Ask Google Ads account questions for small businesses: get clear answers and keep the next step precise

Begin with the business decision

A small-business owner does not need to know the provider's query language to ask a useful account question. Start with the decision you want to make: whether spending fits the plan, which campaign needs attention, or whether a particular offer has suitable results. Everyday language can express that purpose. The assistant's role is to connect it with the relevant available evidence and explain the answer. A clear question is not necessarily a long question; it gives enough context to make the result meaningful.

Separate a request for facts from a request for advice or action. “How many campaigns are enabled?” asks for current account state. “Which campaign had the lowest purchase ROAS in the last thirty completed days?” asks for a ranked performance view. “Suggest better headlines for this offer” asks for creative work. “Pause that campaign” asks for a live-edit proposal once the intended target is resolved. These requests can form one conversation, but each needs a different kind of tool or reasoning. Treating them all as account writes would be inappropriate.

Supply the business facts the account cannot establish. An owner may know that a service is unavailable, that stock is limited or that the calendar is full. Those facts affect whether measured demand is desirable. The assistant can use them as context, while keeping their origin clear. It should not invent a margin, customer quality judgement or operational constraint from clicks alone. Useful conversation combines retrieved account evidence with the owner's actual intention, then identifies the next decision rather than merely producing another set of numbers.

Know which connected account is being read

The Slack channel's connection provides the account context. A business may operate several accounts, and the owner's Google login may reach more than one. A question in one channel should not silently combine them. Confirm the connected account when the subject could be ambiguous, particularly when discussing a market, brand or campaign that exists elsewhere. The assistant should use the actual authorised context rather than infer an account from a familiar channel name or a business mentioned in an earlier unrelated conversation.

If a required connection is missing or no longer readable, the assistant should explain that politely and use the appropriate connection path. It should not report a false zero or substitute data from another account. Some questions also require a connected Analytics property or Merchant account; Google Ads access alone does not make every other source available. The available tools should reflect that context. An answer must distinguish a capability that exists in the app from evidence actually available for this business's connected sources.

Account names and relevant asset parents help the owner recognise the result without exposing raw resource identifiers. Precise internal identity remains important for reading and executing the correct target, but ordinary replies should use understandable names. If duplicate names remain ambiguous, ask for the distinguishing context. An extra clarification is justified when its answer determines the target. It is unnecessary when the question plainly concerns the whole connected account and the current defaults already provide a useful performance scope.

Understand the current question defaults

The current Slack question policy avoids asking the owner to choose a campaign before every broad performance answer. If no target is named, the default is the whole connected account. If no period is given, it uses the last thirty days and states that scope in the reply. This makes a general question such as “How are we doing?” answerable without a sequence of setup questions. The response should still make the chosen account scope and time frame understandable so the owner can correct them in a follow-up.

Wording guides the default measure. Spending or budget questions point toward cost. Leads, sales or sign-ups point toward conversions and acquisition cost. Return points toward ROAS, while traffic or reach points toward clicks, impressions and click-through rate. A broad performance overview uses a short set of relevant measures, with ROAS when conversion value is recorded. These defaults are a starting interpretation, not permission to invent a purchase-only definition or a profit calculation. When the selected outcome matters, name it explicitly.

If the metric was omitted, the answer should state its assumptions and invite a correction after providing the useful evidence. When a metric is already named, it should not ask the owner to confirm the same choice again. Important ambiguities still need resolution: a duplicate asset name, a missing required destination or an unselected lead event can change the meaning or target. The goal is a responsive conversation with visible assumptions, not either endless preliminary questions or silent guesses presented as indisputable account facts.

Use explicit dates when they matter

A rolling period and a calendar month answer different questions. The last thirty completed days end at yesterday in the relevant source calendar, while last month is the preceding calendar month. Around a month boundary, those ranges can diverge. If you need a particular launch or promotion period, name its dates. The assistant should honour that selection instead of copying a dashboard's remembered interval. Exact bounds help the owner reconcile another view and prevent a graph caption from concealing an unintended range.

