Google Ads audits for consultants: establish the evidence, prioritise findings and agree the next work

Begin with the client's audit brief

A useful audit engagement starts with the client's question. The client may be preparing a launch, checking a new agency's account structure, investigating measurement or reviewing a period of weak commercial results. These purposes can use some of the same checks but lead to different priorities. Agree the decision the audit should inform before treating its findings as a universal list of things to change.

Ask what the business wants the advertising to do and what has changed recently. Capacity, stock, service areas, pricing and website work can explain intentional settings that otherwise look unusual. The audit can read configuration and performance evidence, but it does not know every business constraint. A consultant adds value by connecting the findings to that context without inventing it.

Define the engagement boundary. An account review can identify useful tasks without including website implementation, creative production or continuous monitoring. The client should know which work the consultant will perform and which work belongs to another person. An audit result does not automatically grant authority to change the live account or create an ongoing response commitment.

Use a brief that is specific enough to guide interpretation: the connected account, relevant offers, expected activity, measurement questions and intended next decisions. Keep it concise and revisitable. If the client changes the question during the review, update the interpretation deliberately rather than forcing the new question into a checklist built for another purpose.

Create a baseline without premature changes

The initial run should establish evidence before the consultant starts adjusting settings. A baseline helps identify the account's current state and later distinguish reviewed configuration from changes made during the engagement. If edits happen immediately as each finding appears, it becomes harder to explain which conditions existed at the start and why a subsequent report differs.

Confirm the account connection and intended scope. A consultant may have access to several accounts with similar campaign names. The audit should identify the selected account meaningfully, while internal references remain behind the scenes. If access is unavailable, resolve that condition rather than presenting an empty run as a clean result.

Keep the audit's reporting period in view. The headless account audit uses its established recent range to assemble relevant performance context. A separate client question may need another period. Do not pretend the audit snapshot answers every historical question automatically. Use focused reporting for an interval the audit did not inspect.

Record important business facts alongside the baseline. A paused seasonal campaign, a recently replaced checkout or a temporary product shortage can affect the review. Those facts should not be used to dismiss evidence without checking it, but they help identify the actual next task. The baseline is a starting point for reasoning, not a verdict about the competence of whoever previously managed the account.

Confirm what the run could check

An audit is only as informative as its completed checks. Some source reads can succeed while another surface is unavailable. The application preserves failed-section information so a missing read does not become a passing result. Read those limitations before summarising the account, especially when the unavailable surface concerns the client's main question.

A complete fetch can be reused through the shared audit data path. An incomplete run is not cached as though it were complete. Its successful evidence can still be useful, but the consultant should not claim that the whole account passed because a failed section returned no rows. Preserve the difference between an empty valid result and evidence that was not retrieved.

Performance sections also need to finish or show an honest unavailable result. A frozen shared report should not leave a permanent “loading” label that suggests work is still running when it stopped. The current headless path carries completed section assembly and failure metadata into the snapshot. The client handoff should reflect that actual outcome.

If a material section is unavailable, choose the next check deliberately. A focused read may answer the immediate question without rerunning the entire audit. If the audit itself needs a fresh complete run, explain why. Do not repeat the same unsupported request unchanged or describe a local processing failure as proof that the provider has no relevant data.

Read failure, review and pass correctly

Audit labels describe the check's rule, not a complete commercial judgement. A failure means the observed configuration or condition does not meet that check's expectation. A review label calls for human interpretation. A pass indicates that the inspected condition met the check. None of these labels automatically establishes profitability, website quality or the suitability of every campaign.

Read the finding's explanation and target list. The label alone can conceal important scope. A presence check may verify that some exclusion structure exists without judging whether its terms are useful. An ad health check may inspect enabled entities rather than every paused item in the account. The consultant should interpret the actual rule instead of extending its colour to claims it never tested.

Informational findings can appear outside the scored summary. They can still be useful to the engagement, but their presence should not inflate the failure or review totals. Likewise, a diagnostic recommendation is not necessarily an urgent defect. The current summary deliberately separates informational material so the Slack counts and detailed rendering stay meaningful.

Use the labels to organise questions. Ask which failures affect an important business dependency, which reviews need context and which passes answer the client's concern. This turns the audit into a practical agenda. A consultant who merely repeats red, amber and green counts provides less value than one who explains the evidence and the next task each meaningful finding supports.

