Ask Google Ads questions in plain English without losing the meaning of the answer

Start with the decision, rather than the interface

A good account question begins with something you need to understand. You may want to know whether spending increased, which campaign absorbed the increase, or whether a conversion decline is concentrated in one part of the account. The business question exists before you choose a report screen. Plain-English reporting is useful when it lets you express that question directly and receive an answer whose scope is still clear.

The convenience comes from removing the need to translate every request into report-builder steps yourself. It does not remove the need for defined measures. “How did we perform?” can lead to several valid answers because performance may mean revenue, leads, cost or efficiency. “Which enabled campaigns spent the most over the last seven completed days?” gives the conversation a concrete starting point without requiring technical field names.

A connected assistant should obtain current account facts when a request needs them. It should keep general advice distinct from those facts. A suggestion that landing-page clarity may affect results is an explanation to investigate; a statement that a particular campaign's cost increased needs account evidence. This distinction becomes especially important in chat, where both kinds of reply can sound equally fluent.

ppctor's reporting framework supports that separation through defined fields, business measurements, validated reads and requested presentation. Creative questions can remain conversational. Account questions use the relevant connected tools. The benefit is a shorter route from a decision to the information supporting it, while preserving the meaning that a careful analyst would specify in a traditional report.

Include the essential ingredients in ordinary language

Most useful reporting questions contain four ingredients: the population, the measure, the time period and the desired comparison or presentation. The population might be the whole account, enabled Search campaigns or one named ad group. The measure might be spend, purchase conversions or a named lead event. The period establishes the dates, while the comparison establishes whether you want a ranking, a trend or a change against another window.

You do not need to express those ingredients in technical syntax. “Show spend for enabled Search campaigns over the last fourteen days, highest first” is understandable and specific. “Compare primary lead conversions this week with the previous week” supplies another useful structure. If the definition of the lead population has not been established, clarify it rather than rely on an unlabeled conversion total.

A short question can be precise. Extra words do not automatically make a request better. An instruction that mixes several audiences, unrelated measures and a live edit can become difficult to review even when it is lengthy. Begin with the result you need. Then use follow-ups to investigate one part of it or request a suitable visual. This keeps both the analysis and its presentation manageable.

For recurring questions, agree on the definitions that matter to your team. Establish whether ordinary revenue questions mean all primary Ads conversion value or purchase-only value, whether reports use completed days and which GA4 event represents a lead. Those choices reduce repeated clarification. They should still be visible in the resulting interpretation so that a new colleague can understand a familiar shorthand without inheriting an invisible assumption.

Follow one question through to a concrete answer

Consider a fictional retailer, Alder Kitchen, comparing two enabled campaigns. The question is: “Which enabled campaign spent the most in the last seven completed days? Show its spend and clicks.” This asks for a ranking by spend, with clicks as supporting context. It does not ask which campaign is most profitable, which has the highest CPC or which should be paused.

For the example, suppose the returned campaign totals are as follows. These numbers are invented to demonstrate the answer structure; they are not results from a customer account. The highest-spend campaign can be selected from these totals without changing the question or adding an unsupported recommendation.

A suitable answer is: “Kitchen Essentials had the highest spend among the enabled campaigns in the selected seven-day period: £840.00 and 700 clicks.” The answer names the population and measure. It does not call the campaign wasteful because a spend ranking cannot establish that conclusion. If the user then asks for CPC, the illustrative calculation is £840.00 divided by 700, giving £1.20.

The distinction between result and judgement is the main lesson. A natural-language interface should shorten the route to this answer while keeping it faithful to the request. An analyst may investigate conversion outcomes next, but that is an additional question. It changes the evidence needed and may require a more specific conversion definition. Treating the first answer as a complete optimization decision would hide that missing step.

Understand what the underlying read selects

The worked question has a straightforward reporting shape. It selects campaign names, cost and clicks; restricts the campaign population to enabled campaigns; applies the seven-day period; and orders by cost. The following GAQL example shows the read behind that kind of answer. It is explanatory syntax, not an instruction that an ordinary Slack user must write.

