Optimise Google Ads settings for consultants: connect account configuration to evidence and owner decisions

Define the optimisation question

Settings optimisation begins with a business requirement, not a checklist of switches to turn off. A local service, an online store and a destination-based travel offer can need different targeting and measurement choices. The consultant should establish what the client needs the account to do before interpreting a configuration as unsuitable.

State the decision clearly. The client may want to limit activity to a service area, preserve an intended primary conversion population or review a campaign's budget. These are specific questions with different inputs. A broad request to improve settings should be broken into relevant investigations rather than an indiscriminate account rewrite.

Use audit findings as prompts for review. A finding can identify a configuration that deserves attention under the audit's assumptions, while the business context determines whether to change it. Passing every heuristic is not proof of a commercially successful account, and retaining a justified exception is not necessarily neglected optimisation.

Keep authority separate from advice. A consultant can explain alternatives and recommend a supported setting change. The connected owner still confirms the exact live proposal. A conversation about best practice should not become a standing permission to modify every similar setting later.

Establish the current configuration baseline

When several operators share the account, ask which recent changes are already intentional. A campaign may be paused for an offer update, or a new target may reflect an agreed investment plan. The current read establishes the setting, while the operator provides the decision context. Keeping both avoids a consultant reversing legitimate work because it differs from the engagement's older notes.

Read the relevant current values before recommending a change. A client may describe an old configuration, or another operator may already have corrected it. The baseline should identify the selected resource, actual parent context and setting values needed for the decision.

Choose a narrow baseline that fits the question. A location decision needs the campaign's relevant geographic criteria and intent setting. A budget decision needs the current budget relationship and eligibility. Neither requires retrieving unrelated creative or every product in the account simply to begin the review.

Historical performance can add context, but it is not the current configuration. A campaign with spend last month may be paused now. A current setting may have been introduced after the selected reporting dates. Retain the time sequence rather than interpreting every historical result as the effect of today's settings.

Record the baseline in understandable terms. Names, settings and meaningful parent relationships help the client assess it. Internal references remain useful for exact targeting behind the scenes, but ordinary client-facing notes should not be a dump of IDs and provider field paths.

Discover capabilities through the shared framework

The application links readable settings, declared edit actions and provider field capabilities through shared metadata. The AI can discover the appropriate resource and supported operation rather than relying on a dedicated tool for each possible sentence. This lets the same framework handle different combinations of settings questions.

Local settings discovery does not perform a provider read. It establishes available fields, relationships and action links. Account retrieval then obtains the values necessary for the selected request. A catalogue describes capabilities; it does not prove that the connected account has a particular campaign, value or eligible target.

Declared prerequisites matter. An action may require an existing target, a compatible channel or a populated relationship. The consultant should obtain those inputs before preparing the operation. A broadly writable API does not remove the need for the application's supported preparation and verification path.

When a capability is unavailable, name the exact reason. The current auto-apply recommendation subscription entry records an unsupported historical endpoint and directs management to Google Ads. It should not be presented as an available live switch merely because an old audit snippet exists.

New questions can combine fields and conditions through the framework, while new write effects still need valid implementation and verification metadata. The consultant should explain the real supported scope rather than claiming that every API field is instantly editable through chat.

Separate readable, writable and verified effects

A field can be readable without accepting updates. A writable field can lack an independently selectable read projection. A known convenience action can have a stronger preparation or verification path than a generic operation. These are distinct capabilities that affect what the final result can establish.

The settings capability metadata records independently readable effects where they exist. It also preserves cases without an independent projection. A consultant should not promise a checked final state when the actual supported operation can establish only a provider acknowledgement.

Preflight validation checks the proposed request before live mutation where supported. Verification checks the actual recorded result afterward. A valid preview does not establish later completion, and a returned receipt does not automatically prove every downstream effect the business cares about.

Use the outcome's actual wording. Verified execution, accepted acknowledgement and uncertainty require different follow-ups. An implementation limitation should not be described as a failed business result, but it also should not disappear under a universal success message.

