Give the client a useful investigation, not a score-chasing exercise
Google Ads Quality Score audits for consultants: investigate keyword relevance and approve proportionate changes
Understand the diagnostic inventory, preserve its target and evidence limits, and connect each proposed remedy to the client's real offer, message and destination.
Set the consultancy question before reviewing scores
A Quality Score review should answer a practical consultancy question. The client may want to understand an audit finding, improve a message, examine a destination or decide whether a keyword still belongs in the account. Each question needs different evidence. A low diagnostic number can draw attention to a target without establishing the right commercial remedy.
Agree what the engagement will produce. A useful deliverable might be a prioritised investigation, a creative brief or an approved pause for a named criterion. It should not be a promise to raise every score to a particular number. The client needs an improved decision about its advertising, rather than a detached numerical target that says little about the business outcome.
Identify the owners who can supply relevant facts. The account manager can explain the campaign's purpose, the client can confirm the offer and a developer can explain the destination's behavior. A consultant can combine those observations into a focused recommendation. None of those roles should be assumed to know every part of the customer journey merely because they have advertising access.
Start narrowly and expand deliberately. A question about one keyword does not require a new account-wide performance report, complete ad inventory and website crawl. Establish the named target and the unresolved question, then request the evidence that can answer it. This makes the engagement faster and keeps its conclusions tied to observed facts.
Explain what the diagnostic means
Google's Quality Score guidance describes a keyword-level diagnostic on a one-to-ten scale. Its components concern expected clickthrough rate, ad relevance and landing-page experience. Google explicitly states that the reported score is not an auction input or a KPI to aggregate with ordinary performance data.
A consultant should therefore avoid describing a change in the visible score as a guaranteed change in cost, sales or profitability. The number can identify where a relevance investigation might be useful. The actual business result still needs its selected cost and outcome evidence, while an account edit needs its own verified configuration result.
The distinction helps the client interpret an audit recommendation calmly. A keyword with a low diagnostic can still represent an offer the business wants to serve. A higher diagnostic does not certify that the target produces valuable customers. The commercial decision should combine the wanted intention, customer experience and measured outcome rather than use the score alone.
Make the purpose explicit in the report title and explanation. “Review relevance for this keyword” gives the client a task. “Repair low Quality Score” can imply that the app has a writable score setting or a guaranteed numerical fix. The diagnostic is evidence to investigate; the remedy changes an actual supported object or a separately maintained website.
Read the current low-score inventory
The current account reader selects enabled positive keywords in enabled campaigns and ad groups with a returned Quality Score of three or less. It requests the keyword text, parent campaign and group names, score and resource reference. Results are ordered by the diagnostic and limited to three hundred rows.
The audit displays the returned low-score targets with their parent context and provides per-keyword pause actions where a resource reference is available. The finding is an informational review rather than a universal score-based commercial judgement. It gives the consultant an inventory to examine, not proof that every displayed target should be removed from useful demand.
The selected population excludes paused keywords and paused parents. A client asking about a named paused target needs its own relevant read rather than an answer inferred from this enabled inventory. Similarly, a campaign type without keyword Quality Score should not receive a generic claim that its keywords all pass this diagnostic.
A successful response establishes what the reader returned at that time. It does not contain component explanations, historical daily diagnostics or selected-period performance simply because the rows have a score. Retrieve those facts separately when they actually change the requested investigation, while preserving the confirmed target context.
Keep the returned sample and full population distinct
The three-hundred-row limit is important when the account has many low-score keywords. A returned list can be useful without being a complete count of every qualifying target. The consultant should distinguish the displayed inventory from an exhaustive population and avoid calling the visible row count the total merely because the interface shows a confident summary.
The ordering also affects interpretation. The lowest diagnostics appear first, which is different from ordering by spend, business value or commercial urgency. A prominent row is not necessarily the most important intervention. The next prioritisation question may require narrow performance evidence for the selected candidates.
A campaign view reuses the account-wide low-score data and filters the current implementation by campaign name. It does not issue an independent unrestricted inventory for that campaign. The consultant should keep the inherited cap and name-based grouping in mind, particularly when a requested target needs exact current resolution.
