Use the diagnostic to define a precise commercial and implementation review
Google Ads Quality Score audits for enterprise: interpret keyword diagnostics and choose accountable supported actions
Keep low-score observations separate from profitability, unavailable scores and complete account coverage. Connect the keyword, creative and destination review without treating a pause as a quality repair.
Define the enterprise decision
An enterprise Quality Score audit should start with the decision the organisation wants to make. The owner may ask which keywords have low scores, why a particular keyword appears in the finding, how it performs commercially or whether a supported status change is appropriate. Those questions need different evidence. The report should preserve the actual intention rather than translate every low-score mention into an immediate pause proposal.
Establish the customer and relevant campaign. Central teams can manage several regions and similarly named groups, while the same keyword text can appear in many resources. Use the connected authenticated context and actual resource identity internally. Display the readable keyword and its correct parents so the owner can recognise the selected scope without interpreting a raw criterion identifier.
Agree a completion condition that matches the task. A diagnostic inventory can complete when its scope and returned values are established. A creative discussion can produce suggestions without changing the account. A pause needs approval and resulting-state evidence. A landing-page investigation needs its own actual observations. These tasks should not be compressed into one claim that keyword quality has been fixed.
A useful review may conclude that a keyword remains enabled. The business may confirm a valuable purpose, performance may support retention or evidence may be insufficient for a live decision. Record the reasoning and next review period. The audit's usefulness depends on accountable interpretation, not on producing the largest possible pause list or increasing a preferred diagnostic number.
Understand the diagnostic's proper role
Google's Quality Score guidance describes a keyword-level diagnostic on a one-to-ten scale. It explicitly distinguishes the score from a business KPI and an auction input. The enterprise review should use it to guide questions about experience and relevance, rather than treat it as a direct profitability verdict.
A low returned score is a useful observation. It does not independently establish wasted spend, low lead quality or the reason a customer did not purchase. Those are business and performance questions with their own measures. The review can combine them thoughtfully without turning the diagnostic into a substitute for the actual commercial outcome selected by the owner.
A target such as improve quality needs a concrete task before a live proposal. The owner might want to review creative, inspect a destination or pause a named keyword. Each has different effects. There is no supported mutation that directly assigns a desired Quality Score. An action should change an actual editable resource field, with later diagnostic observations treated separately.
Keep organisational dashboards focused on meaningful outcomes. A diagnostic distribution can be useful as a review inventory, but a single invented account-wide score should not become the enterprise's headline success measure. The team needs to know which keyword observations led to which decisions and whether the resulting commercial evidence supports the chosen approach.
Establish the eligible Search population
The current low-score finding is tied to Search. It should not be advertised as a quality report for every PMax asset group, Shopping product or Display audience. If the account does not run the relevant Search population, the finding is not a universal clean-account assurance. The tool should explain applicable scope rather than force unavailable diagnostics into a generic quality metric.
The current reader selects enabled positive keywords under enabled campaigns and ad groups. That scope differs from every historical or paused criterion in the account. A returned list should retain the parent-status conditions used by the actual query. Do not infer a wider population merely from a report title or assume a keyword's own enabled flag describes all of its parents.
The report reader orders the low-score observations and caps its result at three hundred. The visible count therefore needs that coverage context when the cap is relevant. A full selection request cannot safely treat the audit list as complete just because it looks substantial. The conversational bulk-pause path has its own uncapped current candidate read.
If the owner asks about one campaign or group, retain that scope. A focused question should not trigger an account-wide claim or a universal pause candidate set. Shared reporting definitions can select valid keyword fields and relationships locally, while provider reads establish the actual selected observations. This makes the answer both more exact and less wasteful.
Interpret the current low-score finding
The audit highlights enabled keyword observations with scores in the low range of one through three. Its status is informational and does not contribute a scored issue-ledger penalty. The report shows readable keyword, campaign and group context, and it offers supported pause actions. Those controls are available decisions, not proof that pausing is commercially appropriate for every listed resource.
