Turn a spending concern into a defensible keyword decision
Which Google Ads keywords waste money? An agency guide to outcomes, search intent and precise approved changes
Start with keywords that spent money without meeting the client's agreed outcome threshold. Check the reporting period, conversion definition, search terms and business context before deciding whether to pause a keyword or exclude a particular search.
Define the decision behind “waste”
A useful answer identifies the keywords that spent money without producing the agreed business result during the chosen period. It then explains the evidence that makes those rows worth reviewing. The word waste contains a commercial judgment; Google Ads supplies measurements, while the client supplies the objective and acceptable tradeoff. An agency should establish that relationship before presenting a ranked list as a deletion queue. A keyword can have poor short-term purchase value, contribute to a longer sales journey, or expose a broken measurement setup. Those possibilities require different decisions. The first response should therefore state what was measured and why the candidates were selected, rather than declaring that every click without an immediate conversion was money thrown away.
Ask which decision the client intends to make. They might need to protect a limited monthly allocation, investigate a recent change, remove searches for services they do not provide, or review an experiment at its agreed stopping point. These are concrete purposes with different evidence requirements. A service mismatch can be established from the offer and observed search intent even when the performance sample is small. A marginal acquisition cost requires reliable outcomes and enough relevant history. An agency's reporting should make the distinction visible. Otherwise a cheap irrelevant term and an expensive commercially relevant term can appear beside each other under an identical “bad keyword” label, encouraging a broad response that fixes neither problem accurately.
Agree on a candidate rule in plain language. For example, review enabled Search keywords that spent more than the client's chosen amount and produced no primary purchase conversions in the last thirty completed days. Another account might require primary leads and a maximum acceptable cost per lead. The rule should identify the period, currency, status population, outcome family and whether the threshold includes equality. It is a selection for investigation, not a claim that a universal Google threshold has been crossed. Naming the assumptions protects the client from interpreting the assistant's default ranking as their commercial policy. If no acceptable threshold exists, describe the largest spend-without-outcome candidates and ask the client how they want to judge them.
Keep the account's wider purpose in view without losing the individual target. Brand protection, seasonal discovery, recruitment, repeat purchase and new customer acquisition can all involve different expectations. A client may intentionally fund exploratory work that will be reviewed separately. Record that intention before treating its cost as waste. At the same time, do not use “awareness” as an automatic excuse for every weak row. Compare the observed result with the stated objective and the evidence available for it. The agency earns confidence by showing a reproducible decision process: this keyword incurred this cost, this selected outcome was reported, this commercial requirement applies, and this is the next proportionate action.
Establish the client, period and keyword population
Begin with the connected Google Ads account and the reporting dates. An agency can discuss several clients in one working day, and a familiar campaign name is not enough to establish ownership. Use the channel's account context and confirm any explicit account switch before reading or proposing a change. Keep account identity understandable through the client's name and campaign hierarchy; identifiers belong to the underlying request rather than the explanatory table. A finding copied into another channel should still explain which business it concerns. This is especially important when an agency uses repeated campaign names across accounts, such as Brand, Services or Seasonal Offers. The investigation must resolve the actual resource within the authorized account, not whichever repeated name appears first.
Write out the dates behind “last thirty days,” distinguishing a rolling period from the previous calendar month. For a completed-day review, use the account's reporting time zone and identify the final completed date. A client who asks for today should receive a clearly labeled incomplete-day result. Comparisons need equally clear boundaries. A thirty-day candidate list compared with seven days of conversions is not a coherent performance test, even when the columns look tidy. Recent outcomes may also mature after the first report. Save the period with the result so a later follow-up can reuse the same definition and distinguish newly reported outcomes from a genuine change in the keyword's traffic.
Decide whether the question concerns keywords that are enabled now, all keywords that spent during the period, or a particular campaign's current candidates for action. These populations differ. A paused keyword may explain historical spending but require no additional pause. A currently enabled keyword may have been activated only recently. Current status does not reconstruct its status on every reporting date. Include removed or historical rows only when the report supports that population and the purpose requires it; do not turn a historical resource into an editable current target without checking. State the population in the heading or first explanatory sentence. That small discipline prevents clients from confusing an account spending history with a list of changes still available to execute.