A shortlist can support examples and investigations while another complete scoped report supports a total. Keep those outputs clearly labelled. The shared framework should preserve row coverage through formatting, filtering and follow-up so that an attractive table does not accidentally turn a bounded dataset into an exhaustive account claim.
Separate unscored, unavailable and empty results
An absent diagnostic is not a score of zero. Google explains that an unscored keyword can lack enough relevant evidence to establish the reported score. A consultant should preserve that state rather than include it automatically in the low-score group or claim it meets a higher-score standard.
A provider read failure is another state. The current low-score reader allows the rejected query to propagate so the audit can report an unavailable section. Returning an empty array after a failure would create a misleading clean result. Resolve the read before making a diagnostic conclusion about the selected population.
A successful empty low-score result means no qualifying rows were returned by that selected query. It does not mean every keyword in every campaign has a known score of at least four. Paused targets, non-applicable campaign types and unscored entries need their own interpretation.
Use plain language in the handover. The client should understand whether there were no matching enabled rows, the check could not run, or a named keyword has no reported diagnostic. Those distinctions lead to different next steps and prevent missing evidence from being presented as a business success.
Request component evidence when it matters
The low-score inventory alone does not identify which component needs attention. Its selected fields contain the overall diagnostic, not all the component statuses. If the client asks whether the message or destination is responsible, obtain the relevant supported evidence before presenting a specific explanation.
Component observations should guide a focused task rather than a universal reconstruction of the account. A relevance observation can support review of the actual ad and wanted intention. A destination observation can support review of the page and completion path. Neither supplies every operational detail required to prescribe the remedy.
Keep observed provider diagnostics separate from the consultant's hypotheses. The page might omit a key offer detail, but that needs actual page evidence. The ad might set the wrong expectation, but the reviewer needs its current content. A useful report states the observation, the question it raises and the evidence needed to test the explanation.
Do not invent component values from the overall number. A score of three does not authorise a fictional table with every component marked below average. The client can still receive a useful next instruction when the available data is narrow. Precision about what has been read is more useful than an apparently complete diagnosis built from assumptions.
Interpret local summaries without inventing a provider KPI
The app also has an ad-group summary that reviews locally supplied averages for scored groups and flags those below five. That is an application review convention. It should not be described as a native Google Ads ad-group Quality Score field or a business performance KPI.
An average can hide varied keyword circumstances. A group with a few weak diagnostics and several stronger ones still needs examination at the actual keyword and offer level. The summary can help locate an investigation area, but it does not establish which particular keyword the consultant should pause or which destination change the client should fund.
The summary excludes groups without a positive supplied score and can be absent when no scored groups are available. That is not proof that those groups have perfect relevance. Keep the difference between non-applicability, missing diagnostic evidence and a threshold result visible when explaining the card.
A consultant should avoid treating an average increase after pausing low-score targets as proof that the customer experience improved. The composition of the remaining inventory has changed. Evaluate actual message, destination and business outcomes at their own scopes rather than reward an arithmetic improvement created by removing the diagnostic rows.
Keep the real keyword hierarchy visible
Keyword text can repeat in multiple groups and campaigns. Preserve its real ad group and campaign when interpreting a finding or preparing a change. A phrase in an emergency-service group can set different expectations from the same phrase in a scheduled-service group. The target's context is part of its meaning.
The client-facing description should show the keyword, ad group and campaign by readable names. Do not append internal IDs or create a fictional parent ad group. Technical references support ownership and exact API preparation behind the scenes; visible hierarchy helps the client recognise the intended object.
If the conversation identifies a target ambiguously, ask the focused question that resolves it. A prior unrelated mention does not make every repeated keyword refer to the same group. Established context can avoid unnecessary questions, but the live operation still requires canonical resolution in the connected account.
Parents also identify responsibility. The campaign owner may choose the commercial intention, another colleague maintains the ad copy and a supplier maintains the page. A low-score keyword does not authorise changing all those objects. Resolve the actual remedy and show only the relationships needed to understand and approve its effect.