A successful empty low-score result means the selected reader found no qualifying low-score rows. It does not prove that every keyword has an available score or that the whole account is profitable. The enterprise should preserve the distinction between absence of selected low observations and a comprehensive assessment of every keyword's experience and economic outcome.
A populated result identifies a review population. The owner can ask for relevant performance, component evidence or creative ideas. Those are separate tasks. The current basic low-score row contains the score and parent context; it should not be described as if the audit already retrieved every component status or tested every landing page.
The remedy's exact effect matters. Pausing changes keyword status. It does not improve the keyword's diagnostic, rewrite the ad or repair a website. A final result can verify the status field honestly while a later creative or destination review remains open. This clarity prevents a convenient audit button from being mistaken for a universal quality-improvement operation.
Keep unavailable and missing scores distinct
A missing score is not the same as a returned low score. A failed provider read is another state again. Keep these distinctions in the output rather than convert all absent values to zero. A zero-like placeholder can make an unsupported resource appear severely poor or create a false pause candidate when no actual diagnostic was established.
The current reader propagates failures so the audit can report that the low-score section could not be checked. That behavior is important. A rejected query should not become an empty successful list and a reassuring message that no low-score keywords exist. The owner needs the actual evidence state before making a commercial or live configuration decision.
Missing diagnostics should not be assigned an invented average for charts or filters. A distribution can show established values and describe the remaining coverage. If the owner needs a complete inventory of available versus missing scores, use the actual supported query and scope for that question. The low-score sample alone does not supply every denominator.
Preserve unknown information through follow-ups. If a previous answer said a component was not retrieved, the next reply should not confidently diagnose that component from the overall score. Obtain the relevant evidence or keep the gap explicit. Conversational fluency should make the investigation easier, not conceal the difference between observed facts and plausible hypotheses.
Use actual component evidence when requested
The provider's Quality Score reference identifies expected click-through rate, ad relevance and landing-page experience as components, with comparative statuses. The current basic audit query does not retrieve all those component fields. If the owner asks which component is weak, the tool needs the appropriate supported evidence rather than infer an answer from the total score.
A component observation can help define a focused task. The team may review the offer wording, the intended query or the destination's usefulness. The component is still a diagnostic signal, not a complete causal explanation of every business result. Keep the actual returned status separate from the proposed action and its expected commercial purpose.
Google's Quality Score improvement guidance recommends examining components alongside other measures and connecting ads with useful experiences. For an enterprise, the practical task is to identify the actual owner and supported change. Do not turn general guidance into a claim that the app tested the customer's page or proved a particular headline causes the low score.
If a component read is unavailable, explain that specific gap. The AI can still discuss possible review approaches using confirmed context, but should label them as suggestions. A named low-score keyword can remain the established target while the next evidence task changes. Retaining that continuity is better than restarting broad discovery for each question.
Resolve the keyword and its real parents
The same keyword text can exist in several groups, campaigns and match types. A live request must resolve the actual criterion. Internal identity is necessary for precise execution, but client output should show readable text, match context where useful and actual parents. Raw IDs and undefined parent labels do not help the owner understand the selected scope.
A keyword belongs to an ad group, and the group belongs to a campaign. The proposal should display that actual chain. It should not describe an ad group as having another ad group as its parent or confuse a PMax asset group with a Search group. Resource relationships belong in the shared framework so outputs remain correct across different actions.
Use established conversation context when it resolves the target. If the owner has reviewed one keyword row and says pause it, retain that resource when unambiguous. If several matching rows remain plausible, ask which group or campaign they intend. Do not guess a criterion from text alone or lose context because the follow-up omits the full name.
A removed or replaced criterion needs fresh identity. Another resource with the same text is not automatically the original target. A later proposal should establish the intended current resource before execution. This prevents a historical report or similarly named group from silently directing a write to a different keyword than the owner reviewed.