Use the actual keyword reporting surface for Search keyword questions. Performance Max listing groups, Shopping products and Display targeting are different entities with different hierarchies. A campaign's aggregate cost cannot be distributed across imagined Search keywords merely to satisfy a familiar table format. If the requested account contains no applicable keyword population, explain which relevant surface is available and offer that investigation. The same principle applies when the client narrows the scope to one ad group: resolve that group and its campaign, then read its keyword performance. Do not download every account's data and improvise an approximate filter from visible names. Precise scope makes the result more reliable and keeps unnecessary provider work out of the conversation.
Select the outcomes that matter
For a sales account, distinguish primary purchase conversions from all primary conversions. The broader set can include other valuable actions and values, so it should not be labeled purchase revenue automatically. The reporting framework supports outcome selection rather than treating every account as if it had one universal conversion. Use the selected definition consistently across the keyword table, campaign comparison and requested chart. When a client asks whether a keyword made sales, primary purchase outcomes are a clearer starting point than a mixed count of purchases, calls and form interactions. If the client deliberately wants all primary outcomes, say so. The metric remains valid, but it answers a broader question than completed purchases alone.
For a lead account, combine the relevant primary lead categories when the client asks for leads, while allowing a named action when the question is more specific. The account may use several lead stages or collection sources. Their categories and names do not prove that they represent different people. Call, lead and offline groupings can overlap, so adding their group totals can double count the same reported action membership. A candidate rule based on all primary leads should be applied to that selected population directly. If the client wants “Contact Form” only, resolve that action by its exact name and handle ambiguity openly. The agency should not silently choose a similarly named form or substitute an unrelated action merely because it has more data.
Secondary actions are excluded from the normal primary-outcome comparison unless the client explicitly requests them. They can still be useful diagnostics: a product interaction or a secondary call signal may explain engagement without meeting the agreed acquisition goal. Keep those roles distinct. The conversion catalogue can expose configured actions even when they had no activity, helping establish whether the intended action exists before concluding it generated zero results. It can also identify the action's current primary role, category and source. These are configuration facts, not proof that the website sent every event correctly. Where the measurement remains uncertain, label the performance conclusion provisional and investigate the affected implementation rather than presenting a fabricated zero.
If the client requests a GA4 comparison, preserve the source and event selection. Ads leads can be compared with a specifically confirmed GA4 event; a likely name can be suggested from the available event list but should not become a silently assumed correspondence. Ads primary purchase value and GA4 purchase revenue can be presented as separate series with their respective scopes. A difference between them does not prove keyword waste, since collection, attribution and traffic scope can differ. Never attach account-wide GA4 outcomes to each keyword unless the chosen reporting dimensions support that relationship. Use the comparison to answer the measurement question it can actually answer, then return to the keyword's observed Ads performance and the client's commercial rule.
Build a useful spending and outcome table
A first table usually needs keyword text, match type, ad group, campaign, cost and the selected outcomes. Add outcome value, cost per outcome or return only when those measures fit the client. These columns allow a reader to understand both the candidate and its parent context without receiving a technical resource dump. Sort by the measure relevant to the decision: total cost for largest spending exposure, cost per lead for acquisition efficiency, or purchase return for an ecommerce threshold. Include the threshold and population above the table rather than repeating a long explanation in every row. The point is to make the next decision readable, not to reproduce every field the provider can return.
Compute ratios from the appropriate totals. Combined cost per conversion is total cost divided by total selected conversions; it is not the arithmetic mean of individual keyword costs per conversion. Likewise a grouped return uses total selected value divided by total cost. When the denominator is zero, the ratio is undefined rather than a meaningful zero. Show that state explicitly and retain the underlying cost and outcome count. A no-conversion candidate still has useful evidence, but a made-up ratio would distort sorting and thresholds. If several rows are grouped for a client summary, preserve their population and calculate the grouped measure from those rows' totals. A neatly formatted average of averages can be mathematically wrong while appearing entirely credible.
