preserve the requested dates and metric population. It is a new view, not automatically the audit's original evidence.
Record the review time when current configuration matters. A later colleague may read a stored result after someone changed the campaign. The snapshot remains useful as a record, but it should not be treated as a live read.
Use performance to inform a configuration decision where appropriate, without claiming causation from coexistence. A low ratio and a particular network setting can appear together. Their appearance does not prove that changing the setting will improve the ratio.
Preserve the client's commercial events. Promotions, stock changes and landing-page releases can alter performance over the same dates. They may explain why an apparent configuration concern needs a more specific investigation.
After a verified edit, choose a suitable later period for outcome review. The implementation receipt and the later measurement remain separate evidence, even when both are discussed in the same client report.
Review tracking flags with implementation limits
An auto-tagging check reads a setting. It can show whether the account's flag is enabled. That observation does not independently verify that redirects preserve click identifiers or that every destination records the intended outcome.
A tracking template also needs context. It may belong to a measurement arrangement outside the advertising account. Identify its owner and purpose before preparing a supported clear-template action.
Landing-page parameter observations can help the review, but they are not a universal prescription to append the same tags everywhere. A site may use a deliberate attribution convention or another integration.
For a suspected tracking problem, identify the actual symptom. A missing recorded event, changed value and unavailable report are different concerns. Choose the investigation that could establish the relevant fact.
Do not describe a successful flag change as a verified customer journey. The execution can check its own written setting. A form, checkout or downstream sales process requires appropriate separate evidence.
When the client asks for help explaining the issue, keep the language exact. “Auto-tagging was enabled and the setting verified” is supportable. “Tracking is completely fixed” overstates what that operation established.
Check conversion configuration deliberately
Conversion roles, counting choices and assigned values affect interpretation. An audit recommendation can identify configuration worth reviewing, but the client's business meaning is needed before choosing a change.
Ask which actions represent the intended primary outcomes and which are observational. A generic primary-versus-secondary recommendation should not erase a deliberate setup without understanding the objective and applicable provider restrictions.
Counting choices also need context. A repeated purchase and a repeated contact action can have different commercial meanings. Do not apply one preferred counting rule to every action merely because the field is editable for some types.
Supported conversion edits are constrained by action type and policy. Some conversion origins and types are locked. The existence of a metadata field does not imply that every action accepts a role or counting change.
For subsequent reporting, select the intended aggregate or named action. All primary conversions, primary purchases and primary leads can describe different populations. Values can belong to non-purchase actions as well.
Keep an implementation question separate from metadata. Reading an event's name or current role does not prove it fires at the correct moment. The relevant measurement owner may need to validate that behaviour outside the audit.
Review negatives and keyword diagnostics
The negative-keyword check examines presence across applicable enabled campaigns, including campaign, ad-group and attached-list coverage. Presence is not proof that the negatives are commercially appropriate or that every irrelevant query is excluded.
A campaign without negatives deserves context-specific review. The next task is to understand the offer and relevant search evidence, not to invent a generic list and describe it as verified waste removal.
Read the parent scope before proposing a negative edit. A shared list can affect several attached campaigns. Removing one entry there has a different effect from removing a campaign-specific exclusion.
The low Quality Score check identifies enabled Search keywords with scores from one to three. Google's Quality Score optimisation guidance describes the score as diagnostic rather than a business performance indicator.
Use that diagnosis alongside the selected performance evidence and the client's priorities. A low score does not by itself prove that a keyword loses money or should be paused. The client may have a specific reason to retain it while addressing relevance.
When pausing is justified, resolve the named keyword and its parents. A bulk low-score action has a broader effect than one selected pause. The proposal should disclose the actual set rather than hide it behind a cleanup label.
Distinguish empty groups and disapproved ads
An empty-ad-group check concerns enabled ad groups in enabled campaigns that contain no ads. A group with a paused non-removed ad is not empty under that definition. Preserve the check's actual meaning.
Such a group cannot serve an ad from content that does not exist, but that does not establish a measured spending loss. It may be part of unfinished setup or an intentional staging arrangement. Ask why it is enabled before deciding on a pause.
Performance Max uses asset groups and is not covered by that ad-group check. A PMax-only account should not receive a pass framed as though it uses the same structure. Applicability belongs in the explanation.
The disapproved-ad audit check has its own enabled-ad and enabled-parent population. Broader policy reporting can include other non-removed entities. Different counts can be correct when the queried populations differ.
Review a disapproved ad's actual issue and content before choosing a supported replacement. The audit does not supply a universal automatic policy fix or appeal. Policy review and content verification are different outcomes.