Connect the keyword to confirmed commercial intent
The enterprise should confirm what the selected campaign offers. A keyword can be relevant to one division and unsuitable for another. Industry-level assumptions are not enough. Establish the actual product, service, geography and intended customer task before describing a low-score observation as irrelevant traffic or preparing a broader restructuring proposal.
Separate business facts from ideas. The owner can confirm that the company sells a particular item or does not provide a service. The AI can suggest possible interpretations of a query, but those hypotheses should not become live exclusions or status changes without a clear decision. A useful review keeps the observed diagnostic, confirmed offer and proposed explanation distinguishable.
Low score and commercial value can coexist in the selected report. That does not require ignoring the diagnostic; it means the decision should use both layers appropriately. The team can preserve a valuable keyword while investigating experience or messaging. A pause is one available status effect, not the inevitable conclusion of every low-score row.
Likewise, a higher score does not guarantee a useful business outcome. The organisation's actual selected conversions and values remain essential. The article should avoid a simplistic ranking where higher diagnostics always mean better investment. Accountable interpretation asks which task the evidence supports and which commercial question still needs its own measurement.
Brainstorm creative from the actual offer
An owner can ask for headline or description ideas based on the confirmed product and keyword intention. The AI can answer creatively without calling Google Ads merely to generate text. It should not pretend those suggestions were measured winners or that a provider supplied their performance. This conversational support is distinct from an actual creative inventory or live ad request.
Use confirmed claims. Do not invent free delivery, accreditation, a discount or regional availability because a more specific headline sounds persuasive. The enterprise's offer owner should establish what is true. Suggested wording can remain useful while its commercial claims are reviewed, and the output should make clear that no account changes have been made.
A request to upload or replace creative changes the task. It needs the actual supported ad operation, required fields, precise parent and confirmation. The AI should retain the selected group and available creative context rather than jump to another interface or ask for a campaign that was already unambiguously established.
Creative suggestions should not promise a future score. A revised headline can be a reasonable hypothesis for the intended query, but configuration verification establishes text and resource state, not a guaranteed diagnostic improvement. Keep subsequent performance and quality observations as separate review tasks after the approved change.
Review the destination as its own evidence layer
A configured final URL identifies a stored destination. It does not prove the observed browser route, loading behavior or usefulness for the selected intention. The Quality Score audit does not test every destination automatically. If the owner wants a landing-page assessment, define the relevant route and obtain appropriate actual evidence rather than infer a page defect from the overall score.
The website owner can review whether the page fulfils the confirmed offer and intended task. A campaign destination inventory can help choose the page, but it is not itself a completed journey test. Regional redirects, booking applications or other transitions may matter. Keep the configured URL, observed route and implementation conclusion as distinct records.
A supplied address can be normalised for supported ad creation, but that convenience does not certify reachability or business content. If required information is missing, ask for it politely and preserve the established group context. Do not invent a final URL from a product name or claim that a plausible domain was tested when no such evidence exists.
A website task can continue while the account review completes. The owner may retain the keyword, prepare a creative proposal or choose a status change based on the available evidence. Those decisions do not replace the destination investigation. A clear handover assigns the page question to its actual owner and states which observation remains needed.
Examine group themes without automatic restructuring
A group can contain several intentions that are difficult to address with one confirmed offer. The team can discuss whether clearer themes would help, but a group count alone does not establish that the current structure is poor. The audit's low-score row is a diagnostic starting point, not a full account taxonomy assessment.
Use actual inventory when the question requires it. A list of keywords and ads can show what the selected group contains. It cannot independently establish the organisation's preferred structure or prove every query follows a particular intention. The business owner should confirm the intended boundaries before a new group or keyword set is prepared.
Creating new resources is different from moving existing history. A supported new group and keyword creation request produces new identities. Pausing an old keyword changes that resource's status. The proposal should show these effects honestly rather than call the sequence a seamless move that preserves every historical relationship.
Do not create new structure simply to chase the diagnostic. A useful plan should connect the proposed group, creative and destination to an actual offer and selected measure. The enterprise can choose an experiment or preserve current structure while reviewing messaging. The action should follow the confirmed decision, not an automatic assumption that every low score requires more groups.