Keep precision appropriate for the metric. Currency and ratios should appear with readable rounding, while comparison conditions should use the underlying values rather than rounded display text. A row displaying a return of three can lie just below or above a threshold of three. If the distinction changes selection, explain the boundary without adding six decimals to every client-facing cell. Counts of attributed conversions may be fractional, so do not silently force them into whole-person totals. The report is describing attributed measurement, not enumerating unique customers. Use the account currency for Ads cost and do not combine accounts with different currencies into one unqualified spending total.
Separate the full matching population from the displayed examples. A Slack preview may show the most important rows while a count or aggregate covers more. The assistant should state how many rows match, how many are shown and whether a data limit prevents completeness. A top-ten list is not evidence that the account has only ten weak keywords. Conversely, a provider error is not an empty candidate set. If a requested output is too large, narrow the scope or offer a supported fuller report rather than silently cropping the finding and then proposing action against an unseen population. This distinction matters most when the client says “pause all of those”: the referent must be the verified selected targets, not an ambiguous mixture of preview and total.
Separate evidence from an immature sample
A keyword with one click and no reported purchase does not provide the same evidence as a keyword with sustained relevant spending and no selected outcomes. The client's risk tolerance determines when investigation becomes intervention. There is no universal click count that establishes waste across every product, buying cycle and lead process. Show the observed scale and ask how it relates to the account's target. A high-value service can tolerate a different acquisition pattern from an inexpensive replenishment product. The agency should avoid inventing a statistical guarantee from a small table. The immediate decision can still be practical: keep observing, investigate the offer, exclude a demonstrably irrelevant search, or pause exposure under an agreed budget constraint.
Account for the time between the advertising interaction and the reported outcome. A sales journey or offline import can mean the newest part of a period is less mature than its beginning. Align the reporting date basis when comparing cost and outcomes, and explain any deliberate alternative. A conversion-date view answers when outcomes occurred; an interaction-date view relates outcomes to the earlier advertising interactions. Neither should be mixed casually with another source's dates and presented as a precise same-day efficiency verdict. The current reporting framework can choose the relevant basis, so state which was used for the decision. Repeat the review on a mature window when the result depends heavily on recent unresolved activity.
Distinguish learning about a new keyword from evaluating an established one. A recently added term may be testing a known commercial hypothesis and have a planned review date. An old term may have accumulated enough spending to deserve a sharper decision. Inspect the business context and bounded change history when timing matters, without assuming the history surface supplies unlimited retained evidence. A launch date inferred from the first visible click can be wrong. Use current resource information and available history as separate facts. The client can then decide whether the experiment has reached its agreed exposure limit, rather than receiving a simplistic judgment that every fresh keyword must immediately equal the account's mature average.
Do not hide ongoing cost behind endless calls for more data. When the client has a hard spending constraint, a cautious pause can be appropriate before a causal explanation is complete. Describe it as an exposure decision under that constraint, not proof that the keyword could never work. Record the selected targets and the reason so a later reviewer can understand the choice. Similarly, a clearly irrelevant search may warrant an exact exclusion even if it has little spend, because the client's offer rules it out. The important distinction is between what the evidence establishes and what the owner authorizes. Practical caution and honest uncertainty can coexist without turning the report into a refusal to act.
Read keyword and search intent separately
A keyword is the advertiser's targeting term; a search term is the user's reported query. The two are connected but not interchangeable. A weak keyword total can contain both useful and irrelevant searches. Pausing it removes that targeting resource, while excluding a specific query changes eligibility for the exclusion's defined scope and match behavior. Start from the keyword candidates, then inspect supported search-term detail when intent is the unresolved question. Keep the keyword's aggregate performance alongside the query examples so the client can see why the deeper review was opened. Do not label a query's cost as the cost of the entire keyword, or assume that one irrelevant example accounts for every unsuccessful click in the aggregate.