The current question defaults interpret a named month without a year as the latest relevant occurrence: this year's when that month has started, otherwise last year's. A named weekday is the most recent occurrence up to today. Those rules avoid unnecessary date questions, while the answer should still reveal the chosen scope. If you mean an older occurrence, say so. A named current month can include a different cutoff from a completed rolling report, so read the actual returned date basis rather than assuming every ordinary phrase uses identical bounds.

Source time zones also matter. An Ads account and a GA4 property can use different clocks, while the owner reads Slack in another location. Keep data dates separate from message delivery time. Recent measurement values can also change with processing or later recorded outcomes. For an incident investigation, state what observation time and period are actually known. A current account state does not establish when a problem began, and a single failed page observation does not establish the entire duration of an outage.

Distinguish a performance question from a settings question

Performance questions concern recorded activity over dates: cost, clicks, outcomes and their selected ratios. Settings questions concern the configuration of an account object: status, budget arrangement, targeting or relationships. Both can support one decision, but one cannot replace the other. A campaign with historical cost may now be paused. An enabled campaign may currently have no delivery. A configured target ROAS is an instruction within a strategy, while recorded ROAS is a result calculated from selected value and cost. The answer should preserve those distinctions.

The current tool framework has a catalogue for discoverable settings and another path for validated report data. Discovering a field or relationship does not itself read live account state. It tells the assistant which supported data task may be appropriate. Retrieving the relevant evidence then establishes what is true for the connected account. This separation can reduce unnecessary provider requests, but it must not be misrepresented as a completed inspection. A list of available fields is not proof that every field has a meaningful value in this account.

When the owner asks why something happened, begin with the observed performance movement and then inspect relevant state or history. Do not retrieve every possible setting just because a campaign has a weak ratio. A bounded investigation can ask which contributor changed, then obtain the particular configuration needed for the next decision. The audit guide explains wider diagnostic checks; an account question can remain focused when the evidence already identifies a narrower task.

Select the business measure you mean

Use familiar business language with enough definition to select the right evidence. “Purchases” can request purchase-specific conversions rather than every primary action. “Leads” can request the primary lead category population, while a named action can narrow the question further. “Revenue” needs care because configured conversion value can include non-purchase primary actions. If the intended numerator is purchase value, say so. The assistant should use the reviewed business metric definitions available to its tools instead of treating every similarly named provider column as interchangeable.

ROAS needs a selected value and spend scope. Acquisition cost needs cost and the selected acquisition count. Average purchase value needs purchase value and purchase count. Show the relevant inputs when a movement requires explanation. The owner should not have to guess which denominator was used. A request about profit cannot be fully answered from advertising return alone when margins and other expenses are unavailable. The assistant can explain the recorded return and identify the business evidence needed, rather than inventing a profit figure from conversion value.

Raw provider metrics can still be requested when the owner specifically wants that column. The assistant should distinguish that request from a semantic business question. A business metric catalogue exists to keep definitions consistent across language variations, not to prevent a knowledgeable user from inspecting a supported field. The answer should label the actual selection. This makes a follow-up such as “Use purchases instead of all conversions” a change in the data population, rather than a cosmetic rename of the same number.

Keep primary outcome populations understandable

Google's conversion-role guidance distinguishes primary and secondary reporting and optimisation roles. The app's ordinary conversational aggregate deliberately selects primary actions and leaves secondary actions out unless the user asks otherwise. Several primary actions can contribute to that total. This is an intended reporting selection, not an automatic deduplication repair or a claim that every action represents a unique customer.

Purchase-only value differs from all-primary value. If the account assigns value to a primary call or lead action, it can appear in the broader primary value total without being purchase revenue. Calls and offline actions can also be inspected as their own selected views. They may already belong to the broader aggregate, so adding those subtotals onto it can double-count the selected evidence. A useful answer names the population and avoids presenting overlapping categories as independent additions to a new grand total.