Confirm that the client wants the intention
Before trying to improve relevance, establish whether the business wants the demand represented by the target. A keyword might describe a discontinued service, unsupported product or customer situation outside the client's capacity. A better message is not necessarily the right task when the intention itself is unwanted.
For a wanted intention, examine the offer definition. The client may have a specific service model that the generic keyword does not communicate. A useful review can identify what a prospective customer needs to understand before enquiring. The consultant should confirm those business details rather than invent them from the wording.
Separate a permanent offer boundary from a temporary condition. A full diary or unavailable product this week can need a different remedy from a service the business never provides. Record duration and review conditions where they matter. An ordinary confirmed edit does not create an automatic future reversal.
The decision can be keep and improve, investigate, or pause the named target. A numerical diagnostic does not choose that outcome by itself. A clear commercial reason makes the eventual proposal understandable and gives the client a basis for reviewing it later when the offer or circumstances change.
Review observed queries at their own scope
A positive keyword is different from an observed search term. The consultant should not treat its text as a complete transcript of all the searches it reached. Use relevant returned query evidence when the investigation needs to understand actual intentions, preserving dates, campaign context and the report's coverage limits.
A query can be unsuitable locally while the positive keyword remains useful. In that case a precisely scoped negative may be a better decision than pausing the whole criterion. The negative-keyword guide explains presence checks, matching boundaries and shared-list relationships.
Do not invent missing queries to explain the score. A provider category, creative suggestion and observed search term have different meanings. The client needs to know which evidence was actually returned and which themes the consultant is proposing to investigate.
Performance evidence for a query also needs an intended outcome. A short period without recorded conversions does not prove that the client should never serve that intention. Combine the commercial fit, selected outcome and data maturity before recommending a restriction, rather than treating every disappointing row as an automatic negative.
Assess the promise made by the ad
Read the ad as a prospective customer would. Does it identify the offered service, set a credible expectation and make the next step clear? A consultant should look for specific mismatches rather than assume that every low-score target needs entirely new copy.
Confirm operational facts behind the promise. Same-day attendance, a minimum price, free delivery or a particular qualification should not appear merely because it would sound persuasive. The client needs copy that reflects its actual offer. An assistant can improve wording without inventing the business facts the proposed message depends on.
Review the message with its destination. An ad offering a quotation should lead to a useful quotation route, while an appointment offer should make booking understandable. A collection of individually appealing headlines can still communicate inconsistent expectations when considered together.
Give the creative work a purposeful brief: the wanted intention, real differentiator and destination action. Keyword repetition alone is not a guaranteed remedy for the diagnostic. The consultant should define what a clearer message needs to communicate and how the client will judge whether the proposed copy achieves that purpose.
Use creative conversation without unnecessary provider calls
The assistant can brainstorm headlines, descriptions and qualification wording from supplied or already retrieved context. Ordinary creative conversation does not require Keyword Planner, another audit or a fresh performance scan. The client should be able to explore alternatives before deciding whether a live account change is needed.
Label creative output as proposed copy rather than measured evidence. A suggested headline does not have a provider-reported relevance status or proven conversion effect. The assistant can explain how it expresses the confirmed offer, while the client remains responsible for choosing the business promise.
Ask for essential missing facts when they change the creative task. If the client has not confirmed delivery terms, write alternatives that avoid an unsupported promise or request the relevant detail. Do not burden the conversation with questions that can be answered from established context, but do not manufacture claims to make the copy sound complete.
Publishing is a separate stage. A supported ad-creation request needs its format, parent group, landing page, content and initial status. Brainstorming alternatives does not authorise creation. The proposal can turn selected creative into a reviewable operation while keeping its later diagnostic and business performance separate from configuration readback.
Review the destination against the offer
The destination should help the intended visitor understand the offer and act on it. Review service detail, relevant product information, qualification boundaries and the next step. A low diagnostic does not prove a particular page defect; the consultant needs actual destination evidence before assigning a correction.
Use the page the relevant ad really points to. A homepage review cannot certify a different landing page, and a static screenshot cannot demonstrate the entire completion route. Identify the tested destination and scope so the developer knows which observation the task concerns.