Select stable business measures
Define whether the review concerns all primary conversions, purchases, leads, calls or a named action. A request for the worst keyword needs its ranking measure. Highest cost, lowest return and highest cost per selected lead can identify different resources. The AI should preserve the intended metric rather than invent a generic worst-performing label from the diagnostic.
All primary actions in the selected population contribute to the intended aggregate, while secondary actions are excluded. The report should not silently remove one primary action because it resembles another. An implementation review can assess overlap separately. Reporting current selected outcomes honestly is different from applying a preferred conversion configuration inside the arithmetic.
Primary conversion value can include nonpurchase values, including those associated with leads or calls. Keep the label accurate. A purchase-specific selection is narrower and should retain its category boundary. Named actions need actual identity, and the selected definition should survive follow-up chart and threshold requests.
For lead comparisons with GA4, establish the corresponding event. The AI can suggest likely names from the actual event list but should confirm the meaning when uncertain. Do not invent a quality hierarchy from lead categories. Calls and offline outcomes remain available where supported without an automatic cross-source pair or a claim that the tool inspected the business's external systems.
Keep volume and ratio arithmetic clear
Show base cost and selected outcome count alongside ratios. A striking result can come from little activity, and a small commercial amount can have different significance from substantial spend. The enterprise should understand the scale rather than receive an invented certainty threshold or a recommendation based only on a colourful diagnostic row.
Combined ratios need summed compatible base measures. Combined return uses selected value over cost; combined cost per outcome uses cost over selected count. An unweighted average of row ratios is a different calculation. Keep full precision through the calculation and display no more than two decimals so formatting does not distort the actual result.
Missing observations are not zeros. A zero selected outcome is a returned fact under its definition, while a missing field or failed read has another status. Undefined ratios should not appear as zero when the denominator is absent or zero. These distinctions prevent a misleading graph or ranking from turning unavailable evidence into a confident pause candidate.
Do not turn historical candidate cost into guaranteed savings. A verified pause establishes status, but future traffic and account allocation remain separate business observations. The review can explain the historical exposure and intended decision without promising that an identical amount will be saved or that return will necessarily improve.
Separate current diagnostics from historical observations
The current quality field describes the available diagnostic when read. It should not be displayed as a historical daily series merely because the owner selected a performance period. Historical diagnostic measures need their actual supported fields and grain. A current value repeated across every day would manufacture observations the tool never retrieved.
Use available historical evidence when the owner asks how diagnostics changed. Do not infer that today's score applied throughout the last month. A current status inventory is also different from historical serving configuration. A keyword may have been paused or re-enabled during the period, and that boundary can matter to interpretation.
Resolve last thirty days as the intended rolling window, not last calendar month. Preserve the dates through follow-ups and identify a partial current day when included. A graph should not silently use a different end date from the report. Accurate time scope is essential when the enterprise compares observations around an ad or website release.
Record known concurrent changes. A new offer, creative replacement or destination release can influence the interpretation of later observations. A movement in score or conversions after a change does not independently prove causality. The account operation can be verified separately, while the commercial review uses stable selected measures and an appropriate completed period.
Preserve exact operators and grain
Keep strict and inclusive comparisons as requested. Score below a threshold differs from at or below it. Cost above a value differs from at or above. Preserve AND and OR when combining diagnostics and performance. The shared reporting logic should carry these operators rather than simplify them into a different set of keywords.
State the grain. A keyword whose whole-period total qualifies differs from a daily keyword row that qualifies. A campaign aggregate differs from a group aggregate. Current diagnostic fields combined with historical metrics should retain those evidence layers, rather than imply the score held on each day of the selected period.
Coverage belongs in exact counts. The capped audit list can support its displayed finding, but a request for every qualifying resource needs complete appropriate evidence. Do not filter that capped list locally and announce an account-wide exact result. Use the supported filtered or uncapped read required by the question and state any remaining limitation.