An individual conversion can be selected by its name, with clarification when names or context are ambiguous. If you ask which offline actions exist, the assistant can use relevant metadata rather than infer their origin from a performance ratio. If you ask for secondary results, it should honour that explicit selection where supported. These questions need the available account evidence and appropriate metric path. The app should not assume that every account has purchases, calls or imported conversions merely because those categories exist in its catalogue.

Ask what best and worst mean

“Best” and “worst” require a ranking measure. The current broad default ranks by ROAS when conversion value is recorded, otherwise by conversions. That offers a useful initial answer, but it may not match the owner's decision. A campaign with the lowest observed ROAS might serve an intentional test or a different business purpose. A campaign with no conversions might have very little activity. Name the measure when you know it, and read cost, counts and scope alongside the ranked result.

Ask for conditions that make the comparison relevant. You might want enabled campaigns, a specific outcome and a minimum amount of recorded activity over a defined interval. Those choices should be explicit, not silently introduced to manufacture a convincing winner or loser. A ratio with no valid denominator should not be treated as an ordinary zero that wins a lowest-value ranking. Limited activity and unavailable evidence deserve their own interpretation. The assistant should explain why the selected ranking is answerable and what its actual evidence covers.

A ranking is evidence for a decision, not authority to apply a change. If the owner follows with “pause it,” the conversation should resolve the intended target from the ranked answer and prepare the supported exact proposal. The assistant should not rerank under a different metric without saying so, and it should not pause every weak row when only one target was selected. A coherent follow-up preserves the chosen account, period, measure and target until the user explicitly changes them.

Use conditions without changing their meaning

A condition should retain its actual comparison. “Greater than three” excludes equality, while “at least three” includes it. “No more than” and “less than” also differ. The tool framework supports validated conditions for numeric comparisons and other appropriate operators, subject to the source and field capabilities. The assistant should map ordinary phrasing to that condition rather than flattening every threshold into the same rule. A boundary matters when it decides which targets are counted or selected for a later proposal.

Compound requests also need their logic preserved. “Enabled campaigns with cost above this amount and no selected purchases” requires both conditions. “This brand or that category” can describe alternatives. “Everything except this named group” expresses a negation. A validated plan should apply the appropriate scope and aggregate filters at the correct stage. Conditions on a group's total can differ from conditions on individual rows. Do not filter a capped preview locally and then claim it represents a complete account-wide condition result.

If a field or operator is not compatible with the chosen report, explain the actual supported selection needed. Do not silently omit a condition to return something. The owner may accept a narrower question, but the answer must state its new scope. The same applies to string matching: an exact name, a contains request and a broader pattern can select different objects. Clarify when that difference would change the intended target, particularly before an exclusion or live status operation.

Resolve names through their actual parents

A campaign, ad group, keyword or asset can have a display name that is not unique enough for an edit by itself. Use its actual hierarchy. A keyword belongs in an ad group and campaign context. An ad also needs the relevant group and campaign. Performance Max uses asset groups rather than ad groups. Account-level objects do not need invented campaign parents. A readable answer should present that useful chain without raw resource IDs or undefined labels. Internal identifiers still matter for precise execution, but they are not the owner's ordinary decision language.

When a user says “the campaign you just enabled,” the assistant should use the established conversation target if that reference is clear. When several campaigns were discussed, it should ask the question that resolves the ambiguity. The follow-up should not lose the previous context simply because the user used a pronoun. Equally, it should not force an ambiguous reference into one target to avoid asking. The correct balance is a coherent conversation with precise resolution whenever the next step depends on the target.

Creation requests also require the right parent dependencies. A new group needs an eligible campaign context, and new keywords need the appropriate group. A local dependency catalogue can help order supported steps before provider work, while live state and required fields still need evidence. Do not assume every campaign type uses the same group or asset operations. The assistant should prepare the actual supported chain and explain any required input, rather than claiming all creations work simply because a campaign and a group share familiar names.