Separate content work from account work. A Google Ads mutation can change a supported destination field, but it does not automatically rewrite the website or repair a form. The proposal and handover should identify the actual owner and workflow for each correction.
Keep the review practical. The client needs the page detail that affects the chosen intention, rather than an unbounded redesign recommendation attached to every low-score row. A concise task might clarify the service boundary or make the booking route easier to understand. Its acceptance condition should describe the demonstrated result.
Test the actual customer completion path
A useful first page is only part of the journey. A visitor may continue to a booking widget, payment page or contact form with separate behavior. Test the path that supplies the selected outcome before claiming that a destination change has repaired the whole experience.
Distinguish successful page loading from successful enquiry or purchase completion. A visible button does not establish what happens after submission. The consultant should record which steps were demonstrated and which remain untested, particularly when a third-party transition introduces another owner or implementation.
Use authorised test evidence without exposing real customer identity values in Slack. The client-facing task can name the page, completion behavior and observed issue without showing private lead records. Account identifiers and personal emails also do not belong in report URLs as convenient lookup shortcuts.
Keep measurement verification separate from usability. A person may complete the form successfully while the selected event is missing, or an event may fire before the intended business completion. The relevant implementation owner needs to confirm the actual event behavior. A keyword status readback does not certify either part of that website process.
Consider structure when it serves a real purpose
An ad group can contain several intentions that need different promises or destinations. If the evidence establishes that mismatch, a structural review may be useful. The consultant should explain the business purpose of the change rather than recommend splitting every low-score keyword into its own group.
Creating a group, adding keywords and creating an ad are distinct logical actions. A live proposal needs supported types, valid parents and the intended initial statuses. The local dependency definitions should establish the order, while current target reads establish actual account ownership and prerequisites.
Do not imply that copying the target into a new group will itself produce a better diagnostic or business result. The useful change is the coherent offer, message and destination the structure supports. Those elements still need confirmed facts and appropriate implementation.
Preserve historical context in the handover. A new object is not the same reporting resource as an old one, and a removed or paused target remains part of earlier performance evidence. Describe what changed and why so that later comparisons do not silently merge different populations under one familiar keyword text.
Keep commercial performance beside the diagnostic
The client ultimately needs to understand cost and valuable outcomes. Define whether the report uses all primary conversions, purchases or a named action. A low diagnostic can coexist with a commercially useful result, while a higher diagnostic can coexist with an unsuitable offer. Neither measure should silently replace the other.
Standard conversion value can include assigned value from primary leads or calls as well as purchases. Several primary actions contribute within the chosen scope, while secondary actions remain outside a primary-only measure. If the client wants purchase revenue, select and label that narrower definition rather than call every standard value contribution a sale.
A comparison with Analytics needs the corresponding event when lead meaning differs by site. The assistant can inspect available names and suggest a likely candidate, then ask for confirmation where naming does not establish the business meaning. An individual named action and specified event can be compared without pretending all conversions represent the same stage.
All-traffic revenue divided by advertising cost is blended ROAS or MER. That broader ratio can inform the client, but it does not establish the return of one keyword. Keep the population and revenue definition visible so a favorable account total does not become evidence of the selected target's individual commercial performance.
Preserve thresholds and logical conditions
A request can combine diagnostic and performance conditions. “Score at most three and cost above 100” differs from a rule joined by OR. Greater than differs from greater than or equal to. The shared reporting definition should carry the requested comparisons and grouping through selection, filtering and presentation.
Apply performance conditions to valid evidence. Undefined ratios, missing outcomes and unavailable reads should not become zero values that accidentally qualify targets for a pause. Keep full precision for threshold evaluation, then round client-facing metrics to at most two decimal places.
Preserve the relevant reporting grain. A keyword condition is different from an ad-group condition, and the same text in several groups represents several resources. A combined total should not erase the hierarchy required to identify which object qualifies for the requested change.
Explain preview limits when they matter. The three-hundred-row diagnostic inventory and a small displayed shortlist do not automatically establish every target meeting a combined rule. A complete count needs suitable complete evidence. The consultant can still prioritise observed candidates while stating the bounded population honestly.