SELECT campaign.name, metrics.cost_micros, metrics.clicks
FROM campaign
WHERE campaign.status = 'ENABLED'
  AND segments.date DURING LAST_7_DAYS
ORDER BY metrics.cost_micros DESC

Google's query-language overview describes selecting resources, attributes, metrics and segments, together with metadata describing compatible fields. Adding a segment changes the rows returned. A daily question therefore needs a date segment; an overall campaign ranking does not become a daily report merely because a date range filters it. That distinction explains why the question's requested detail matters.

ppctor's ordinary reporting workflow uses registered fields and validated plans rather than expecting the user to supply this query. It returns measured data with source, period and presentation information. A requested chart can reuse the returned data within that request instead of reading the same plan again. Technical identifiers remain useful internally for precise selection; they are not needed in a normal reply that names a campaign clearly.

The actual account must supply the values. The illustrative table above demonstrates how to interpret them, not how to invent a response when a read fails. If the connected account cannot be read, the reply should explain that failure or the selection needed. A syntactically plausible query and a fluent sentence are insufficient evidence that a campaign really spent the amount quoted.

Make the reporting period explicit enough to check

Time phrases carry business meaning. “Last month” means a calendar period. “Last thirty days” means a rolling window. “This month” includes the current calendar period, potentially with an incomplete current day. A useful answer should not silently substitute one for another because the substituted period is easier to display. Ask for completed days when you want to avoid partial-day comparisons.

The account's time zone also matters. A date boundary is part of the report definition, not simply a label added after the data returns. If a team operates across regions, agree which account or property time basis the question uses. When comparing Ads and Analytics, keep the reported period visible. A similar-looking date label does not automatically establish that both sources count activity at the same moment.

For a trend question, specify whether daily detail is essential. “Show the trend over the last thirty days” can support a readable weekly presentation when the native chart's point capacity would otherwise be exceeded. “Show every day for the last thirty days” asks for preserved daily detail. The latter can require multiple native charts with an explanation. The system should not drop dates simply to make one tidy graph.

Comparisons need the same care. “Compare with the previous period” usually implies a comparable window length, while “compare with last year” introduces a calendar relationship. Say which comparison you need and inspect the returned dates. Clear period wording makes follow-ups more reliable because the conversation has an established window to preserve when you ask for another measure or a more focused breakdown.

Keep the account and asset scope clear

The same campaign name can exist in more than one account, and the same ad group name can exist under more than one campaign. A plain-English request should use the established account context and enough parent information to identify deeper assets. “Show results for Summer Offers” may be adequate when the selected campaign is unique. “Show the Accessories ad group in Summer Offers” is clearer when group names repeat.

A linked Slack destination provides account context, but it should not make every ambiguity disappear. The assistant still needs to distinguish campaign, ad group, asset group, keyword and product relationships. Performance Max asset groups and Standard Shopping ad groups are different structures. A result about products needs to preserve the relevant relationship rather than flatten everything into a list of names that loses where the product is included or excluded.

Status is another population choice. Enabled assets, paused assets and historical activity are not the same selection. An enabled-campaign report can omit spend from a campaign that is currently paused but spent during the reporting window. That may be correct for the question, but it will differ from a full account total. Ask for all campaigns when that is the intended historical population.

When the result seems surprising, inspect the scope before assuming the metric is wrong. Check the selected account, parent asset, status filter, period and source. A narrow question can produce a much smaller number than a dashboard's broader total without either calculation being incorrect. The reply should make the selection understandable so that the user can decide whether it answers the business question they meant to ask.

Choose the business measure deliberately

A metric name is not always a complete business definition. Spend and clicks are relatively direct, while conversions can include different action categories and roles. An account may record purchases, forms and calls. Some actions are primary and others secondary. A useful answer should follow the requested population instead of treating every conversion-like field as an equivalent measure of success.

The current business measurement catalogue distinguishes all primary conversions, purchase conversions, leads, calls and named actions. Secondary actions are excluded by default unless explicitly requested. All primary conversion value can include value attached to non-purchase actions. Purchase-only revenue is a narrower choice. If your question is about sales revenue, specify whether you mean primary purchase value or the broader primary conversion-value total.

Ratios also need defined components. CPC uses cost and clicks. Google defines average CPC as total click cost divided by clicks in its average CPC explanation. A combined account CPC should therefore be calculated from the relevant total cost and clicks, rather than averaging campaign CPC figures without their weights. The same principle of preserving components matters when interpreting efficiency measures.

For ROAS, establish the revenue and spend scopes. Purchase value divided by advertising spend is different from all-traffic GA4 purchase revenue divided by selected advertising cost. The latter should be described as blended ROAS or MER. A conversational answer can explain that distinction without forcing the user to know provider field names, but it should not hide the definition simply because both calculations are ratios with revenue in the numerator.