For standard RSA changes, review the complete replacement effect, including text, URL, paths and pins. A small wording request can use an operation that replaces more than one field. The human proposal must make that clear.
Interpret location and network recommendations
A location-option recommendation needs the client's service model. A local repair business and a travel business may have different reasons for reaching people inside or interested in a place. The audit cannot settle those commercial facts alone.
Read the current selected locations separately from the location option. Changing an option is not the same as replacing the country list or defining a service radius. The proposal should name the precise intended field and campaign.
Network recommendations also need evidence and scope. A campaign setting can be read, while performance contribution requires its own relevant data. Do not infer a quantified saving simply because a setting is enabled.
Keep channel applicability explicit. A recommendation for one supported campaign configuration does not automatically extend to every channel type. Discover the supported action and its target requirements before promising execution.
If a client confirms a geographic restriction, use that fact in the decision record. It can justify a supported change without implying that the setting guarantees perfectly accurate geographic delivery.
Evaluate later outcomes under the selected geography and business measure when appropriate. A verified settings receipt confirms the checked configuration, while later location performance remains a separate observation.
Review assets and product context
Asset coverage recommendations concern the declared associations and content. An asset's existence at account level is different from a campaign's own attachment. Read inheritance and target context before treating a missing association as a provider rejection.
A call asset may be relevant to a business relying on calls, but the correct number and its business use must be confirmed. A catalogue check cannot invent opening hours, phone routing or conversion settings.
For Performance Max, review asset groups and their listing selections in the correct hierarchy. A product can be included in one enabled group and excluded in another. A group-level occurrence count differs from a unique-product count.
Broader brand and category nodes are different from explicit item leaves. Do not describe an aggregate exclusion as a verified enumeration of every individual product beneath it without the required product evidence.
When a product decision uses historical ROAS, keep dates and the value definition explicit. Current exclusion and past performance can be reported together, but that does not establish the effect of the exclusion on the past result.
A freelancer may need the feed owner or creative owner for the next action. Keep that dependency visible. An audit can identify an asset-related decision without making it a setting you should change unilaterally.
Prioritise the next consulting decision
Prioritise findings by the client's objective, evidence strength, affected scope and practical next step. A large finding count does not automatically identify the most important work. One unresolved measurement definition can affect several performance interpretations.
Separate corrections from experiments. A confirmed obsolete template and a network setting worth testing are different decisions. The first needs ownership and the desired configuration; the second also needs a defined later observation.
Consider how much of the account a change affects. An account-level operation can influence several campaigns. A named keyword pause has a narrower target. The amount of visible text in the proposal is not a measure of its impact.
Resolve missing business facts before assigning urgency. A location choice might be wrong for the service area, but that requires the client's actual coverage. An apparently missing asset can be intentional while its content is under review.
Choose a coherent sequence rather than a simultaneous cleanup of unrelated settings. If conversion configuration, budget and creative all change together, later results become harder to interpret. Staging justified decisions can produce a clearer review.
Document deferred findings with an owner and reason. A freelancer should not have to reopen the whole audit because one item was left unexplained. “Waiting for the tracking owner” is a useful status when it accurately reflects the dependency.
Work through an invented onboarding audit
The following situation is invented. A freelance consultant begins work with Brook Workshop, a fictional local business. The client wants booked service appointments, serves a defined area and has an external provider maintaining its measurement setup.
The audit finds an enabled empty ad group, a location option worth reviewing and a tracking template whose purpose is unclear. A performance section cannot be checked. These observations should not be bundled into a claim that the account is generally broken.
The consultant confirms that the empty group belongs to an abandoned setup and is not awaiting creative. A supported pause can be proposed for that exact group. Its absence of ads is not converted into an invented estimate of lost spending.
The client confirms the service-area requirement. The consultant reads the current geography and location option before deciding whether a supported option change is justified. The selected locations remain a separate configuration question.
The tracking template remains unchanged until its external owner explains the dependency. The unavailable performance section is recorded as not checked, with a follow-up to resolve the read. Neither item is presented as a pass.
The audit therefore produces two settled decisions and two open investigations. This is a useful onboarding result because it preserves scope and uncertainty. It does not require every card to become green before the client can understand the review.
Make the fictional handoff usable
The empty-group note includes the named group and campaign, the client's confirmation that the setup is abandoned and the proposal status. The location note includes the actual field being considered and the supplied service-area requirement.
The tracking note identifies the owner who must answer the question. The performance note identifies the section that could not be read. Another consultant can continue those tasks without treating them as approved edits.