Review the result measurement first

Agree how a measurement change will affect later reporting. Making an action secondary changes the selected primary population, so a later total can fall even if the underlying business activity is similar. The client needs that context before assessing the consultant's effect. Preserve the dates and definition revision in the record rather than treating the next smaller total as an unexplained campaign decline.

Settings decisions need an appropriate success measurement. Total conversational Ads conversions include primary actions by default and exclude secondary actions unless requested. Total primary value can include non-purchase actions. Purchase-only questions select the primary purchase population instead.

Inspect conversion-action metadata when the population is unclear. A client may have several intended primary outcomes, including leads, calls or uploaded actions. The tool should count the selected primary population as configured rather than silently redesign the business measurement while producing a report.

Changing a conversion role is a separate live decision. Show which explicit action will become primary or secondary and why the client wants that change. A reporting comparison does not grant permission to alter the action configuration.

Counting settings also require their supported action context and business meaning. The available action can set one-per-click or many-per-click where eligible, while the consultant establishes the intended outcome with the client. Do not apply a blanket counting recommendation to every action without reviewing what it represents.

For Analytics comparisons of leads or a named action, obtain the selected corresponding event. The AI can suggest a likely observed name, but the user must name or confirm the mapping. That choice belongs to the measurement definition, not an invisible assumption inside the settings review.

Investigate tracking settings and implementation

For an external measurement dependency, name the owner and acceptance test. A website specialist can inspect the journey or implementation while the account operation changes its selected setting. These tasks can progress at different times. The consultant should not use a verified account field to close an unfinished implementation check, or demand another account mutation when the remaining issue is outside Ads.

Auto-tagging and tracking templates are account settings with supported convenience actions. Their current values can be inspected and an exact operation prepared. Enabling auto-tagging or clearing a template changes the relevant stored field; it does not test the complete website measurement implementation.

Before clearing a template, establish why it exists. The client may use it for a legitimate workflow that needs review. A heuristic finding can prompt investigation, but it should not erase an intentional measurement dependency without a business decision.

Keep syntax, stored state and customer journey separate. A URL field can have the intended value while a redirect or downstream page behaves differently. A website measurement task may need a separate owner and acceptance check. The account operation cannot certify those effects solely through readback.

If the reported issue concerns missing results, inspect the selected conversion population and implementation context rather than assume the template is the cause. A settings recommendation should resolve a known dependency or test a supported hypothesis, not replace several unrelated fields until the numbers look different.

Fit location configuration to the offer

Location criteria identify the selected geography, while the positive location-intent setting determines the relevant targeting mode. These are separate choices. A request to target a country with Presence should resolve both the geographical criteria and the intended setting rather than implementing only one.

Presence can fit a business serving people in a chosen area. An interest-based setting can be deliberate for an offer whose buyers plan from elsewhere. The consultant should obtain the business purpose before treating the audit's preference as a universal instruction.

Explicit additions and removals preserve the intended selection. A country change should not accidentally remove an unrelated radius or retain a geography the client asked to replace. Show the supported steps and current context clearly so the owner understands the actual scope.

Store-linked radius work has its own prerequisites. Creating a business location group is an account task that can require provider population before later targeting. The radius action requires an existing populated group; it should not silently create that prerequisite during an ordinary campaign targeting change.

After a verified location edit, later performance can show observed activity under the resulting configuration. The stored setting does not guarantee future traffic distribution or a commercial improvement. Keep the business follow-up attached to its own dates and evidence.

Distinguish campaign language and research defaults

Campaign language criteria and keyword research language choices are different contexts. The supported campaign action sets the desired complete language selection, with an empty selection meaning all languages for that operation. Do not infer a live campaign edit from a research conversation.

A consultant should establish the campaign's intended audience and offer presentation before changing its languages. The client's website content and service capability can inform the decision, but they are business inputs rather than facts guaranteed by the current criteria.

Keyword research has its own default and explicit-instruction behaviour. If the client asks for research restricted to a particular language, that is a research selection. It does not automatically request a campaign-language change or prove that the campaign currently uses the same scope.