Define what best and worst mean before ranking assets

“Worst campaign” is a decision phrase, not a metric. A campaign can have the highest spend and also the largest profitable sales contribution. Another can have low cost but no useful outcomes. A ranking needs a measure, a direction and often a minimum evidence threshold. Without them, the assistant risks turning an unstated assumption into a confident operational recommendation.

A more useful question is: “Which campaign has the lowest purchase ROAS among campaigns with at least £300 in spend over the last thirty completed days?” A lead-generation team might instead ask for the highest cost per primary lead conversion among campaigns with enough recorded leads. The two questions use different business outcomes. Neither should be reduced to a universal “worst” score.

Minimum thresholds protect interpretation. A campaign with one click and no conversion should not automatically outrank a large campaign as the most important issue. A high-cost campaign with a recent conversion lag also needs context. Thresholds do not prove that an asset is healthy; they define the population being compared. Keep those thresholds visible in the answer so the reader understands which candidates were considered.

After a ranking, investigate before editing. Ask whether the relevant conversion population changed, whether the date window is comparable and whether the result is complete. A request to pause the selected campaign can then use the reviewed edit flow. The evidence question and the live-action question remain distinct, even when the user expresses them in consecutive messages within the same thread.

Express conditions without losing their boundaries

Everyday language can describe precise conditions. “Spend above £500” differs from “spend at least £500.” “ROAS below three” differs from “ROAS three or lower.” A framework that supports inclusive and exclusive operators can preserve those distinctions instead of rounding them into the same filter. The user should be able to inspect the answer's population and confirm that the intended boundary was used.

Logical relationships matter too. “Campaigns with high spend and low purchase ROAS” requires both conditions. “Campaigns with no purchases or purchase ROAS below two” defines a broader population, provided each measure is available and meaningful. A request can also exclude a named campaign from a comparison. These instructions should remain structured selection choices rather than being applied loosely to a small preview after the data returns.

ppctor's reporting plans support structured comparisons and logical conditions, with compatibility checks for the selected source and dataset. An unsupported combination should produce an explanation rather than quietly ignore one condition. Missing performance is also different from a measured zero. Where the requested relationship cannot establish a metric for an item, the result should preserve the unknown coverage instead of asserting that the item passed or failed the threshold.

Keep complex requests readable. If several conditions are essential, express them as a short sentence with clear grouping: “Enabled campaigns, excluding Brand, with spend at least £500 and purchase ROAS below two.” Then check the reply's description. Precision belongs in the selection; it does not require the final Slack message to look like a technical object or expose internal implementation fields.

Distinguish a count from the examples shown in a reply

A short Slack answer often shows examples. That is sensible presentation, but the number of examples is not the total number of matching items. A reply listing five campaigns should not establish that only five campaigns exist. Ask “how many” when you need a count, and expect the result to come from the complete supported selection rather than from the visible preview.

The current reporting instructions explicitly distinguish counts from capped examples. They also keep completeness and coverage information available to the answering workflow. A limited preview can support an explanation, but it should not be treated as an exhaustive population for an edit. If the result is incomplete, the next step is to narrow the scope or resolve the limitation before making a claim about every matching asset.

Relationships introduce another counting question. A product excluded in two enabled asset groups represents two exclusion occurrences in that group-by-product population. It is one distinct product identity. Both numbers can be useful, but they answer different questions. Ask for the relationship count when you want to understand exclusions across groups, and ask for distinct items when that is the intended measure. The wording should survive into the answer.

For large results, separate the headline from the examples. A useful response can state the complete count, name the campaign and group coverage, and then show a few representative entries. It should explain any unknown performance coverage when a metric filter is involved. This preserves readability without allowing a compact Slack layout to change the substance of the analysis.

Use follow-ups that preserve the established question

A conversation is useful because the second question does not need to repeat every part of the first. After a campaign spend ranking, “Now show clicks for the same period” should preserve the period and campaign population while changing the requested measure. “Only Search campaigns” changes the population. “Make it daily” changes the grain. Each follow-up should alter the part named by the user and retain the rest unless the instruction changes it.

ppctor's reporting framework stores the report plan for supported follow-up context. It tells the answering workflow to preserve scope and series populations unless the user changes them. This is different from assuming that every past conversation is permanently available. Thread context has a bounded lifecycle, and a new conversation can need the important details restated. Precise context handling is still necessary within an active exchange.