Google explains that some low-activity queries are omitted from the search terms report for privacy, while search terms insights can group such activity without exposing individual queries. A visible query list is therefore not necessarily an exhaustive inventory of all searches. Use the provider's actual reporting surface and acknowledge its coverage. If visible term cost does not reconcile to the wider keyword or campaign total, investigate the scope and coverage instead of manufacturing missing terms. The agency can still identify actionable observed searches. What it cannot do is claim that every unseen query was irrelevant or that the returned list completely explains all keyword spending.
Read intent in the client's offer context. A query for repairs is irrelevant to a business that only sells replacement equipment, but relevant to a service branch of another client. A “free” term can describe a free consultation that the client intentionally offers; it is not automatically waste. The same applies to brand, competitor, instructional and location words. Ask for missing offer facts when they materially affect the decision. Creative discussion can help formulate likely relevance categories without needing a provider call for each phrase, but the distinction between AI interpretation and observed account activity must remain clear. A brainstormed negative is a suggestion, not a search proven to have consumed spend.
Negative match behavior deserves explicit attention. Google's negative keyword guidance distinguishes negative match types from positive targeting and explains that negative exact match does not exclude searches with extra words. The dedicated additions in ppctor prepare exact-match negatives. Do not tell the client that one exact exclusion removes an entire theme or all variants. Review each proposed text, its match type and its parent scope. If the client intends a broader exclusion, first establish whether the specific supported action can express it. An unsupported breadth should receive a clear explanation, rather than an exact exclusion described as though it fulfilled the broader instruction.
Preserve campaign and ad group ownership
The same keyword text can exist in several ad groups and campaigns, with different match types, offers and destinations. Treat each actual keyword resource as a separate target. A report can group text for a summary when the user requests that perspective, but it must retain the underlying parents before any edit is prepared. “Pause garden pump” is ambiguous if the term appears in Brand Equipment, Repair Services and Seasonal Sales. Present the candidates by keyword, ad group and campaign, then resolve the intended occurrence. Avoid replacing that hierarchy with a visible numeric identifier and expecting the client to recognize it. Names and meaningful parents are the context the person needs to approve the change.
Campaign purpose can make identical terms perform differently. One group may send users to a service page while another promotes products. One campaign may target a different country or serve a different commercial goal. Review these differences before interpreting the poorer row as a duplicate that should disappear. Duplicate wording does not mean duplicate targeting purpose, and aggregate text totals can hide a useful branch. A client who asks to pause the weakest occurrence should receive the specific occurrence selected by the agreed measure. A client who asks to remove a phrase everywhere needs an explicit broader target set and confirmation of each supported effect. The two requests should not collapse into one default account-wide action.
Negative scope also follows ownership. An ad group exclusion can preserve eligibility elsewhere in the campaign; a campaign exclusion affects the campaign's eligible matching scope; a shared list has effects through its attachments. Creating a list alone does not attach it to campaigns. The current dedicated adapter checks the selected list's type and enabled state, handles the attachment relationship and verifies the resulting membership. Explain those effects separately. A client should never approve “create the list” believing that every campaign has already been protected. Where a request combines creation, terms and attachments, show the sequence and resolve the actual supported relationships before offering execution.
Keep observed performance and planned target identity linked through the conversation. A follow-up such as “pause the worst one” should refer to the established ranked population and measure, with a proposal that repeats the resolved target and effect. If several rows tie or the reference is unclear, ask a focused question. Do not repeat the entire account discovery when the previous verified context remains available, but do recheck the relevant current settings before execution. A durable, understandable target description lets the client see continuity without exposing internal request syntax. It also helps an agency reviewer catch a misplaced campaign or ad group before an authorized change affects the wrong commercial branch.
Check creative, destination and measurement
Poor keyword outcomes can originate after the search rather than in the targeting text. Review the applicable ad and landing destination when the term appears commercially relevant but repeatedly fails the selected objective. A mismatched offer, unclear pricing, unavailable product or broken form can impair results. The report should distinguish observed Ads performance from facts about the page that have actually been inspected. Provider destination findings are useful evidence, but a destination audit is not proof that every live page and form works. Assign website checks to the appropriate owner and include their verified findings in the investigation. Do not promise that pausing a keyword repairs an unrelated checkout failure.