Ask for a true count rather than counting examples

A count request should obtain the count for the intended population, not count a preview of returned examples. The current report tool instructs the assistant to use a count operation when that is the question. A response showing ten examples can be useful, but those ten do not prove the account contains only ten matching objects. The answer should distinguish the total from the displayed subset. This matters for campaign status, product exclusions and any broad account question with more matches than fit comfortably in Slack.

Define the unit being counted. Unique campaigns, keyword criteria, product-targeting occurrences and unique inventory products are different units. The same product can occur in several enabled groups with different inclusion choices. A branch-level exclusion entry is not automatically a resolved count of every underlying product. If the user asks for the number of group-specific exclusions, preserve those occurrences. If they ask for unique products across inventory, obtain the necessary resolved evidence rather than reusing another count under a new label.

A count can be accompanied by a compact breakdown when it helps the owner understand its scope. State which campaigns and group types are involved and whether relevant parents are enabled. Keep unread or unavailable evidence visible. A successful count in one dataset cannot certify an unrelated surface. If a provider result is incomplete, the assistant should explain what is known and what needs another check. That protects the owner from a confident small count produced merely because a query or preview stopped early.

Inspect product targeting as a hierarchy

Product inclusion questions need their campaign and group context. A Performance Max campaign can contain multiple asset groups, each with its own listing tree. Standard Shopping uses ad groups with its applicable product-group structure. The hierarchy can narrow from all products through category or brand branches to individual identifiers. The same product can be included in one group and excluded in another. A question about manual exclusions should inspect that advertiser choice rather than substitute Merchant disapprovals or other reasons a product is not showing.

For an exclusion-occurrence count, the app's intended business definition counts the excluded product occurrence in each relevant enabled group. In an illustrative two-group case, if one product is excluded in both and another in just one, there are three excluded occurrences and two unique products. Those figures answer different questions. Name the enabled parent scope and the relevant group types. Do not report a count of broad branches as a fully resolved count of individual excluded products unless the evidence actually supports that resolution.

Performance-filtered requests require both the targeting hierarchy and the relevant performance evidence. “How many excluded products have purchase ROAS above three?” needs an agreed period, purchase definition, valid performance inputs and the exact occurrence population. Missing activity is not automatically a zero ratio. A request to include or exclude a selected supported leaf then needs its exact target and current tree state, preserving unrelated branches. The owner should see what will change and where, rather than a generic promise to alter every appearance of a product across the account.

Compare sources using agreed definitions

Connected sources can provide useful comparisons when their definitions are explicit. Google Ads purchase conversion value and GA4 purchase revenue can be shown as separately labelled measures of related business activity. Their attribution or scope can differ without making a comparison meaningless. The assistant should select a supported pairing and retain its source labels. It should not add overlapping revenue from both systems into a new total. Nor should it claim the two measures must match exactly merely because both are monetary outcomes.

For all-primary conversions against GA4 key events, state those aggregate populations. For leads against a specific GA4 event, the owner must select the corresponding event because its name and meaning can differ between sites. The assistant can inspect actual available names and suggest a probable candidate; a suggestion is not confirmation. Current named-comparison output restrictions still apply. The answer should identify the valid supported next step rather than guessing an event, discarding the user's selection or claiming universal comparison delivery.

If the owner asks for all-traffic purchase revenue divided by agreed advertising spend, describe the resulting business measure as blended ROAS or MER. Ask which spend source is intended if that is necessary. This differs from a narrower attributed campaign return. Keep dates, time zones, traffic scope and currency aligned where the question requires it. A comparison can reveal differences worth investigating, but the assistant should verify the definitions before inventing a tracking fault to explain them.

Handle unavailable evidence honestly

A failed read does not establish zero activity. If account access, a provider request or a required source is unavailable, the answer should state the actual limitation and the relevant next step. A successful empty result has another meaning: no matching recorded activity or targets in that scoped read. A missing denominator has another meaning again: a ratio cannot be calculated on the selected basis. These distinctions should remain visible in ordinary language rather than being collapsed into a reassuring zero or an unexplained refusal.