Keep the proposal clear when replacing a complete set. Adding one desired criterion and setting the whole desired selection have different implications. The supported operation and displayed steps must match the client's actual request, with current-state checks preserving the prepared decision.

Review network and automation choices

If the client chooses to retain a setting, record the observation that would prompt another review. A future campaign contribution report, a changed offer or a new budget decision can supply that trigger. This makes the exception accountable without calling it a defect. It also avoids repeatedly reopening the same recommendation merely because the audit continues to apply its generic heuristic.

The catalogue includes supported operations for selected network and automation settings, with their actual fields and prerequisites. A consultant can investigate whether the configuration fits the campaign's purpose. The existence of a disable action is not evidence that disabling it is always beneficial.

Obtain performance context when the client asks about the effect of a network choice. A current switch value does not establish the network's historical contribution. Retrieve the appropriate supported breakdown and dates before recommending a change based on claimed performance.

Automation settings need the same discipline. Explain which supported choice would change and what the client expects from it. Do not treat a broad dislike of automation as a factual diagnosis of every campaign, or promise that a disabled feature will produce a particular return.

Some declared paths still require provider preflight because historical live acceptance is not established. Final URL expansion is one such catalogue note. Preserve that requirement rather than presenting the action as a guaranteed successful write on every eligible-looking campaign.

Choose budget and target changes deliberately

A budget change affects an eligible live amount, while a business allowance supplies a planning reference. The convenience daily-budget path requires a non-shared daily budget with the supported campaign context. A shared budget cannot be treated as a narrow single-campaign setting without reviewing its effect.

Set operations use the relevant monetary or display unit, while raise and lower operations use their declared percentage interpretation. A target ROAS input of four hundred means four hundred percent on the supported display input path. Make the intended value understandable in the proposal rather than relying on an ambiguous raw number.

Existing target changes require an eligible existing target and supported non-portfolio context. They are not a generic instruction to create a different bidding strategy. Inspect the current strategy and field path before preparing the operation.

Use pacing to explain trajectory and reporting to explain commercial results. Neither observation automatically executes a budget or target adjustment. The owner reviews the exact supported value and target separately.

Review status and structure in context

A campaign, group, ad or keyword can be paused intentionally. Current status is configuration evidence; it does not prove historical absence of activity or commercial unsuitability. The consultant should obtain the offer plan before treating every paused resource as a problem to enable.

The empty-group convenience action discovers enabled groups without non-removed ads and freezes that selection. That rule differs from a finding that a group lacks an enabled ad. A paused non-removed ad can make the group non-empty under the former definition while it still lacks active creative.

Low-quality keyword diagnostics likewise need interpretation. The convenience selection can identify enabled positive keywords with a low Quality Score, but the score is not a profit measure. Google's Ad Rank overview describes several auction factors, reinforcing the need to avoid reducing the whole business decision to one diagnostic number.

Creation and enablement are distinct steps. A new group with keywords can begin paused under its supported creation behaviour, while a later enabled state requires the appropriate requested operation. Show the full intended sequence rather than assuming that creating the children automatically activates their parent.

Scope exclusions and negative controls

Exclusions should follow a confirmed mismatch or deliberate targeting choice. Poor short-period performance alone does not establish that a query, placement or product is irrelevant. The consultant should review the evidence and intended offer before recommending a broad removal of eligible activity.

The negative-keyword convenience actions add exact negatives at their selected campaign, group or shared-list scope. Creating a shared list is a separate action from attaching it. An unattached list does not establish that a campaign's query control changed.

Placement exclusions distinguish websites, applications, video channels and individual videos, with account, campaign or group scope where supported. The proposal must preserve the selected kind and target. Do not broaden a single campaign request into an account exclusion for convenience.

Excluding all app categories is account-wide and does not require campaign discovery. The framework can select that action locally, then prepare its required account operation. This illustrates why each action's valid target belongs in the shared capability metadata.

Review assets and creative dependencies

A creative setting can belong to an ad, an asset or an attachment. Resolve the actual resource before preparing a change. A standalone asset used in several contexts can have a broader effect than one ad replacement, while detaching a relationship does not necessarily change the asset's content.