Resolve the exact change scope
Before preparing a change, resolve the intended entity and operation. Names help human review, while execution uses the account's actual identities internally. Similar names in different parents must not be treated as one target.
A keyword action needs its ad group and campaign. An asset-group selection needs its parent campaign. An account-wide setting can need no campaign discovery at all. The target requirements come from the supported operation, not from a universal sequence of reads.
Distinguish adding from replacing. An audience signal addition is different from replacing a whole selection; a language-set operation uses the desired complete set. Reviewers need the actual effect rather than a vague promise to optimise.
For bulk actions, freeze and disclose the intended affected set. A proposal should not silently grow because more matching entities appear after the reviewer reads it. Current-state checks can stop a stale proposal when relevant settings change.
Shared resources deserve extra attention. A negative-list edit can affect several attached campaigns, while a campaign-specific exclusion has a narrower relationship. Use the actual parent scope in the decision record.
If a target cannot be resolved confidently, ask for the missing information. A precise clarification is part of the workflow. Guessing a parent because it appeared earlier can turn a useful audit into the wrong live edit.
Run and discuss the audit in Slack
Use the audit workflow in the channel connected to the intended account. The Slack result summarises the shared audit evidence, while the full result preserves sections and details. Start with that run rather than request another full audit merely to see its existing findings.
The headless path can reuse available cached data and finishes the required performance phase before assembling its result. It is not a claim that every read is instant or that the returned snapshot remains current indefinitely.
Read the summary as a navigation aid. Issue, review and pass counts refer to their defined findings. Inspect affected entities and unavailable cards in the detailed result before explaining the account to the client.
Ask a focused follow-up tied to a finding. “Show the campaign and ad group for these low-score keywords” preserves the target context. A broader performance question should specify its dates and chosen outcome.
Use conversational explanation to make the finding clearer, but do not let a rewrite add unsupported business claims. The assistant can explain the returned evidence without inventing a cause or a quantified saving.
For the shared workflow and result presentation, see Google Ads audits in Slack. Treat the audit as a review input whose decisions still need client context and appropriate authority.
Ask focused questions from the findings
Each follow-up should close a defined gap. A setting recommendation may need a current configuration read. A keyword decision may need performance for a chosen outcome. A tracking question may need a fact from the measurement owner rather than another account query.
The assistant can inspect local field and operation catalogues to identify supported choices. That discovery does not itself read every campaign. Actual account evidence should be fetched only for the question and its required context.
Preserve the date range when narrowing a performance view. If you switch from thirty days to yesterday, say that it is a new diagnostic slice. Different results under different windows do not demonstrate that one answer was incorrect.
Make filters explicit. “Keywords with primary purchase ROAS at or above four” identifies the outcome and includes the boundary. Unsupported or unavailable inputs should remain visible rather than become guessed matches.
A complete count and a table preview are different results. If you need the exact affected total, use the supported aggregate or complete population. Do not count the visible examples as if they represented the whole selection.
When the client needs a clearer presentation of the same returned data, formatting can reuse it. A table or chart request should not trigger duplicate identical reads merely because the first explanation was too dense.
Ask the client for the fact the account cannot establish
An account cannot tell you whether a service is still offered simply from a keyword's presence. A location list does not establish where technicians can actually work. Ask those questions directly and keep the supplied answers with the decision.
Do not disguise missing commercial context as a data limitation that another provider query will solve. Some decisions depend on the client or an external owner, and the handoff should say so.
Prepare a supported proposal
An audit finding and an executable action are separate records. Discover the current supported action and valid targets before promising that the assistant can apply the recommendation. Some checks have no automatic fix.
The conversational proposal should name the target and precise change in everyday language. Put each logical step on its own row or section. Do not compress several effects into a technical string that makes the parent entity ambiguous.
Include important replacement effects. A complete RSA replacement can change text collections, remove pins and clear omitted display paths. A product-selection replacement can have a broader effect than adding one item. Those effects belong in the review.
Use the intended units for budgets and bidding targets. Currency amounts, percentage adjustments and target ROAS percentages are different inputs. A reviewer should not have to interpret a provider representation to understand the proposed value.
No live change has happened when the proposal is merely prepared. The confirmation button exists to apply the settled instruction after authorised review. A strong recommendation should not bypass that stage.
For action boundaries and the general sequence, see Google Ads fixes in Slack. Read its distinction between supported edits and findings that still need manual work.
Read the execution result precisely
A verified execution result confirms what the operation checked. It can establish the intended setting or created content. It does not establish later performance, successful policy review or a complete repair of an external measurement system.