Separate current diagnostics from historical periods
The current diagnostic field does not become a historical daily series merely because the client asks for last month's performance. A historical diagnostic request needs its relevant supported fields and dates. Do not repeat today's score across earlier dates or infer a trend from one current observation.
Use the account time zone and requested period boundaries for performance. Last thirty days is distinct from last month. Keep the resolved dates consistent through the table, chart and recommendation so that a follow-up refers to the same evidence population.
A recent outcome may be incomplete because the business records a later sale after an initial enquiry. The consultant should consider that process where it changes the decision. A diagnostic observation and a mature commercial result have different evidence timelines.
Record actual changes when evaluating a later result. A new ad, changed destination or paused criterion has an execution date, but other offer and measurement changes may also occur. The client can compare periods without claiming that every difference was caused by the single action under review.
Follow an illustrative consultant investigation
This example is invented. Consultant Owen Price reviews a consultation keyword for Hazel Studio Kitchens. The low-score inventory identifies the keyword in the Consultation group within the Kitchen Planning campaign. The business confirms that this is a wanted intention rather than an unavailable service.
Owen reviews the actual ad and destination instead of pausing immediately. The client confirms that consultations require a scheduled appointment, while the existing message suggests an immediate walk-in discussion. The relevant task is to align the promise with the real service, not to make an unsupported numerical guarantee.
The assistant can help draft alternatives from that confirmed context without another provider call. If the owner selects a supported ad-creation request, its proposal names the group and campaign, content, destination and initial status. Website clarification remains a separate task for the site's owner.
After approved account execution, the readback establishes the resulting ad configuration. The developer's destination task and subsequent diagnostic observation remain separate checkpoints. Owen's handover records the intended service and acceptance criteria, rather than claim that one successful write has already improved the score, reduced cost and produced more consultations.
Choose a manageable first intervention
A long low-score list can distract the client from a few meaningful decisions. Choose a small first investigation based on business relevance, available evidence and the feasibility of a useful remedy. The lowest number alone is not a complete priority rule.
Consider the owner effort needed to resolve the question. A clear unsupported-offer boundary may be straightforward, while an ambiguous destination issue needs developer evidence. Group related investigative work when it genuinely shares the same offer or page, without assuming that all similar keyword texts need one identical action.
Preserve the difference between investigation priority and execution scope. A consultant can discuss several candidates in one report, but a live proposal needs an explicit supported request. A ranked list does not authorise changing every object it contains.
Give the first intervention a clear next checkpoint. The client should know whether it needs to confirm the offer, choose creative, demonstrate the page or review a precise pause. This keeps the engagement moving through answerable questions rather than leaving a broad low-score report that nobody knows how to act on.
Retrieve evidence through shared definitions
The reporting framework should select valid metrics, dimensions and resources from local definitions. A score question, a cost question and a selected-event comparison have different data needs. The assistant should request the supported evidence that answers the user's intent instead of applying a dedicated template to every wording variation.
Reuse established facts when they remain appropriate. A follow-up asking for a clearer explanation or table can use the same complete evidence. A changed period, new component question or live edit prerequisite can justify another narrow read. Repeating the entire account audit for a format change adds latency without supplying new facts.
Local dependency knowledge can establish structural relationships without provider queries merely to learn that a keyword belongs to a group. Actual ownership and current resource state still require appropriate evidence when preparing execution. This distinction reduces unnecessary API usage while retaining the safeguards needed for a live change.
Keep routing details behind the scenes. The client needs the answer, scope and next decision, not model instructions or raw query syntax. A generic capability framework is useful when it lets the assistant adapt to the requested evidence without inventing fields or silently changing the measure.
Distinguish one-keyword and bulk pause scope
The individual audit pause action validates that the criterion resource belongs to the audited customer and submits its paused status. The conversational lifecycle path also prepares the target and records a relevant status check. Pausing one positive keyword does not add negatives or pause its parent campaign.
The existing audit bulk action has a broader selection than the displayed enabled-parent inventory. Its fresh query selects enabled positive low-score criteria without repeating the campaign and ad-group enabled filters or the three-hundred-row display cap. The consultant must not describe it as necessarily changing only the rows currently visible in the card.