A chart or export should retain the selected conditions. Changing format is not permission to drop the named action, parent scope, period or threshold. Reusable data definitions should support this continuity across subjects so the AI can choose valid fields and operators from the framework instead of requiring another patch for each natural-language variation.
Treat pausing as a status effect
The supported keyword pause action updates the criterion's status. It does not remove the keyword, change its match type or alter its text. The proposal should say Pause keyword and show the actual group and campaign. A vague fix-quality action would conceal the concrete effect the owner is authorising.
A pause can be a deliberate business decision even when the diagnostic is only part of the evidence. The owner may confirm the intention is outside the offer or choose to stop activity while another review proceeds. Preserve that reason. Do not present the status change as proof that the diagnostic itself was repaired.
Current parent state matters to scope and interpretation. The keyword's enabled field is not a guarantee that its campaign or group is serving. The selected conversational bulk path explicitly reads enabled parents. A single-resource proposal should still identify the actual resource and applicable current state rather than infer live delivery from one flag alone.
A later enable request is a separate supported lifecycle effect with its own current checks and approval. It should not be assumed from a conversation about improved creative. The tool should preserve the previous target where unambiguous while establishing current state, rather than create a new keyword because the old one is paused or has the same text as another resource.
Review the actual bulk-pause candidates
The conversational bulk preparation reads enabled positive keywords under enabled campaigns and groups with the low-score condition. It has no audit-sample limit and filters the qualifying score range before building named status effects. This is different from using the capped report list as all candidates. The owner should see the actual proposed population.
The legacy audit action and conversational preparation have different query details. A public guide should not imply that every interface uses identical candidate selection just because the action names are similar. The displayed proposal and current selected tool determine the scope being approved. Review that concrete candidate set before execution.
A bulk effect should preserve unrelated fields. It pauses the selected keywords rather than rewriting creative, clearing URLs or changing conversion roles. If the owner also asks for those changes, show them separately under the shared logical-action safety limit. Do not hide several consequences inside the word optimise or execute only an arbitrary subset.
Partial provider outcomes need precise reporting. Establish resulting states for the requested resources and distinguish completed effects from unresolved ones. An accepted batch is not a universal quality or profitability certificate. If some results remain unclear, reconcile their current state before retrying rather than blindly replay the full bulk request.
Understand supported RSA creation and replacement
The dedicated RSA creation path targets a standard Search ad group and creates a paused ad from the supplied required creative and final URL. It does not enable the group or campaign implicitly. The owner should see the actual created status and any separately requested status effect, rather than assume a successful creative upload means live delivery began.
Existing RSA editing replaces the desired creative fields through its actual operation. The current path retains ad status, unpins the supplied text and clears omitted display paths. Those effects matter to review. A request to suggest one headline should not silently become a complete replacement that discards other desired creative or pinning behavior.
Supply the full desired supported creative when preparing a replacement. The proposal should show headlines, descriptions and final destination in readable rows. Preserve the established group and ad identity. A normalised URL is not a tested destination, and a successful creative readback does not establish policy approval, future score or commercial uplift.
A creative idea remains separate from a live request. The AI can brainstorm from confirmed offer context without account calls. When the owner explicitly asks to create or edit, resolve the supported resource and required values, then prepare the clear proposal. This distinction keeps normal conversation useful while making live changes accountable.
Sequence actual resource effects
A new ad group must exist before its dependent keywords or ads refer to it. An existing group already supplies its identity, while a supported planned-resource chain can use the framework's actual dependency references. Do not assume a new resource's name alone proves the parent exists or that the provider can resolve an uncreated relationship.
The local dependency table can establish valid structural relationships without API calls. Current evidence determines whether a particular campaign or group exists and is eligible. Keeping these roles distinct avoids unnecessary discovery while preserving the precise live request. The table should be comprehensive for declared capabilities, not a special Search-only workaround for one example.