Separate finding counts from affected targets

The summary counts findings, not necessarily unique assets. One finding can contain several affected campaigns or groups. The same target can appear in more than one finding. A consultant should not tell the client that ten failed checks mean ten unique broken objects. Read the underlying lists before describing the workload or reach of an issue.

Keep repeated relationships meaningful. A product can appear in more than one enabled listing group. A destination can be used by several ads. A shared asset can affect several attachments. Depending on the question, the useful count may be findings, affected occurrences, unique targets or shared dependencies. State which population the report describes rather than silently deduplicating everything.

Displayed examples can be shorter than the full list. Use complete counts or focused queries when the client asks how many. Do not count a capped Slack preview and present that as the account total. A concise summary can still be accurate if it distinguishes the total from the examples shown and preserves the relevant parent context.

Use dependency grouping for workload planning. Several ads pointing to one failing page may share a website task. Several unrelated policy reasons may require separate creative or account tasks even if they appear under one heading. Counting colours cannot establish that distinction. Investigate the relationships before estimating effort or assigning responsibilities.

Apply campaign-type relevance

Account structure depends on campaign type. Search keywords, Standard Shopping listing groups and Performance Max asset groups are different structures. A check that concerns one should not produce a reassuring pass for an account that does not use it. Read which applicable entities were inspected and whether the section is relevant to the client's actual account mix.

Empty ad-group checks do not describe Performance Max asset groups. Quality Score diagnostics concern Search keywords. Network settings and creative requirements also differ by campaign type. The consultant should preserve these distinctions when explaining the result, rather than using a generic campaign vocabulary that makes every action sound transferable.

A supported edit needs the same type awareness. A new Search keyword belongs in an appropriate Search group, while a product selection belongs to the relevant listing structure. A generic dependency framework can order valid work, but it cannot make an incompatible target valid. Resolve the campaign type and parent before preparing a proposal.

When the client has several campaign types, organise the review around their purposes. An account-wide measurement question may span them, while a structural task may be specific to one. This keeps the report readable and prevents a recommendation for one channel from being applied mechanically to every campaign in the account.

Investigate measurement before performance advice

Measurement affects how performance findings should be interpreted. Inspect the relevant conversion actions, their roles and the reported activity before concluding that a campaign is producing no useful outcomes. An account's conversion setup can exist while its website journey still needs validation. Conversely, a period with no conversions does not by itself prove that tracking was never configured.

The no-tracking check uses specific account evidence: a not-tracked status or no actions, together with no recorded conversions. It is not a live test of every form, purchase or call path. Read the finding at that scope. If the client needs implementation assurance, assign the relevant website and event checks separately.

Primary and secondary roles change aggregate meaning. All-primary counts and value can include several action types. Purchase-only and named-action questions are narrower. Do not relabel all conversion value as sales revenue or assume that a lead action records a completed sale. The audit can identify setup questions, while focused reporting establishes the selected measurement evidence.

Some requirements cannot be fixed with an account toggle. Customer data terms, website tags and event implementation involve separate consent or technical work. The supported edit framework should preserve those boundaries. A consultant's task is to explain the actual dependency and owner, not to promise an automatic tracking repair from a setting visible in the audit.

Treat Quality Score as a diagnostic

Google's Quality Score guidance describes it as a keyword-level diagnostic, rather than a KPI or an auction input. Use it to investigate relevance and the customer experience. A low score is not by itself proof of wasted cost or a command to pause the keyword.

The audit lists enabled Search keywords with low scores as informational review material. Read the named keyword, campaign and group, then ask what it contributes to the business. A commercially important keyword can deserve better creative or landing-page work rather than immediate removal. A keyword with little relevant activity may require a different investigation.

Keep unavailable scores distinct from low scores. Missing diagnostic evidence does not justify inventing a number. The consultant can ask for relevant components or performance where supported, but should not rank an unscored keyword as worst merely because the report has a blank cell. The question is what the observed diagnostic supports.

A supported pause is still a separate owner decision. If the client chooses it after reviewing the business context, the proposal should name the exact keyword and parents. Do not apply a bulk pause simply to improve the audit's appearance. A cleaner diagnostic list is not a commercial success metric, and verification of the pause does not prove improved results.