Policy and configuration also affect interpretation. A keyword can have historical spending while its group or ad is paused now; a disapproved creative can restrict current serving. These current facts explain available actions and possible exposure, not necessarily the cause of all earlier performance. Read them when the decision requires them, rather than appending a full policy audit to every keyword question. The agency can use the disapproved-ad review for that specific branch and return to the original candidates. A narrow investigation should expand because evidence points to another surface, not because the assistant has a habit of invoking every available tool.
Quality Score can prompt a component review, but the displayed score is a diagnostic rather than an auction input or a business target. A low score is not enough to establish wasted spend, and a higher score does not guarantee profitable outcomes. Inspect keyword-level expected click-through rate, relevance and landing-page experience where the evidence warrants it. Keep the business measure alongside those diagnostics. An agency can improve a relevant destination while keeping the keyword, or pause exposure temporarily under a spending constraint. These choices depend on the offer and outcomes, not a rule that every keyword below an arbitrary diagnostic value must be removed. The Quality Score guide supports that narrower review.
Measurement checks should answer a specific concern. If the selected action disappeared, the count changed sharply, or Ads and GA4 diverged around a site release, review the configured action, event mapping and implementation evidence. Distinguish role changes, counting settings and duplicate collection; they are different mechanisms. A mixed primary-value total is not proof that purchases were duplicated. A GA4 event count is not a unique lead count simply because its name contains “lead.” Where the issue remains unresolved, report the spending and available outcome facts without upgrading them into a definitive waste estimate. The client can then prioritize a measurement repair and a bounded exposure decision as separate tasks with separate owners.
Make comparisons and filters explicit
Apply the client's operators precisely. “More than,” “at least,” “below” and “no more than” do not describe the same boundary. A condition of cost greater than the selected amount excludes equality; at-or-above includes it. Combining conditions requires the same discipline. “High cost and no purchases” selects rows meeting both conditions, while “high cost or low return” creates a wider union. State the expression in readable language near the result. The framework can represent grouped conditions instead of inventing a special tool for each sentence. That lets the agency reuse the same reporting logic for different thresholds while preserving the client's exact instruction.
Review nulls and zero denominators separately. A missing outcome field is not a confirmed zero, and an undefined cost-per-outcome value should not be compared as if it were an ordinary number. A provider failure must remain a failed read rather than an empty set of weak keywords. If the client asks for no purchases, the system needs a verified purchase selection and the relevant report coverage before applying that condition. Explain unavailable measures when they affect the candidate set. This is especially important when a report joins multiple sources: a missing GA4 side should not cause an Ads keyword to be classified as having no business activity.
Compare like periods and populations before ranking a change in efficiency. Seasonality, different weekday composition and changing campaign scope can affect a before-and-after view. Name the periods and preserve the outcome definition. A keyword's decline within its own matched scope is different from the account average moving because more cost went to another branch. Inspect grouped totals and selected row contributions as needed. Do not present a difference in averages as proof of a keyword-level change when the populations differ. The agency should use the comparison to locate the branch requiring attention, then read that branch's actual evidence rather than solving a mixed-scope arithmetic problem with an immediate pause.
Keep historical facts and future estimates apart. The report can say how much the selected keywords cost during the chosen period. It cannot guarantee the same amount will be saved next month by pausing them. Auctions, traffic allocation and other campaign behavior can change after an edit. The account might spend elsewhere, spend less, or produce a different outcome mix. Call the historical amount reviewed spending, not guaranteed recoverable savings. If the client wants a budget plan, use an explicit scenario with assumptions and separately supported budget actions. That distinction produces a more useful recommendation than multiplying last week's cost by four and presenting the result as assured money returned to the business.