Show each logical effect clearly. Creating a group, adding keywords and creating an ad are distinct consequences even if they form a supported sequence. The shared safety limit can require a split or rephrase for a larger request. Do not omit an enable effect while claiming a multi-step setup completed.
If a dependent effect fails or remains unverified, report the established earlier resources separately. A created group can exist even when an ad was not confirmed. Reconcile actual state before another attempt so a retry does not create duplicate parents. The owner should continue from verified evidence rather than repeat the entire chain because the final step's message was ambiguous.
Coordinate account, creative and website owners
A Quality Score review can involve several enterprise teams. The account owner manages status and resource configuration. The creative owner confirms claims and desired text. The website team investigates the relevant destination. The business owner defines useful outcomes. Assign the concrete questions rather than ask every team to fix quality without specifying what was observed.
A diagnostic signal is useful common context, but each team needs its own task boundary. A creative suggestion does not certify the landing page, and a page improvement does not automatically authorise pausing keywords. Preserve what each owner confirmed and what remains open so the handover does not become a broad assurance assembled from unrelated partial checks.
Organisational planning can happen without new provider calls. The AI can draft a review checklist from confirmed context. Actual account configuration or performance still needs evidence, and live edits need the supported approval flow. This division keeps the conversation responsive while retaining the exact checks necessary for a concrete account request.
Do not claim the app enforces an enterprise approval hierarchy it has not implemented. The organisation may complete its own internal review before the connected owner confirms the displayed proposal. That process can complement the app's authorisation boundary without being represented as a built-in feature. The final message should describe the actual approved and verified effect.
Follow an illustrative enterprise review
The following example is invented. Rowan Workshop sells equipment and runs separate installation campaigns. Its account owner, Nora Flint, asks which enabled Search keywords appear in the low-score finding. The app returns the scoped observations with correct campaign and group labels. It explains the audit's bounded coverage rather than claim every account keyword was fully assessed.
Nora selects one installation keyword and asks for primary appointment performance. The tool preserves the named resource, measure and completed period. The result shows actual base measures rather than treating the diagnostic as profitability. Nora decides to keep the keyword enabled and asks for more specific headline ideas based on the confirmed service offer.
The AI offers creative suggestions conversationally without claiming measured winners. Nora later requests an RSA replacement with the full desired text and confirmed final URL. The proposal makes replacement, retained ad status, pin removal and display-path effects visible. It waits for Confirm and Execute instead of uploading during the earlier brainstorming exchange.
After confirmation, visible progress appears while checks and provider work run. Resulting creative fields establish the configuration outcome. The final message does not claim Quality Score improved or policy review completed. The website owner still reviews the selected route, and Nora schedules a later business comparison using stable appointment definitions rather than a fabricated diagnostic uplift.
Reuse context and narrow evidence reads
A question about one low-score keyword should not automatically read every asset and product in the account. Use the established resource identity and retrieve fields needed for the next question. The local capability definitions already describe valid relationships and operations. Provider calls should establish actual current facts rather than rediscover those structural definitions.
A presentation-only follow-up can reuse returned observations. Drawing a table or chart from the same period and selected measure does not require repeating broad discovery. A new period, component request or live edit can justify focused additional evidence. Preserve that distinction when many enterprise conversations share provider and model capacity.
Short follow-ups should retain the keyword, group and campaign. If the owner says use the same group after reviewing one unambiguous target, the AI should resolve it from context. Ask a focused clarification when several candidates remain. Do not restart an unrelated planning flow or claim the target is unknowable merely because the follow-up lacks a full name.
Creative support can proceed from confirmed context without provider calls. Actual measured claims still require evidence and writes require approval. This general intention routing should apply across headlines, descriptions, keyword ideas and explanations. A narrow exception for one brainstorming phrase would leave other creative conversations unnecessarily slow and tool-driven.
Make proposals clear before confirmation
The proposal should show the actual readable target and exact effect in a visible step section. A keyword pause needs its group and campaign. An RSA replacement needs the desired creative and relevant retained status. Avoid raw IDs, undefined parents and technical payload walls that make the owner hunt for the action being approved.