Review exclusion presence and relevance separately

A negative-keyword presence check asks whether applicable enabled campaigns have an exclusion route, including relevant campaign, group or attached-list structures. It does not establish that the selected terms cover every irrelevant query or preserve every valuable one. The consultant should use the finding as a configuration question rather than a full judgement of search intent.

Review actual search evidence and the business offer before selecting new exclusions. A phrase that looks irrelevant without context may describe a service the client provides. A term with no conversions in a short period may simply lack enough evidence. Ask the owner about the commercial meaning and choose the appropriate supported scope.

Creating a shared list and attaching it are separate tasks. An existing list name alone does not establish that the relevant campaign uses it. Likewise, a campaign negative differs from a group negative. Read the relationships before concluding that a proposed addition will have the intended effect. The confirmation should identify the chosen texts, match treatment and parent or attachment scope.

Do not treat a passed presence check as completion of ongoing query review. The offer and search landscape can change. A consultant can recommend a focused review routine while acknowledging that the audit checked only the stated structural condition. That interpretation is more useful than telling the client their exclusions are complete because one card is green.

Understand empty groups and paused creative

An empty group has no non-removed ads under the applicable check. A group containing paused creative is different: the creative exists but is switched off. A group containing disapproved creative is also different. These conditions can all affect readiness, but they need different explanations and tasks. Do not call every non-serving group empty.

The audit's active-parent population can differ from a bulk action's discovery population. Review the exact candidates in a proposed action rather than assuming its selection is identical to the displayed audit finding. The owner needs to see which groups would change. A frozen target list makes that review concrete and prevents execution from silently expanding to newly found targets.

An intentionally paused parent can explain why a structure is not currently relevant. A returning seasonal offer needs its own readiness check before enablement. Creating new creative, changing an existing ad and enabling the group are separate actions. The consultant should not infer that adding an ad automatically resumes the full parent chain.

Use structural findings to ask the business question. Is the group still needed? Does it have suitable creative? Is the offer available? A supported pause may be appropriate for an unused group, while a creative task may be appropriate for a needed offer. The audit identifies the condition; the client context determines the useful response.

Separate reported policy from destination testing

The audit's disapproved-ad check concerns enabled ads in enabled parents. A dedicated policy read can inspect a broader non-removed population, including paused items when relevant. State which one was used. A clean active finding is not proof that every dormant ad is ready for a later launch.

Obtain the specific reported reason before choosing a repair. Creative wording, destination behaviour and account requirements have different owners. A campaign status change does not remove a policy reason. The consultant should identify the affected ad or asset and its actual context, then separate the account task from website or provider work.

Reported destination policy is not a live browser crawl of every page. A stored URL and an HTTPS prefix do not certify that forms or checkout work. Assign direct journey tests where the client's question requires them. The absence of a reported destination issue does not prove continuous uptime or a complete customer experience.

After a supported creative or URL edit, verify the stored result and later inspect the provider review separately. An executed edit is not immediate policy approval. If the ad remains paused, approval does not itself enable it. Keep these milestones visible so the client knows which part of the repair is complete and which dependency remains.

Review targeting against the business model

Targeting recommendations require business context. A local urgent service may prefer a different location-intent arrangement from a business selling travel bookings to people planning a visit. A review can identify the current setting and a supported alternative, but the consultant should not apply one model's preference to every account automatically.

Inspect location criteria and intent separately. Changing to Presence does not necessarily replace the selected geographic criteria. Adding a city does not remove a broader country target. If the client wants a narrower service area, identify all required steps and display them clearly. A partial edit can leave the original broad coverage intact.

Languages, networks and audience settings also need their actual scope. A desired complete language set differs from adding one language. A campaign network setting differs from a group-level targeting control. Explain the operation that matches the client's intent rather than grouping all targeting work under a vague “optimise reach” instruction.

Keep later performance interpretation separate from configuration correctness. A setting can accurately express the service boundary without guaranteeing more conversions. A broad audience signal can be appropriate without proving high-quality demand. The consultant should verify the requested state, then report commercial evidence over a suitable period with the definition and context intact.