Responsive search ad replacement uses the complete desired creative. Review all headlines, descriptions, final URLs and relevant path choices. The supported replacement removes pins, clears omitted paths and preserves the existing serving status. The owner should understand those effects before approving a seemingly small wording change.

Creating a new paused ad is a different operation. Its successful creation does not repair an old ad automatically or establish provider approval. If the client also requests enablement, that action needs to appear in the supported prepared sequence.

For a new PMax asset group, required existing assets and their declared roles are part of the supported input. The workflow does not invent missing media or treat a Search-ad input set as universal. The local dependency rules identify the valid parent and provider validation checks the applicable combination before live work where supported.

Preserve product targeting relationships

Product targeting is a tree within each relevant group. PMax uses asset groups and Standard Shopping uses the relevant ad groups. Different enabled groups can include or exclude the same product independently, so an occurrence count preserves each group membership rather than deduplicating inventory IDs.

Explicit IDs and broader branches need different evidence handling. The framework can match inventory to selected partitions for a product occurrence view, while a listing-node view describes the structure itself. A category exclusion is not merely one individual-product exclusion because it appears as one node.

Current targeting can be combined with historical performance through declared relationships when the question requires it. Performance enrichment is unnecessary for a simple current exclusion count. It becomes necessary when the client asks for a return or spend condition, with unavailable history preserved as unknown.

The supported inclusion action uses exact complete selections, preserves unrelated partitions and expands broader leaves only for the selected products where applicable. Retargeting a group's broader selection is a different operation. The consultant should keep those effects clear rather than describe every product change as the same enable switch.

Merchant eligibility remains separate from manual targeting. An included product can have another reason it does not show, and a manually excluded occurrence can exist while the inventory product remains healthy. The account operation should address the requested targeting state, not promise universal product visibility.

Follow an invented consulting settings review

Consider an invented service business, Ash Meadow Repairs. The consultant reviews an existing Search campaign after the client asks whether its settings fit a local appointment offer. The client confirms the actual service area, selected primary enquiry action and intended investment reference. These details illustrate the method, not a customer result.

The current read shows the campaign's location criteria and intent choice, while a performance report describes the selected primary result over named dates. The consultant recommends a supported Presence change based on the client's service model. It does not claim that the setting caused every historical enquiry outside the area.

The review also finds an account tracking template. The consultant asks whether it serves a legitimate measurement workflow before recommending removal. The client provides context, and that issue remains a separate task until the dependency is understood. A heuristic does not authorise an account-wide edit.

The owner reviews an exact proposal for the agreed location operation. After execution, the stored setting is checked. The consultant then plans later performance and operational review using the selected enquiry definition. Verified configuration is the immediate milestone; improved appointment quality is a later business question.

If another operator changes the campaign before execution, the stale preparation should be refreshed. If a dispatched request becomes uncertain, the target state should be reconciled before another related proposal. Both cases require evidence rather than a casual repetition of the earlier instruction.

Investigate settings in Slack

The client can ask a normal account question such as which location settings this campaign uses. The AI discovers the relevant readable selection locally, retrieves the selected account evidence and explains it in meaningful language. It should preserve the campaign already named in the conversation.

A follow-up can ask what the consultant recommends. That explanation uses the retrieved setting and business context; it does not need another identical read merely to change the wording. A question about a newly changed current value is different and needs relevant fresh evidence.

Creative requests remain conversational until account work is requested. Suggesting headlines, keyword phrases or a client explanation does not automatically call a research or mutation tool. Measured traffic estimates and live creation are separate requests with their own supported prerequisites.

When a choice is unresolved, ask a focused question. A campaign-language replacement needs the desired selection. An Analytics lead comparison needs the named or confirmed event. A duplicate group name needs parent context. Avoid requesting broad background when the thread already contains the exact target and period.

Requested charts or tables use returned provider data and supported native outputs. Their formatting should preserve metric units, names and no more than two decimals. A visual answer is a presentation of evidence, not a reason to invent arrays or silently change the population.