One confirmation covers the supported bounded set of logical actions. If the request exceeds the shared safety limit, ask the owner to split or rephrase rather than applying an arbitrary subset. A status change, creative replacement and new resource creation remain separate effects. Calling them improve-quality should not obscure the requested consequences.
State “No changes have been made yet. This proposal expires in 15 minutes.” Confirm and Execute authorises the displayed scope. An explanation, creative suggestion or prior unrelated approval does not execute this proposal. If the target or desired effect changes, prepare an updated reviewable message before live execution.
Use the enterprise approval guide for the decision boundary and the enterprise audit guide for wider planning. The owner should understand the actual configuration effect without assuming the audit's pause control is a mandatory or universal quality repair.
Verify configuration without promising score improvement
Show prominent change-in-progress feedback immediately after confirmation and make the controls inactive. The owner should know the click registered while current checks and provider work proceed. Do not mention the user in the interim state. An unchanged proposal followed by a later result can make correct execution feel uncertain and encourage repeated clicks.
Validation concerns the prepared request and applicable operation. Verification concerns the resulting resource fields and relationships. A successful submission alone is not always complete independent readback. Establish each requested status or creative effect before describing the entire request as executed and verified.
If a write may have succeeded but readback is incomplete, reconcile current state before retrying. Show established effects separately from unresolved verification. Do not say nothing changed when a resource may already be in the requested state, and do not blindly replay a creation or replacement because the final confirmation was unavailable.
Make the verified result prominent when evidence passes. Describe the actual status or creative outcome. Keep Quality Score, policy approval, website behavior and commercial performance outside that claim unless their own evidence was obtained. The account operation can close while later diagnostic and business reviews continue with their appropriate definitions.
Present readable diagnostics and comparisons
A diagnostic table should show readable keyword text, campaign, group and actual score with useful scope context. A performance comparison needs selected measures and dates. Keep internal identifiers, emails, provider payload fields and operational timestamps outside ordinary client output. The same display discipline should apply to Slack, web reports and downloadable files.
A requested native chart can combine compatible labelled series where supported. Respect the series limit and explain it when the request exceeds it. Longer periods can use stated weekly or monthly aggregation when daily detail is not explicit. Explicit daily requests beyond twenty points per series need split native charts with the reason, rather than silent fragmentation or an unrequested image.
Use readable dates and actual week ranges. Current-year daily labels can omit unnecessary years, while older periods retain enough context. Display metrics with no more than two decimals and compute ratios from full-precision base measures first. Missing scores, missing observations and undefined ratios should remain distinct from genuine zeros.
Offer a fuller readable file when the detail exceeds a useful conversational answer. Preserve the same selected population and avoid multiplying repeated rows into unnecessary bulk. In ordinary AI replies, place the user mention before feedback text and controls so feedback remains last. Clear output should show what the evidence establishes without suggesting the score is an account-wide performance KPI.
Leave a defensible review record
The handover should identify the inspected population, actual diagnostic observations, selected commercial measures and owner's decision. Record the supported approved effects and verified results separately from creative hypotheses or website tasks. A later reviewer should understand why a keyword was retained or paused without reconstructing the discussion from raw provider records.
Preserve deliberate non-actions. If a valuable keyword remains enabled while the team investigates relevance, record that choice and the next evidence task. If creative was brainstormed but not uploaded, say so. If the destination remains untested, assign its owner. The absence of a mutation can be an accountable outcome rather than an unfinished automatic fix.
Define the next review around a completed period or actual implementation change. Reuse the established method while obtaining current evidence where needed. A prior verified status or creative field is useful historical evidence, but it is not perpetual proof of current state before another live request.
An enterprise Quality Score audit is strongest when it uses a diagnostic to guide precise questions without turning it into a profitability verdict. It can support creative discussion, focused evidence and approved configuration changes. The result leaves actual field effects verified and diagnostic or commercial outcomes open to their own measured evaluation.