Connect findings to commercial dependencies

A finding matters through the business dependency it affects. A disapproved ad for the client's main available offer can deserve attention before a minor naming issue. A measurement gap can limit every performance recommendation. A website failure shared by several ads can have a broader practical effect than a long list of unrelated diagnostics. Establish those relationships before deciding the order of work.

Ask which offers are commercially important now. Historical spend can help identify where the account invested, but it does not automatically identify the client's current priority. A newly launched service may have little history and still matter greatly. A high-spend campaign may be winding down intentionally. Combine the account evidence with the owner's actual plan.

Keep value and profit distinct. Primary conversion value can include assigned lead or call values, and purchase revenue does not include every business cost. An audit cannot calculate reliable profitability from a configuration label alone. The consultant can ask for margin or operating context when the decision needs it, while keeping the provider-backed figures accurately named.

Use dependencies to avoid duplicated work. Several creatives sharing one page may need one website repair plus separate policy checks. Several campaigns using a shared arrangement may need a carefully scoped setting decision. Do not estimate effort from the number of red rows alone. The useful workload is the set of actual tasks and their prerequisites.

Prioritise work without counting colours

Prioritise by the importance of the affected offer, certainty of the evidence and the task's dependencies. A well-established blocking condition on an important offer can come first. A heuristic that needs business clarification can remain a review task. An unavailable section may need a focused check before it can be ranked at all. This approach is more useful than ordering every item by its displayed colour.

Distinguish urgency from ease. A simple setting change can be easy to approve but less important than a measurement investigation. A website task may take longer while blocking the client's main journey. The consultant should explain why a task comes first, including which later decisions depend on it. Completing easy items is not necessarily progress on the client's central problem.

Separate immediate, planned and investigative work in the task discussion without hiding known defects. A confirmed issue should have a concrete response. A question that needs evidence should have a named next check. A future optimisation idea can be considered later, but it should not be used to defer a known issue that the client asked the consultant to fix within the agreed scope.

Keep the first action manageable. A long audit does not require one giant edit proposal. Prepare related supported work in a size the owner can review, and keep the remaining tasks visible. Do not execute only part of an oversized request and imply the whole audit was addressed. A clear sequence of complete tasks is easier to assess and verify.

Follow an invented audit engagement

This example is invented. A consultant reviews Larch Point Learning before a new course launch. The audit reports an active disapproved ad, a keyword diagnostic and a campaign without a relevant exclusion structure. The client also explains that one older course is intentionally paused. These observations form the review agenda, but they do not all have the same urgency or response.

The consultant identifies the disapproved ad's actual group and campaign and obtains the reported reason. The new course is the client's immediate priority, so that finding comes first. A website colleague checks the destination while the owner reviews truthful replacement copy. The keyword diagnostic remains a separate investigation rather than being paused automatically to improve the audit summary.

The exclusion finding prompts a scoped review of actual queries and the course offer. The consultant does not upload a generic negative list without the client's commercial judgement. The paused older course is noted as intentional, with a readiness check planned only if the business later wants to resume it. The audit is interpreted in the context of the launch brief.

For the chosen supported creative edit, the owner reviews the complete desired content and confirms the exact target. Execution verification checks the stored result, while the later provider review remains a separate task. The final handoff states what was observed, what changed, what was checked and what remains. It does not claim that the audit guaranteed launch sales or repaired every possible account issue.

Investigate findings in Slack

Slack can provide the initial audit summary and a place to ask focused follow-ups. Ask which findings need review, which targets are affected and what evidence supports a particular recommendation. The current headless summary carries review details as well as counts, so a consultant can discuss them without pretending that every meaningful issue is a failure.

Resolve targets from the current conversation, preserving the account context. A follow-up about “that group” should refer to the identified group when it is unambiguous. If several groups were listed, clarify the intended one. The response should show its campaign parent rather than exposing internal IDs or inventing a relationship to make the answer sound complete.

Use requested charts or tables for relevant performance evidence. A finding about configuration may need a read of the selected setting, while a question about commercial effect may need a period and metric. Do not run a broad report merely because the client asks for a clearer explanation. Conversational interpretation and creative suggestions can use established context without an unnecessary provider request.