The bulk action uses partial failure and reruns its selection to identify remaining enabled qualifying resources. That supports an account-state result, but it is not the same as proving every originally displayed resource changed under concurrent edits. Preserve actual scope and remaining uncertainty rather than presenting an unrestricted bulk operation as a narrow shortlist action.
If the client requests only named targets, prepare that exact requested scope through the supported path. A low-score finding does not authorise the wider bulk action. The distinction should be visible before confirmation so the owner can make a deliberate choice about what stops serving.
Present exact steps for confirmation
The proposal should show each target and effect on clear separate lines. A keyword pause names the keyword, ad group and campaign. An ad creation names its parent group, content, destination and initial status. The client should not infer whether a campaign is being created or paused from ambiguous arrows and technical status fragments.
One confirmation can cover a supported sequence with clear action sections. If the full request cannot be prepared safely, explain why narrower proposals are needed before execution. Do not execute a subset silently or treat a hidden status change as incidental to a creation step. The client should review each intended live effect.
State “No changes have been made yet. This proposal expires in 15 minutes.” Confirm and Execute authorises the displayed scope. Creative brainstorming, an audit recommendation and a prior approval for another object do not authorise this new mutation.
After confirmation, immediately make the controls inactive and show prominent change-in-progress feedback without mentioning the user. Relevant live state is checked again before execution. The client then knows the click registered while the tool validates the current request rather than waiting through an apparently unchanged message.
Verify the actual setting and reconcile uncertainty
Validation concerns the prepared request and its prerequisites. Verification concerns the resulting resource state. A valid pause request has not necessarily executed, and a submitted operation is not fully verified until the relevant evidence establishes its result.
A successful pause readback establishes the criterion's status, not improved relevance. An ad readback establishes the created object and requested configuration, not a future diagnostic number. A website task needs its own implementation evidence. Keep these milestones separate in the final response and consultancy record.
Ambiguous or partial writes need reconciliation before repetition. An ad may already exist even when verification was incomplete. Replaying a creation blindly can produce duplicates. Inspect the affected state through the supported workflow and report established effects alongside the remaining uncertainty.
The final successful execution should be prominent, with clear achieved steps and readable names. If verification remains incomplete, describe the actual known result rather than declare every step failed. The client needs a trustworthy account record before asking for another change.
Show readable evidence and honest chart scope
A current low-score list is usually better explained as a scoped inventory than an invented time trend. A historical chart needs actual dated observations. A performance chart needs its selected metrics and outcome definitions. Do not spread an aggregate value evenly across dates or repeat a current diagnostic to fill missing points.
For requested compatible performance measures, try one native chart with up to three metrics or segments. More than three needs a readability-limit explanation and a selection. The client should not receive separate charts merely because the framework failed to recognise a valid shared evidence request.
The native chart workflow permits twenty points per series. Use a clearly stated weekly or monthly breakdown for longer periods unless daily detail is explicit. Explicit longer daily requests need split native charts with the point-limit explanation, rather than an unrequested image or missing dates.
Display at most two decimal places and human-readable dates, using actual week ranges where appropriate. Calculate grouped business ratios from summed base values. Put the user's mention before feedback so feedback remains last. These rules keep the evidence clear without presenting the diagnostic itself as an aggregate business KPI.
Leave a clear ownership and review record
The handover should identify the diagnostic observation, exact target hierarchy, confirmed business intention and evidence period. Record the selected remedy and its reason, including cases where the decision is to retain or investigate. Another consultant should be able to understand the work without guessing from a list of scores.
Assign separate account, creative and website tasks where their owners differ. Each task needs an acceptance condition describing what demonstrates completion. A verified account write should not automatically close a developer task or a later commercial evaluation.
Record the next review question and relevant change date. The client can inspect later diagnostic or business evidence without promising that the original action causes every subsequent difference. Any ongoing monitoring needs its actual separately configured workflow rather than an implied automation attached to this one-time conversation.
The consultant audit guide provides the broader review context. A useful Quality Score engagement turns a bounded diagnostic observation into an evidence-based decision, a proportionate approved action where needed and a clear record of what remains to be demonstrated.