An accepted provider request can have a narrower result than independently verified completion. Read the actual status rather than replace qualified wording with a blanket success claim in the client explanation.
An uncertain result requires reconciliation. Some writes may already exist even when complete verification did not finish. Inspect the named target before repeating creation or another action that could duplicate work.
The visible change-in-progress state acknowledges the execution phase. Repeated button presses should not become repeated live mutations. Wait for the outcome or inspect the state rather than create another proposal solely because delivery takes time.
Keep the receipt with the audit decision. A later reviewer should distinguish the finding, the approved proposal and the checked result. That chronology is useful when the account has several people making changes.
Rerunning an audit can confirm a check's current state, but it is not the same as evaluating the business impact of the edit. Use the selected later performance question for that separate review.
Explain the audit to a client
A client-facing explanation should start with the most relevant observed conditions and their significance. Use the client's objective to organise the message rather than copy every provider field into a wall of technical text.
Separate settled decisions from open questions. A supported change prepared for confirmation, a deliberate retained setting and an unread section are different outcomes. Each should have a clear status.
Use names and parent context instead of unnecessary resource identifiers. A keyword in its named group and campaign is easier to review than a technical path. Internal identities still support execution without needing to appear in ordinary client prose.
Do not quantify waste or lost revenue without evidence that supports that exact claim. An empty group, a missing negative and a low diagnostic score do not independently supply a financial-loss estimate.
When a recommendation is heuristic, say what needs review and why. That is more useful than presenting the setting as categorically wrong. The client's commercial requirements can make retention the justified decision.
End with the next concrete task: confirm a business fact, approve the prepared operation, resolve an unread section or review later results. The audit becomes useful through that decision, not through a stronger-sounding verdict.
Choose a sensible later review
Choose the later review based on what changed. A configuration correction can be checked after execution. Its performance effect needs an appropriate observation interval and the same selected outcome definition.
Do not rerun the full audit repeatedly to obtain an existing result in another format. The shared run already supplies its sections. A focused current-setting question can be more suitable when only one field needs checking.
Revisit the audit when the client changes its offer, geography, measurement or account structure. Those events can alter the interpretation of settings that previously made sense.
Use alerts for defined threshold transitions and budget pacing for its separate spending projection. Neither is a substitute for every audit check, and neither automatically authorises a conversational live edit.
Configured Rules are also separate. Current condition-based Rules ask for approval, while automatic mode is restricted to a time-based primary trigger. A periodic audit does not silently become permission for recurring changes.
Keep the freelance routine proportionate. A purposeful review at the right moment can be more useful than constant broad reads that generate more observations than you or the client can interpret.
Leave a usable audit handoff
The handoff should preserve the review purpose, observation time, affected names, business context and decisions. Another consultant needs enough information to continue without treating every finding as a new emergency.
Include unavailable sections explicitly. A summary count alone can hide what was not measured. The recipient should know which areas need a later successful read and which issues were actually established.
Record the complete status of proposed actions. Pending, expired, cancelled, verified and uncertain are different states. An earlier recommendation must not be mistaken for an applied setting.
Assign open dependencies to the relevant owner. A client confirmation, feed correction and measurement-system explanation require different people. A generic “follow up” note leaves the next consultant to repeat the triage.
Keep the future observation question with completed edits. If the decision was intended to address a particular lead outcome, the later reviewer should not substitute a broad conversion-value measure without saying so.
At the end of a freelance engagement, explain which review responsibilities continue elsewhere. The audit record should support continuity without implying an ongoing personal monitoring service that is no longer agreed.
Freelancer audit questions
Does a pass mean the account is fully healthy?
No. It means the specific returned evidence met that check's expectation. Unread sections, other checks and external business processes remain separate. Read the full result before making a broader claim.
Should every low Quality Score keyword be paused?
No. The score is diagnostic. Review the selected performance evidence, relevance and client objective before choosing an action. A supported pause still needs a clear target and authorised proposal.
Does an empty group prove money was wasted?
No. The check identifies enabled groups with no ads under its defined population. It does not quantify a spending loss. Establish whether the group is abandoned, unfinished or intentionally staged.
Can the assistant apply every audit recommendation?
No. Use the supported-operation catalogue and its target requirements. Some issues need manual investigation or provider review. A recommendation does not guarantee an executable fix.
What should I do with a not-checked card?
Retain it as an evidence limitation and read its reason. Resolve the temporary read problem or logged fault through the appropriate path. Do not classify the account area as clean or defective without its data.
Does verified execution establish better performance?
No. It establishes the checked implementation result. Evaluate later outcomes under a defined period and measure, with commercial context and other changes disclosed.