Use explicit references when a conversation has several possible targets. “That campaign” is clear immediately after a single campaign result. After a comparison of four campaigns, “the campaign with the highest spend” is safer. A repeated name under several parents may require the parent campaign or group. A short clarification is preferable to a live action against an asset selected from ambiguous language.

If a follow-up answer changes unexpectedly, compare its period, scope and metric definition with the earlier result. The difference may be a requested change, a clarification or a failure to preserve context. A reliable reply should make that distinction apparent. The goal is not to create a conversation that never asks a question; it is to avoid making the user restate established choices while still resolving genuine ambiguity.

Request the presentation that helps the decision

A chart is useful when the shape of the data matters. A line chart can show change over time; a bar chart can compare categories. A table can show exact values and several identifying columns. Ask for the format you need rather than receiving a visual automatically with every answer. A simple account total can be clearer as one sentence than as a chart containing one point.

The current conversational renderer uses native Slack charts and tables for requested provider data. It does not ask the model to invent numeric arrays. A returned data reference can be used to render the result within the same request, reducing unnecessary repeat reads. A chart request should remain connected to the measured dataset, its period and the selected populations.

Native chart capacity affects long trends. The ordinary chart limit is twenty points per series, with up to three series. Automatic grain can choose a readable weekly or monthly presentation when daily detail was not required. If the user explicitly asks for every day over a longer period, consecutive charts can preserve that detail with an explanation of the point limit. A single chart should not be achieved by silently omitting the final days.

Multiple measures also need readable labels. Cost and revenue can appear together when requested, while a rate and a count may be difficult to interpret on the same scale. Preserve the user's intent and explain the measures involved. Monetary values and ratios should be presented clearly, with no more than two decimal places in ordinary output. A chart that looks tidy but hides the source or units is less useful than a slightly fuller explanation.

Compare sources without pretending they are identical

Some business questions need both Google Ads and GA4. A user may want to compare advertising purchase value with site purchase revenue, or a named Ads lead action with a GA4 form event. The conversation can support those comparisons when the definitions and selected populations are clear. A difference between sources is not automatically an error and should not make a legitimate comparison impossible.

The current measurement catalogue supports reviewed cross-source relationships. All primary Ads conversions can be compared with GA4 key events. Purchase-only measures have their own definitions. Leads need a corresponding GA4 event, because event naming and implementation can vary by site. The assistant can list observed events and suggest a likely candidate, but a named or confirmed event establishes the selection rather than an unverified guess.

A request such as “Compare Contact Form conversions with form_submit_contact” supplies the individual measures directly. The result should label each source and preserve the period and date basis. It should not add the two counts together as though they represent separate business outcomes. Comparing records is different from combining them into a total, particularly when both systems can describe the same underlying activity.

Revenue questions need the same discipline. All primary Ads conversion value can contain more than purchases. GA4 purchase revenue describes the selected purchase-revenue scope. If the business wants all-traffic revenue divided by advertising spend, say blended ROAS or MER and clarify the spend source. The reporting guide explains how source labels, populations and periods keep these comparisons useful.

Interpret missing results before drawing a conclusion

An empty result can mean several concrete things: the selection matched no rows, the relevant source was not connected, the requested field combination was invalid, or the provider read failed. Those are different outcomes and should receive different explanations. A failure to retrieve conversion data is not evidence that conversions were zero. A failed report plan is not evidence that Google never offers daily values.

The reporting tools return structured errors and selection information so that the answering workflow can respond to the actual problem. A non-retryable invalid plan should be corrected rather than repeated unchanged. A missing source should lead to the relevant connection requirement. A genuinely empty selection should be described with its scope. The answer should remain grounded in the returned outcome instead of substituting a broad limitation claim.

Unknown coverage also matters in filtered results. If a product relationship exists but its requested performance measure cannot be established at the required level, the framework should not treat that missing value as zero. The answer can separate examined, matched and unknown coverage. This lets the user see what the condition actually established rather than receiving an apparently complete count built from incomplete evidence.

When investigating an empty answer, change one meaningful selection at a time. Check the account and period, inspect status restrictions, confirm the named conversion or event and review the source connection. Do not broaden everything simultaneously and then claim the original question was answered. A careful follow-up preserves the business question while identifying the precise condition that prevented a reliable result.