Prepare a coherent supported sequence

For a staged implementation, distinguish dependencies that can be planned locally from those that need a real completed result. A parent type relationship can be validated in code; a store group awaiting provider population is a current prerequisite. The latter may require a later preparation after the account evidence exists. Treating both as immediate local ordering would create an apparently neat chain that cannot execute reliably.

A multi-step request should include every intended action before preparation. The local dependency graph validates target kinds, parent references, channel compatibility and ordering. Parent creation precedes the child operation that depends on it; incompatible references are rejected rather than guessed.

Different operations can have dedicated-service restrictions. The graph's ordering does not promise that every combination is universally atomic or that every API method supports preview validation. Follow the actual supported preparation path for the selected request.

Show logical actions separately. Adding keywords and enabling their group are different steps even when the keywords form one list. If the request is too large for safe review, ask the client to split it without exposing internal numerical caps. Do not prepare an undisclosed partial sequence.

Complete dataset selections matter for filtered edits. A visible preview is not the full list of matching products or keywords. Carry the exact complete selection or narrow the scope before preparation. The owner should approve the same population that execution will use.

A supported chain remains bounded by actual capability. Reading an API catalogue does not turn every proposed resource lifecycle into an implemented edit. If a needed step is unavailable, identify it before preparing the rest as though the whole request were ready.

Review exact steps before execution

Keep the business rationale alongside the reviewed value when handing the proposal to another stakeholder. The owner needs to understand both the exact operation and the reason for considering it. Neither a raw payload nor a broad promise of better performance provides that decision context.

The confirmation proposal names the target and actual relevant parents. It states each action and intended value in plain language, with readable step sections for a multi-action request. Internal IDs and undefined parent labels should not substitute for the client's recognition of the asset.

The pending message states that no changes have been made yet and that the proposal expires after fifteen minutes. Only the connected-owner confirmation starts live execution. A consultant's recommendation, report or audit finding is not that confirmation.

Review the consequences as well as the immediate value. A complete language set, a shared relationship or a full creative replacement can have effects beyond a small displayed field. The proposal must accurately describe the supported operation so the client can make an informed account decision.

If the owner asks for a revision, prepare the revised request rather than relying on an unrecorded verbal correction to the old button. The stored proposal should represent the exact reviewed instruction. Expired or superseded proposals need a fresh preparation.

Use current checks and supported validation

The execution workflow checks authorisation and relevant current state again. Another operator can change the target between preparation and confirmation. A stale result should prevent the old proposal from overwriting the changed state and direct the client to a fresh preparation.

Supported provider preflight validates the proposed operation before live mutation. A rejected preview can establish that no live change was made by that proposal. Some methods lack the same preview mode, so the executor uses declared support rather than pretending every path had an identical dry run.

Account write coordination and current-state checks have different roles. Coordination reduces conflicting app writes, while the selected preconditions protect the prepared decision. Neither replaces the owner's authority or guarantees control over every external account operator.

Keep a rejected preparation separate from a dispatched interruption. The former can require corrected input or an eligible target. The latter can require reconciliation because a write may already have applied. The next action must follow the actual evidence, not a generic try-again instruction.

Interpret verified and acknowledged results

The progress message visibly shows change in progress and removes pending controls. It does not mention the user or claim completion. The final result reports the actual workflow outcome, including readable target and action context.

Executed and verified establishes the supported checked account result. Accepted acknowledgement means Google accepted the request while final completion lacks independent verification. Uncertain execution means the request may have applied but the complete outcome remains unresolved. These statuses require different follow-up.

Do not blindly replay an uncertain dispatched write. Inspect current state and available receipts, then prepare only confirmed remaining work if appropriate. Reposting the recorded Slack status belongs to the notification path and should not dispatch another mutation.

A verified setting is not a guarantee of provider approval, actual traffic or improved return. Website behaviour and commercial results require their own evidence. Preserve those remaining questions so a clear execution message does not close the entire business task prematurely.