Keep the conversation's next step specific. The AI should explain the known condition and the evidence still needed, then prepare a supported proposal only when the owner selects an action. A vague “fix it” can require clarification about the repair rather than a guessed mutation. That is especially important when the finding could lead to website, creative or account work.

Reuse evidence and refresh deliberately

The web and headless audit paths share the same underlying run logic rather than maintaining separate sets of checks. This helps the Slack summary and detailed audit describe the same evidence. A completed dataset can be reused, and the detailed sections can be retained with the run instead of fetching the whole account again merely to show more detail.

Reuse is valuable when the scope and snapshot still answer the question. A request for the details behind the existing finding can use the already obtained section evidence. A new date range or current target state can require another read. The consultant should understand which question concerns the snapshot and which question concerns the live account now.

Do not confuse cached evidence with a guaranteed fresh state. Another person can change the account after the audit. Before executing a supported proposal, the relevant settings must be checked again. That focused safety check is different from rerunning every audit section. Efficient work removes duplication while retaining the evidence necessary for the intended action.

Incomplete runs need honest handling. Their successful fields can support narrow observations, but their failed sections should not be cached as complete or transformed into passes. If a later complete run becomes available, compare its actual evidence rather than assuming that every changed count represents an edit. Improved coverage itself can explain why the findings differ.

Assign the correct task and owner

Translate findings into tasks based on the thing that needs to change. A creative task belongs to the offer and content decision. A website task belongs to the destination or measurement implementation. An account setting belongs to a supported Ads operation. A provider requirement may need a separate official process. Assigning every finding to “the ads person” can leave important dependencies unowned.

Write each task with a meaningful target and completion check. A website colleague needs the relevant page and behaviour to test. The connected owner needs the selected account proposal. A consultant needs the metric definition and period for follow-up. Avoid raw identifiers and vague instructions that force the recipient to repeat the investigation before acting.

Keep decision authority distinct from implementation assistance. The consultant can draft recommendations, a colleague can suggest wording and a developer can repair the site. The connected owner still approves the live account proposal. A completed website task does not approve a pending ad edit, and a conversational acknowledgement does not replace Confirm and Execute.

Track dependencies explicitly. A later policy check may depend on the creative edit. A commercial comparison may depend on settled data after the change. A radius action may depend on a populated linked location group. The task plan should reveal these relationships so work can proceed in a valid order without unnecessary provider calls to rediscover rules already available locally.

Prepare supported actions precisely

A finding's suggested treatment is not a universal execution tool. Use the actual supported operation, its valid target and prerequisites. Some findings require manual provider work or website changes. Some convenience actions are limited to particular budget, bidding or asset arrangements. A broad API schema does not establish that every field can be changed in every account.

Show each logical action clearly in the review. If the owner asks to add keywords and enable the group, both steps must appear. If a chain creates a parent and child, the dependency order should be planned locally and the created targets should remain readable. Do not hide a missing step inside a generic success message for the part that was completed.

Review selection scope carefully for bulk actions. The audit display and action discovery can have different populations. Freeze and show the actual selected candidates before approval. The owner should not have to assume that “all” means precisely the examples visible in the summary. A broad action needs enough target evidence to make its consequences reviewable.

Preserve the confirmation boundary. The proposal says that no change has been made yet and expires after fifteen minutes. The connected owner confirms, relevant settings are rechecked and execution follows the supported path. Cancellation of a pending proposal is not an undo for an executed change. A later reversal needs a fresh supported request based on the observed current state.

Distinguish an edit from resolved risk

Execution verification checks the supported requested account result. It can establish that a status, value or selected content was stored. It does not automatically establish that a website works, a provider approved the ad or the campaign now achieves the client's commercial goal. Keep those later checks separate in the task record.

A narrower acknowledgement should be described as such. Some operations can be accepted without independent verification of their full final result. The consultant needs to know whether the next task can rely on a verified state or requires inspection. Do not use an emphatic success label to hide incomplete evidence.

If the result is uncertain after dispatch, inspect before retrying. The provider may have applied the write even when the application did not receive a complete result. Repeating creation can duplicate assets; repeating a relative adjustment can alter the value again. Reconcile the intended target and then decide whether another proposal is needed.

Recheck the original risk using the appropriate evidence. A disapproval needs its later policy state. A measurement question needs the relevant implementation and reporting checks. A configuration preference needs the selected setting. A commercial hypothesis needs a suitable later period. The audit finding and edit receipt are useful records, but neither replaces every one of these checks.