Keep creative work conversational when no measurement is requested

Not every useful advertising conversation needs a provider call. Asking for five headline ideas, alternative product descriptions or campaign names is creative work. The answer can use the product context and the user's constraints without reading current campaign performance. Starting a research report simply because research tools exist would make that interaction slower and change what the user asked for.

ppctor's current instructions explicitly separate creative assistance from measured account claims. Ordinary keyword suggestions, copy drafts, naming and general explanations can be answered directly. If essential product information is missing, the assistant should ask for it rather than invent product facts or run unrelated research. Suggested keywords should not be described as high-volume or low-competition unless provider evidence supports those claims.

The distinction changes when the user asks for ideas and historical metrics. “Suggest keywords for this product” asks for ideas. “Suggest keywords and show their average monthly traffic and CPC range” asks for both creative selection and research. The second request can use the appropriate research path, while keeping its targeting and returned measures clear. Research adds evidence; it should not be an unsolicited requirement for brainstorming.

Creation in the conversation also differs from publication. Drafting five headlines does not upload an ad. Suggesting five keywords does not add criteria to an account. If the user then requests the live addition, prepare the supported proposal for review. This boundary lets the conversation remain useful and flexible without turning every creative exchange into either an API call or an unapproved edit.

Distinguish observations from explanations of what caused them

A measured change establishes what happened in the selected data. It does not always establish why. If CPC increased, the account read can show cost and click changes. A breakdown can show which campaigns contributed. A historical check can identify relevant recent changes within its supported scope. An explanation about competition or user intent may still be a hypothesis until the appropriate evidence supports it.

Use questions that progress from observation to investigation. First ask whether the change exists over comparable periods. Then ask where it is concentrated. Next inspect relevant conversion definitions, status changes or settings. This sequence prevents an assistant from inventing a plausible cause before it has established the pattern. It also keeps provider calls connected to a specific investigative need.

Consider a fictional result where clicks fall by twenty percent and cost stays similar. CPC rises as a mathematical consequence. That observation does not reveal whether bids, competition, traffic mix or eligibility changed. A campaign breakdown may show the shift occurred in one campaign. A later settings read can help investigate that campaign. Each step narrows the uncertainty without presenting the previous step as a final diagnosis.

General guidance can still help. The assistant can explain common reasons for a trend and suggest which evidence would distinguish them. It should label that guidance as general rather than as a fact read from the account. The CPC investigation guide provides a more detailed example of separating the measured increase, its contributing segments and the checks needed before acting.

Turn an answer into a reviewed action without skipping the boundary

Some conversations naturally move from analysis to action. The user asks which campaign is inefficient, reviews the evidence and asks to pause the selected campaign. A reliable workflow carries the resolved target forward while introducing the approval step. It does not treat the earlier analytical question as implicit permission to change the account.

For supported edits, ppctor prepares a proposal showing the exact target and action. A deeper asset includes the relevant parent context. The user confirms and executes the proposal before the live API edit occurs. The proposal expires after fifteen minutes, and relevant authorization and settings are checked again. These checks matter because an account can change after the earlier read.

Multi-step work should remain visible. A request to add keywords and enable an ad group contains two logical actions, even if both concern the same group. A request to create a campaign and its dependent assets requires the correct dependency sequence. The proposal should present clear steps rather than hide them inside a technical paragraph. Unsupported or overly broad work should be narrowed before execution.

Execution evidence also has a specific meaning. A successful verified setting change confirms the supported operation, not a guaranteed business outcome. An ad accepted by the API can still need policy review. An uncertain execution should not be replayed casually. The approval workflow guide explains how a conversational request becomes a concrete, reviewable action while keeping analysis and authorization separate.

Use the account-linked Slack workflow deliberately

Begin in a Slack destination connected to the intended account. Mention the app to start the account question, then use the same thread for related follow-ups. The account link supplies context, while the wording supplies the measure, period and population. If the required source is not connected, resolve that connection rather than asking the assistant to infer unavailable account data.

A practical opening question is: “Show enabled campaign spend for the last seven completed days, highest first.” A follow-up can ask: “Add primary purchase conversions for those campaigns.” Another can request: “Show that as a table.” Each step has a clear change. The conversation should preserve the original dates and campaign population unless the user changes them.