Google's Analytics freshness guidance notes that reporting values are processed at different intervals and recent results can change. A completed day is a useful period boundary, not a guarantee that every outcome is final. Consider that evidence when investigating recent cross-source differences. The owner does not need a long technical disclaimer in every reply, but a conclusion about a recent discrepancy should not ignore the source's actual reporting basis.

Keep partial coverage honest. Policy and destination tools can preserve successful findings while identifying unread surfaces. A zero among the successful readers does not prove the unread surface is clear. A preview also does not establish full population coverage. The assistant should use these returned limitations when explaining the answer and choosing the next check. If the limitation changes the user's intended decision, obtain the required evidence rather than smoothing it away to make the conversation look complete.

Investigate a change without guessing its cause

“Why did cost increase?” requires more than repeating the percentage. Start by confirming the selected account, period and metric, then identify where the observed movement occurred. A campaign breakdown can show contributors. Relevant change history or settings can show facts that help interpret them. The owner's trading calendar can add another layer. Keep those observations separate from a causal conclusion. A budget edit near a cost increase is relevant evidence, but it does not automatically prove it caused every additional amount.

Use a bounded sequence instead of reading the entire account repeatedly. If one campaign explains much of the movement, inspect its relevant current state and history. If the change is widespread and concerns outcomes, measurement may need investigation. A creative or landing-page issue can require another kind of evidence. Select each follow-up because it resolves a particular question. This helps the owner understand the reasoning and avoids a slow series of unrelated calls whose outputs do not support the final explanation.

Verify the conclusion that matters to the decision. If the business asks whether a campaign was paused, inspect that actual state rather than infer it from zero cost. If the concern is a website fault, distinguish provider-reported destination evidence from a live journey test. If the question is commercial profitability, obtain the required economics instead of calling ROAS profit. A careful answer can still be direct: describe the observed change, the verified relevant facts and the concrete next action supported by them.

Follow an invented service business conversation

Consider an invented service business, Ashfield Window Care. The owner asks which of two enabled campaigns had the higher selected primary lead acquisition cost over a defined completed period. Suppose Home Visits recorded 300 of cost and ten selected leads, while Shopfront Visits recorded 240 and six leads. The calculated costs per lead are 30 and 40 respectively. These figures are invented to explain the flow, not customer performance or a recommended benchmark. The account question has a clear population and ranking measure.

The owner adds that Shopfront Visits serves larger jobs and asks whether it should be paused. The first ranking does not answer that commercial question on its own. The assistant should use the known job-value context and ask for the evidence needed to assess suitability, rather than assuming the higher acquisition cost proves waste. If the owner deliberately decides to pause it for capacity reasons, that intention can support a specific proposal. The business decision is separate from the arithmetic observation.

When the owner says “pause that campaign,” the conversation resolves Shopfront Visits from the established context and prepares the supported status action. The proposal names the campaign and pause step. Only after the authorised owner confirms should the live write occur. A verified receipt establishes the requested status, while later cost and business effects need their own review. If the next message says “enable it again,” retain the same clear target and prepare a new proposal rather than treating the earlier approval as ongoing permission.

Keep brainstorming conversational

The assistant can help with ordinary creative requests without querying an advertising provider. If the owner asks for five keyword suggestions around a known product, clearer headlines or a description of an offer, the conversation context may be enough. Those are ideas, not measured demand. Label them accordingly when needed. Do not automatically route every mention of keywords into Keyword Planner, or treat every headline suggestion as a request to inspect and edit live ads. Creativity should support the owner's discussion without pretending to be provider research.

A request for traffic history, competition or bid estimates changes the task. It needs the appropriate research tool and connected account context. A phrase-based traffic request can require idea discovery followed by historical metrics for the resulting selected keywords. The answer should describe the actual returned research rather than attach invented volumes to conversational ideas. Conversely, an explanation of an existing research result can reuse it. The assistant should choose the tool path based on the requested evidence, not solely on whether the sentence contains an advertising-related word.