Write a useful client handoff

Lead with the important account questions and completed work. State the scope and any material coverage limitation. Then explain the highest-priority findings, their business relevance and the selected next tasks. The client should understand why the work matters rather than receiving an unranked copy of every check.

Separate observed facts, client context and recommendations. A reported disapproval is an observation. The client's launch priority is business context. Replacing selected creative is a chosen response. Keep these categories clear in connected prose so the handoff is easy to assess without implying that the audit alone proved the whole recommendation.

Use readable target context and concise task descriptions. A keyword should retain its group and campaign; a shared website repair should explain which offers depend on it. Remove technical clutter rather than removing the information needed to recognise the work. A polished layout cannot compensate for an incorrect parent relationship or a hidden unresolved dependency.

Include the actual outcome of live work. Pending, expired, verified, acknowledged and uncertain proposals need different follow-up. A later reader should not infer execution from an earlier “looks good” message. Keep unresolved tasks visible with their owners and checks, so another consultant or client colleague can continue without restarting the entire review.

Recheck at a meaningful point

A follow-up should answer whether the intended task is complete and whether the relevant evidence changed. Some configuration results can be checked immediately after execution. Provider review and commercial reporting may need a later check. Choose the timing from the dependency rather than running the whole audit repeatedly in the hope that its colours improve.

Use focused checks where appropriate. If one campaign setting changed, inspect that setting. If a website repair was completed, test the relevant journey and later reported state. A full audit can be useful after several material changes or at another agreed review point, but it is not necessary for every narrow follow-up question.

When comparing audit runs, account for changes in coverage and applicability. A newly enabled campaign can introduce relevant checks. A previously unavailable read can reveal evidence that was not observed before. A changed finding count is not automatically a changed number of defects. Read the actual findings and affected targets before describing the movement to the client.

Connect the completed review to consultant alerts and consultant reporting when ongoing attention is part of the engagement. Alerts can identify defined transitions, while reports can evaluate selected periods. These workflows complement the audit without turning it into a promise of continuous automatic repair.

Keep the audit's boundaries visible

An audit is a structured review of available evidence under defined checks. It is not a full legal assessment, a complete website crawl or a profit certification. The consultant should explain the actual scope where it affects the client's decision. Avoid generic caveats that bury the result, but preserve specific limitations such as an unavailable surface or an unchecked implementation journey.

A passed check has a bounded meaning. Negative-keyword presence does not prove complete query control. No active disapprovals does not certify paused creative. A healthy tracking setting does not test every site event. A valid URL string does not prove destination availability. These distinctions should be part of interpretation rather than discovered only when a later problem contradicts an overbroad claim.

Keep unsupported operations out of promises. If a relevant fix requires Google Ads or another responsible person, describe the exact task and follow-up. The application can still support investigation and a useful handoff. A local capability catalogue should guide valid work, not create confidence that every provider setting has a safe accepted mutation path.

The consultant's value lies in making the evidence and decision clearer. That includes recognising when a heuristic is appropriate, when business context changes its meaning and when another check is necessary. The audit becomes a strong starting point when its rules, coverage and task consequences remain visible, rather than a score to improve at any cost.

Answer common consulting questions

Does a failed check always need a live edit?

No. Read the rule and business context first. The response may be a website task, measurement investigation, provider process or supported account change. Prepare a live proposal only for the selected valid operation.

Are the summary totals counts of broken assets?

No. They count findings in their categories. A finding can affect several targets, and a target can appear in several findings. Read the affected lists before describing the number of unique assets or estimating workload.

Should low Quality Score keywords all be paused?

No. Quality Score is diagnostic. Review the keyword's relevance, performance and business role before choosing a response. A cleaner diagnostic list is not proof of improved commercial results.

Does an empty section mean everything passed?

Only a completed relevant check can establish its result. An unavailable read is not a pass. Preserve failed-section information and obtain the material missing evidence before closing the client's question.

Can the audit repair every finding automatically?

No. Supported account edits use exact targets and owner confirmation, while other tasks belong to website, measurement or provider workflows. Verify the actual edit separately from later policy, journey and commercial checks.