Keep the destination audience in mind. Channel members can see posted replies and files. A normal answer should use readable asset names and genuine parent context instead of displaying raw identifiers or private owner information. If the output is too large for an ordinary reply, ask for an appropriate detailed report or narrower selection. A large file should still go only to an intended audience.

Use a new explicit question when the task changes substantially. A product-feed investigation and a campaign revenue comparison can share an account, but they do not share every selection. A creative request can remain in the same thread when it uses the established product context, without being forced through the earlier report plan. Good context handling preserves what remains relevant and changes what the user actually changed.

Build a team habit around answerable questions

A team does not need a rigid prompt library to benefit from conversational reporting. It needs a few shared habits: name the period, choose the business measure, identify the population and distinguish analysis from action. Those habits improve ordinary questions while leaving room for follow-ups. They are especially useful when several people use the same connected account for different decisions.

A fictional agency, Birch Reach, agrees that its weekly client check begins with spend, primary outcomes and the actual reporting dates. Analysts can ask additional questions in their own words. Before posting a recommendation, they confirm whether the result is complete and whether the conversion definition fits the client. This creates consistency without requiring every analyst to memorize provider field names.

The agency also gives staff examples of changes in intent. “Suggest a headline” is creative. “Which existing headline performed best?” requests measured evidence and depends on the supported reporting surface. “Upload this headline” requests a live edit. These examples reduce unnecessary calls and prevent the model's access to tools from becoming a reason to use them for every sentence.

Review misunderstandings as definition problems where appropriate. If colleagues disagree about ROAS, establish the revenue and spend scopes. If a count seems too small, check whether it came from examples or a complete selection. If two sources differ, inspect their definitions before calling either incorrect. The conversation then becomes a practical analytical interface rather than a place where ambiguous labels are repeated more confidently.

Keep the result useful after the conversation ends

A useful answer can be understood later by someone who did not participate in every follow-up. It names the relevant measure, source and period, and describes the population when that matters. A statement such as “Kitchen Essentials had the highest spend” is incomplete without the comparison set. “Highest spend among enabled campaigns over the selected seven completed days” preserves the analytical meaning.

When a result leads to a decision, record the decision separately from the supporting observation. A campaign ranking is evidence; a choice to investigate or pause a campaign is an action. Keeping that distinction makes handovers clearer. The next analyst can inspect the evidence and understand what was decided without assuming every remark in the thread was an approved instruction.

For recurring reporting, consider the configured Reporting workflow rather than relying on a conversation to become a schedule implicitly. For a continuing condition, consider the relevant monitoring configuration. A one-time question and a saved schedule have different lifecycles. The user should intentionally choose when the same account check needs to run again and where its output belongs.

Conversation context is bounded, while useful business definitions may need a longer-lived home in team documentation. Keep agreed conversion choices, reporting periods and source scopes somewhere colleagues can review. A future question can then restate the necessary selection quickly. This approach preserves the flexibility of plain English without depending on an assistant to remember every historical exchange indefinitely.

Frequently asked questions about plain-English account reporting

Do I need to know GAQL to ask a question?

No. Describe the measure, population and period in ordinary language. The connected reporting framework uses defined fields and validated reads behind the conversation. A technical query can explain how a report works, but it is not a prerequisite for an ordinary account question. The answer still needs actual returned data rather than plausible invented values.

Will every message call Google?

No. Creative suggestions, copy drafts and general explanations can stay conversational when no measured account evidence is needed. Questions about actual settings, existing assets, performance or explicit research use the appropriate tools. Asking for a draft and asking to upload it are separate requests, with the latter requiring the supported edit approval flow.

Can I ask for a chart after receiving a total?

Yes, when the requested data and supported chart are available. The chart needs the appropriate breakdown or time series; a total alone does not supply daily points. Within a request, the framework can reuse a returned dataset for rendering. Follow-ups should preserve the established scope while obtaining any additional detail that the new presentation genuinely needs.

Can the assistant decide which GA4 event means a lead?

It can inspect observed event names and suggest a likely candidate, but the corresponding event must be named or confirmed for that comparison. Sites use different names and implementations. A plausible name is not sufficient evidence of the business definition. Once selected, the event should remain explicit in the comparison.

Does asking which campaign is worst pause it?

No. The analytical question retrieves and interprets the relevant information. A live action requires a separate supported proposal and confirmation. Define the performance measure first, review the selected campaign and then request the action if appropriate. This preserves the distinction between a conversational recommendation and permission to change the account.