Creation and editing are separate again. After choosing creative text, the owner may ask to add it to a specific supported ad or create an asset. That requires parent context, required fields and approval. Suggestions in the thread do not become a live change automatically. The assistant should preserve the chosen wording and prepare the exact supported operation without adding unrequested content. This lets a small business brainstorm freely while retaining a precise handoff when it decides to apply an idea to the account.

Request the chart or table that answers the question

Ask for a visual when it improves understanding. A daily trend can show timing; a contributor table can support target selection; a small comparison chart can show related metrics on the same interval. The current framework renders requested provider evidence into native Slack charts and tables. It should use the returned data record rather than model-invented numerical arrays. An explanation can reuse the same evidence, while a new period, segment or metric requires the appropriate data selection.

Native charts currently allow up to three series and twenty points per series. For a longer interval without explicit daily detail, a coarser weekly or monthly grain can keep the answer in one readable chart. State that breakdown. If daily detail is explicit and splitting is necessary within the supported capacity, explain the point limit and split. Do not silently discard dates or turn a thirty-day request into two unexplained periods. A longer explicit request can require a narrower interval, coarser grain or table.

Use readable labels and precision. Weekly labels should show real date ranges rather than raw year-week codes, and current-year dates can omit unnecessary year text. Monetary and ratio values should not spill several decimals into the reply. Series labels must preserve source and definition. A purchase-value series is not a generic all-primary value total, and a ratio should be aggregated from its valid inputs at the displayed grain. Tables should disclose previews and respect current supported delivery restrictions rather than claiming every possible layout is available.

Preserve scope through follow-up messages

A follow-up should build on the established account, period and selected populations unless the user changes them. “Show that as a chart” requests another presentation of the same evidence. “Use purchases instead” changes the outcome selection. “Compare the previous period too” changes the data task. The assistant needs to distinguish those intentions rather than restart every reply from generic defaults. The owner benefits from a coherent thread in which the answer remains attached to the question that produced it.

Pronouns and references need target resolution. If one campaign was just discussed, “pause it” can refer clearly to that campaign. If the conversation considered several groups or offers, ask which one. A user naming a parent can resolve duplicate display names. The assistant should not claim it cannot understand an ordinary reference merely because it is not a raw identifier, nor should it choose a target without enough context. Exact resolution matters most at the transition from explanation to a prepared live operation.

Retain the user's corrections. If the owner says to use the total paid-traffic metric, that selection must change the plan rather than only the wording of the answer. If they choose an explicit GA4 event, preserve it for the supported comparison path. If they correct the period, update the dates. A later reply should not quietly revert to the earlier default. This is how the conversation becomes useful for iterative investigation instead of a series of isolated answers that repeat the same misunderstanding.

Reuse evidence when the task is presentation

The current report tools return a reusable data reference for rendering. A validated record can support a requested chart or table without another provider read for the same plan. Within a request, identical data tasks can also be memoised. This is an efficiency mechanism, not permission to treat every previous observation as current. The assistant should reuse evidence when the scope and required values match, and obtain fresh evidence when the period, population or live-state requirement changes.

Local catalogues also help choose the right task. Discovering available metric definitions, settings and relationships can often be done without a live provider call. A catalogue tells the assistant what it can request; it does not tell it the account's actual value. This separation avoids unnecessary discovery reads while preserving the need for account evidence. A count operation should obtain the real population count rather than require a huge example list. A narrow selected field read can be better than retrieving unrelated settings that do not affect the decision.

Creative explanations and arithmetic on existing evidence do not inherently need another call. In the Ashfield example, dividing 300 by ten and comparing it with 240 divided by six requires no fresh account query. Determining the campaign's current status before a write does. The owner should experience a responsive conversation with a clear evidence basis, not repeated provider work whenever the assistant changes presentation. Keep the plan and returned coverage together so efficient reuse does not accidentally remove the limitations that made the original answer honest.

Turn an answer into a specific supported proposal

An account answer can lead to an action when the owner requests it and the tool supports the target. Prepare an exact proposal with the understandable asset name, actual parents and intended operation. For a budget change, show the relevant amount and currency; for a status change, show the requested status; for creative work, show the final intended content and related field effects. A vague recommendation to optimise the account is not a concrete operation the owner can safely confirm.