Follow an illustrative agency investigation
Consider an invented example: Mira Cole at Alder Media reviews Search advertising for the fictitious Juniper Equipment Workshop. These people and businesses are illustrative, and the situation is not a customer result. The client asks which keywords waste money because the month feels expensive. Mira first establishes that the question concerns purchases of replacement equipment, not service inquiries, and selects thirty completed days in the account's currency. The assistant retrieves the relevant keyword cost and primary purchase outcomes with ad group and campaign names. It presents the largest spending candidates without declaring that every row is useless. The review has a defined population and a commercial objective before any action is discussed.
One candidate belongs to Repair Enquiries, where the client's selected purchase definition omits the campaign's intended leads. Mira separates that branch from the ecommerce ranking and reviews it under the lead objective. Another keyword belongs to Replacement Pumps and has meaningful cost without primary purchases. Its visible search terms include requests for a repair manual alongside product-specific queries. The client confirms that manuals are outside the offer, while product queries remain relevant. The evidence supports a focused intent review rather than removing the whole positive keyword immediately. The report names the observed searches, preserves their parent scope and distinguishes their visible spending from the keyword's complete aggregate.
A third candidate has a plausible buying phrase but sends users to an old category page. The website owner verifies that the desired equipment is difficult to find there. Mira does not describe a pause as a destination repair. She asks for draft page and ad ideas conversationally, then separates that creative work from the live edit proposal. The client chooses an exact negative for the irrelevant manual query and a temporary pause for one specifically identified keyword while the page is repaired. The proposal displays each effect with its campaign and ad group. Creating broader campaign exclusions or changing unrelated budgets would exceed what the client actually chose.
After confirmation, the assistant reports execution and the available verification for the selected settings. Mira records the website task separately and schedules a mature follow-up using the same purchase definition. She compares the remaining keyword branch and campaign outcomes without claiming the earlier reviewed cost was guaranteed savings. If the paused keyword is later re-enabled, that status change receives its own precise review. The case illustrates the agency's real value: translating a vague spending complaint into correctly scoped evidence, preserving useful targeting, involving the right owner and approving only the intended changes. None of those steps requires pretending that the original ranking alone established the cause of every unsuccessful visit.
Investigate efficiently in Slack
Use a conversational request that carries the important scope. For example: “For the last thirty completed days, show enabled Search keywords with cost above our chosen review amount and no primary purchases. Include campaign and ad group, and rank by cost.” The amount should be supplied in the account currency rather than invented by the assistant. If the client has not agreed a threshold, ask for the largest candidates and make that ranking explicit. The reporting framework discovers available fields and settings, reads supported data and retains a reusable result for follow-up questions. A straightforward keyword question should not trigger unrelated Merchant scans, broad account audits or redundant reads of the same verified reporting slice.
A useful follow-up narrows the unresolved branch: “Show the reported search terms for the largest candidate in Replacement Pumps.” The assistant should resolve the reference from the verified result and preserve its parents. If the current capability or provider coverage does not support the requested breakdown, explain that specific limitation. Do not fabricate the missing query detail or route the client into a different metric while claiming success. The available data library and shared condition logic support many combinations, but field existence alone does not guarantee every resource-and-segment combination is valid. Local compatibility checks can avoid unnecessary API calls for combinations that cannot work.
Creative help belongs naturally in the same conversation. “Suggest a clearer headline for this product,” “brainstorm alternative buying phrases,” and “draft a client explanation” can be answered from established context without consulting Keyword Planner automatically. If the client requests Google keyword history or measured ideas, the relevant research tool is appropriate and the account/location context should be established. Keep AI suggestions separate from observed spending evidence. A suggested keyword has no proven account performance merely because it sounds relevant. Similarly, a drafted explanation should not invent customer results. The conversation can move between interpretation, creativity and measurement while maintaining an honest boundary around each.
Reduce tool use by reusing the verified dataset for sorting, grouping, counts and requested visuals when its scope already answers the follow-up. Read again when the client changes dates, requests a new surface or needs current settings for an edit. That is a more useful efficiency rule than avoiding all follow-up calls indiscriminately. Provider failures, stale context and incomplete selections still need explicit handling. The client should receive a concise explanation of the unresolved part and a concrete next step. The agency can consult the account-question workflow for the general pattern, then apply it to this specific keyword investigation.