Avoid unnecessary provider and storage work

The local capability and dependency catalogues describe settings and relationships without scanning account data or storing a new dependency table for each conversation. They provide the plan; provider reads supply current values. Discovery should not become an unnecessary account-wide query.

Selected relationship enrichment occurs only when the requested fields or conditions need it. A product exclusion count does not automatically retrieve performance, while a return threshold needs the declared historical relationship. Validate filters first so invalid requests do not waste provider work.

Reuse a returned performance dataset for the explanation and requested visual. A different presentation should not retrieve the same totals again. A changed population, period or current-state question justifies fresh evidence; reuse should not hide stale observations.

Durable state still matters for approved writes and configured scheduled work. Removing duplicate reads does not remove the need to retain proposal transitions, execution outcomes or configuration revisions. Those records prevent repeated controls or delayed delivery from becoming repeated mutations.

Use provider filtering for the selected scope where supported, while preserving the exact framework condition when pushdown is unavailable. An unsupported OR translation must not silently drop one alternative. Efficiency is useful only when the optimised request still answers the client's original selection.

Review later outcomes with the same definition

Choose an observation plan before confirming a material change. Name the result definition, relevant scope and practical later review. If operational lead quality matters, identify who will provide that business feedback. The account data can measure its selected conversions, but it cannot independently certify the suitability of every enquiry. A clear follow-up combines the appropriate evidence without pretending the advertising metric covers the whole business outcome.

A later report should retain the agreed measurement and name its dates. If the client changes from all primary value to purchases, explain that definition change before comparing results. A more suitable definition is useful, but it should not appear as an unexplained performance movement.

Before-and-after periods can show observations without proving causality. The offer, demand, website and other settings may change at the same time. A consultant should identify known context and remaining uncertainty instead of attributing every movement to the most recent verified edit.

Review the current setting when the client asks whether it remains applied. That is a factual configuration question, separate from historical performance. An earlier verification does not certify every future state after other operators can change the account.

Use monitoring or an explicitly saved reporting configuration where recurring observation is appropriate. A one-time optimisation conversation does not create a permanent automated rule or ongoing account-management commitment.

Leave a clear settings handoff

A declined proposal belongs in the handoff as a decision, not a failed execution. The client may reject a change because it conflicts with an offer plan or requires further evidence. Retain the rationale and any remaining investigation. This helps a successor understand why the current configuration differs from a recommendation without assuming that the confirmation button malfunctioned or that the consultant forgot the task.

Where execution is uncertain, distinguish it from a declined or expired proposal. A dispatched request may have affected the account and requires reconciliation. A fresh recommendation after that reconciliation should state only confirmed remaining work. These labels directly determine the next safe practical action, so they should remain explicit in the client record.

Record the business question, original setting, chosen action, execution outcome and remaining checks. Name actual targets and parents. A successor should be able to understand the decision without reading internal payloads or reconstructing the entire conversation.

Preserve justified exceptions. A client can intentionally retain interest-based targeting, a paused resource or a tracking dependency. The record should identify the rationale and review owner. Passing every audit heuristic is not the objective when the business has a supported reason for a different configuration.

Identify scheduled configurations separately from one-time edits. Explain which report, monitor or Rule remains active, who maintains it and what its intended behaviour is. A conversational recommendation is not standing authority for future automatic work.

Link to audits, account questions and approvals where they support the next task. The settings framework coordinates evidence and valid actions while the client retains the business decision.

Answer common settings questions

Does every audit finding need a live edit?

No. Review its assumptions, current evidence and business purpose before choosing a supported action.

Does a readable field prove it can be edited?

No. Readability, write capability and independent verification are separate declarations.

Does keyword research change campaign languages?

No. Research selections and live campaign criteria are different contexts.

Does a new negative list affect a campaign immediately?

Creation and attachment are separate operations. The selected campaign relationship determines its effect.

Does local dependency planning require a provider call?

No. The catalogue establishes ordering and capability. Current target values still require their relevant evidence.

Does verified execution guarantee improvement?

No. It establishes the checked account result. Later performance and business outcomes remain separate observations.