Check the supported operation and required inputs before provider work that is unnecessary. Local dependencies can order a parent creation before its child and prevent impossible requests from reaching a live write. Current account facts and valid target state still need the relevant reads. Support differs by asset and setting. Do not promise universal campaign creation, arbitrary shared-budget edits or an automatic website repair simply because a readable field appears in a catalogue. Explain the actual task and any remaining required input without substituting a different action silently.

Confirm and Execute is the approval point for the authorised connected owner. Before confirmation, no live edit has been made. The proposal expires after fifteen minutes, and relevant settings are checked again before execution. The approval guide explains this handoff. Multi-step requests need separated visible actions and correct dependency order, without claiming every operation is universally atomic or automatically reversible. The owner should recognise what will happen and where before authorising the prepared request.

Separate confirmed execution from business results

After a valid confirmation, a visible in-progress state tells the owner that execution is underway. The final receipt should make verified success clear and describe the target and change in ordinary language. If the provider accepted a write but the complete result was not independently verified, state that limitation. Do not turn it into a false success or blindly replay the operation. Inspect the relevant target before another mutation, especially when an asset creation might already have succeeded.

A verified edit establishes account state, not a guaranteed commercial outcome. A paused campaign's status can be correct while recorded cost takes time to reflect the new period. A created ad can be present and still await provider review. A destination edit can be applied while website implementation work remains open. A budget reduction can protect funding without improving lead quality. Keep these stages separate in follow-up answers so the owner knows exactly what has been completed and which business question still needs evidence.

Later performance should use a defined interval and the agreed outcome population. If several changes or trading events happened together, do not automatically attribute every movement to one edit. The decision record can show what was applied, when and why, while the next account question obtains the relevant result. This keeps the conversation accountable. The owner can choose another supported proposal based on later evidence, but a previous confirmation should not become standing permission for the assistant to keep altering the account until a metric improves.

Build a manageable account-question routine

A useful routine combines recurring evidence with focused questions. An owner can review cost and selected outcomes, then investigate a material contributor or current readiness issue. Scheduled reports, alerts and anomaly checks have their own configured purposes; they do not replace every conversational question. The client-reporting guide explains a clear owner–adviser handoff. The assistant can support that discussion without turning each follow-up into a whole-account audit or making a live edit from an unexplained warning.

Keep the business context current. Service availability, product stock, capacity and funding intention can change. A targeting exclusion sensible last month may deserve another review when the offer changes. A selected conversion can cease to represent the intended outcome after a website update. Name these changes when asking the next question so the assistant can select the relevant evidence. It should not assume the earlier context remains permanently true, nor invent a new business rule to explain a metric that moved.

Close investigations with a concise note: question, selected evidence, known business context, decision and owner of the next check. An explicit no-change decision can be appropriate. An unresolved data limitation should remain a defined task rather than disappear from the conversation. This makes later follow-ups easier and reduces repeated ambiguous requests. The value of account questions is a sequence of informed decisions, not the number of tool calls or the amount of data pasted into Slack.

Answer common conversational account questions

Do I need to name a campaign every time?

No. The broad performance default is the whole connected account. The assistant states the chosen scope and period so you can correct them. Name a campaign or parent when the question concerns that target or when duplicate names make the intended object ambiguous.

Can I ask for ideas without research?

Yes. Keyword brainstorming, headline suggestions and explanations can use the conversation context. Provider research is appropriate when you ask for measured history, traffic estimates or other current account facts. Ideas should not receive invented provider numbers.

Does a displayed example list give the complete count?

Not necessarily. A true count uses the selected population, while a preview shows a bounded set. Ask for the count and its unit. Product occurrences across groups, unique products and broad listing branches are different counts.

Can I move from an answer to an edit?

Yes, for a supported exact action with the required target and inputs. The assistant prepares a proposal for authorised confirmation. Execution occurs after that approval, and verified account state remains separate from later business results.