Present findings without overstating savings
Lead with the answer the client needs: how many verified keyword occurrences meet the stated review condition and how much historical cost those occurrences represent. Follow with a short table of the most useful examples and the action distinction. A name alone is insufficient when the same text exists in several groups; include the ad group and campaign in readable columns. Label the selected outcome and dates once above the table. Avoid raw resource names, internal field keys and long explanations of implementation details. The client should understand the scope and next decision quickly, while the fuller evidence remains available for a deeper agency review.
A requested chart should clarify a real question. A time series can show whether the candidate set's cost rose recently, while a ranked bar view can show spending contribution. Use the selected dataset and suitable grain instead of generating separate charts for each portion of the period without explanation. Native chart limits may require weekly or monthly aggregation for long periods; explicit daily detail can require more than one native chart with the limit stated. Keep cost, conversions and ratios labeled as different measures rather than placing incompatible units under one unexplained scale. A chart's purpose is to reveal the pattern, not to make a weak conclusion seem more authoritative through decoration.
Offer a fuller supported output when the matching population would overwhelm Slack, preserving the same selected metrics and definitions. The larger output should not add irrelevant monthly repetitions or technical columns simply because they exist in the source. A keyword performance report and a Keyword Planner history report answer different questions and must not be confused. Planner volumes do not prove that the client's own keyword generated profitable traffic. For account performance, retain actual cost and selected outcomes; for research, label the planning data appropriately. The agency can summarize a few candidates in the conversation while ensuring the downloadable or web evidence does not quietly change the metric definition.
Close the result with the most proportionate next step. That may be an intent check, a named conversion review, a destination task or a precise proposal. Avoid turning every table into an immediate “pause all” invitation. When a client deliberately requests a pause, the proposal should still explain the exact selected occurrences and their parents. When no rows match, say so within the verified scope rather than suggesting that the whole account is free of waste. When the read failed or the population is incomplete, retain that distinction. Clear presentation is an accuracy requirement: it prevents a small preview, a missing source or an undefined ratio from becoming a misleading account-wide verdict.
Choose the precise supported action
Choose between changing a positive keyword's status and adding a negative at the appropriate scope. These are distinct effects. A pause stops that keyword occurrence from being enabled; an exact negative excludes the specified matching search within the supported parent scope. The dedicated keyword status action resolves the actual resource, while the negative adapter prepares the specified texts and match behavior. Explain the choice using the investigation's evidence. If the issue is a whole targeting occurrence under an agreed exposure rule, a pause may fit. If the issue is a particular irrelevant search within otherwise useful traffic, a focused negative may preserve the valuable branch.
The negative additions support campaign, ad group and shared-list targets through their respective adapters. Existing matching criteria are checked so the proposal can add missing terms rather than blindly creating duplicates. The preparation also defines a read-back check for the desired membership. A shared-list change does not itself attach the list to a campaign; the attachment is a separate relationship action with its own target. Campaign type and API capability must be respected. Do not generalize one supported Search workflow into a promise that every inventory type accepts the same target and negative operation. Establish the exact setting and relationship before presenting it as executable.
If the client decides to replace weak targeting with new positive keywords, the dedicated positive creation action requires a standard Search ad group. It accepts the requested supported match type and checks existing terms. Creation, status and destination work remain distinct. New keywords added to an existing group do not automatically enable a paused group or repair its ad. A planned new parent's dependencies can be resolved in the action sequence, but the confirmation must show the resulting states clearly. The negative-keyword guide and empty-group guide explain the neighboring workflows without making them implicit consequences of this keyword review.
Use the local capability and dependency framework to plan supported effects before making unnecessary provider requests. A keyword needs its actual ad group and campaign context; an account-level unrelated action should not trigger keyword discovery. Existing targets still require the relevant read and authorization checks. Unsupported settings receive a clear boundary, and overly broad requests should be split into understandable smaller requests without silently losing requested effects. The agency can discuss a broader strategy conversationally while preparing only the specific actions the owner chose. That separation keeps the tool useful for complex work without allowing a vague instruction such as “fix the waste” to become an uncontrolled collection of account mutations.
Review the target and every effect
The proposal is the client's last opportunity to catch a wrong target before a live change. Give each logical effect its own clear step. For a positive keyword, show the keyword text and match type, its ad group, its campaign and the requested status change. For a negative, show the exact text, negative match type and parent scope. For a shared list, show the list and any separately requested campaign attachment. Avoid arrows that make the child look like its own parent, numeric identifiers in explanatory labels or a technical operation dump. The person should be able to read the intended account effect without understanding the provider's resource syntax.
Use the confirmation notice: “No changes have been made yet. This proposal expires in 15 minutes.” The Confirm and Execute button authorizes the displayed proposal through the connected-owner flow. Preparing a proposal is not a live write, and drafting a strategic recommendation does not authorize a change. If the client modifies the targets or parameters, prepare the revised effect rather than executing an earlier proposal under a new conversational assumption. A multi-effect request needs all intended steps displayed together in their valid order. If the request must be divided, say so clearly and preserve what remains to be requested.
On confirmation, the app should make progress visible promptly and disable or replace the actionable state so the click does not look ignored. The live execution still checks authorization and the relevant current settings, performs supported validation and sends the prepared request only within that approved flow. This protects against a stale proposal when another agency member or the client changed the account meanwhile. These checks do not promise that every future auction will behave as expected. They establish that the requested configuration change remains appropriate to execute against the resolved target. The agency can read the approval workflow for the general safeguards.
Separate execution evidence from business success. A verified pause establishes the status of the selected keyword; a verified negative establishes the desired criterion or relationship. It does not establish that acquisition efficiency improved, all irrelevant searches disappeared or a future spending amount was saved. If the write was sent but complete verification failed, explain that state and avoid blindly retrying the mutation. The client may need a focused account review before another change. Clear statuses help the agency distinguish a successful configuration edit from an uncertain outcome without treating uncertainty as proof that nothing happened.
Verify the change and review the next period
After execution, review the exact target and each effect in the result. Confirm that the desired keyword status or negative membership was read back where supported, and that any requested list attachment was verified separately. A partial result must not be described as complete because the first step worked. Preserve the readable parent hierarchy so the final message can be compared with the proposal. If a change was unnecessary because the desired state already existed, make that distinction rather than implying a new mutation occurred. The agency's record should show the approved intent and the verified configuration, with outstanding external tasks listed separately.
Choose a follow-up window that fits the business and outcome maturity. Review remaining cost, selected outcomes and commercial quality under the same definition used for the original decision. If the metric changes, explain why and avoid a misleading before-and-after claim. A keyword pause can change the distribution of traffic to other eligible targeting; a negative can reduce one intent branch without guaranteeing an account-wide spending reduction. Look at the relevant campaign and group totals as well as the selected keyword occurrence. This wider follow-up helps determine whether the intervention addressed the intended problem or merely moved the visible cost elsewhere.
Keep reversible operational choices and durable commercial learning distinct. The owner may later re-enable a paused keyword after repairing a destination, or retain an exact negative because the excluded service remains outside the offer. Each later live change requires its own accurate proposal and approval. Record why the original action was chosen so the next reviewer does not interpret a pause as proof that the keyword can never be useful. The same applies to exclusions: their meaning comes from the offer and scope, not an agency convention that every low-performing query should remain blocked indefinitely. Periodic review keeps the account aligned with the client's current business.
Maintain an investigation routine that begins with cost and selected outcomes, then expands only where evidence requires it. Reuse verified reporting results, read the relevant deeper surface, involve the website or measurement owner when necessary and prepare exact supported changes. This approach scales across clients because the underlying tools, conditions and relationships remain reusable while the commercial judgment stays specific to each account. The final answer to “which keywords waste money?” becomes a defensible set of candidates and decisions rather than a dramatic label pasted onto a ranking. The client sees what happened, why it matters, what was approved and what the next review will measure.