Google Ads Interview Questions & Answers for Experienced Marketers

By data maturity and goal. Max Conversions/Max Conversion Value for volume while learning or without a firm target; Target CPA when lead economics are known; Target ROAS when conversion values vary and you have roughly 30+ conversions a month with accurate values. Targets should come from unit economics — break-even ROAS is 1 ÷ margin — not from aspiration.

Start from break-even (1 ÷ gross margin), add the profit cushion you need, then set the initial target near recent actual ROAS and tighten gradually. Setting it far above what the account currently achieves collapses traffic, because the system only enters auctions it predicts will hit the target.

Mostly no — Smart Bidding already weighs device, location, time and audience per auction, so manual percentage adjustments are ignored. The exception is a device adjustment of −100%, which still excludes that device entirely. Schedules still control when ads run at all.

Advanced Smart Bidding tools: a seasonality adjustment tells the system about a short, sharp conversion-rate change (like a flash sale) so it doesn’t mislearn; a data exclusion tells it to ignore a period of corrupted data (like a tracking outage). Both protect the model from learning the wrong lesson.

When several campaigns share one economic goal — a shared Target ROAS across product lines, for instance — so the system can balance spend across them, optionally with min/max bid limits. It trades campaign-level control for portfolio-level efficiency.

Only two remain: data-driven (the default) and last click — the rules-based models were retired. Data-driven is the right default because it credits assisting interactions; last click survives mainly as a comparison lens. The deeper point: attribution shapes what Smart Bidding optimises toward.

First-party data (like hashed emails) sent with conversions to recover matches that cookie loss breaks. They restore measurement accuracy, which directly improves Smart Bidding — and with offline conversion import, you can optimise toward closed deals and revenue rather than form fills.

Different attribution scopes and models (Ads credits its own last touchpoints; GA4 sees all channels), different counting (conversion time vs event time), and consent/modelling differences. The fix is choosing one source of truth per decision and, crucially, not importing GA4 conversions alongside native tags for the same action — that double-counts.

With experiments: campaign experiments or drafts for tactics, geo holdouts for channels, and conversion-lift tests where available. Auction wins and ROAS don’t prove causation — an experiment with a control group does. I keep brand and non-brand separated for exactly this reason.

Consent signals gate which data can be used; where consent is denied, modelling fills gaps. With ad_storage becoming the controlling signal for ads features, misconfigured consent silently shrinks audiences and conversions — so consent setup is now part of measurement QA, not a legal afterthought.

Asset groups per theme or product line, audience signals as strong hints (they guide, not restrict), and the newer controls: campaign-level negative keywords (up to 10,000), brand exclusions, and channel-level reporting to see where spend goes. Feed quality still does most of the work for retail.

Separate brand first — brand exclusions stop PMax taking credit for demand you already own. Then look at incremental conversions versus the campaigns it cannibalises, channel distribution, and search-term insights. If PMax ‘wins’ only by absorbing brand and remarketing, it isn’t adding value.

A feature suite on Search campaigns — search-term matching beyond your keywords, AI text customization, final-URL expansion — not a separate campaign type. I’d test it on one strong campaign (30+ conversions/month) via a one-click experiment for 4–6 weeks, with Smart Bidding, tight negatives and text guidelines, before wider rollout — especially since DSA and campaign-level broad match auto-upgrade into it from September 2026.

As demand creation, not capture: view-through and assisted conversions, brand lift, and Attributed Branded Searches — under data-driven attribution. Judging it on last-click ROAS systematically undervalues it, because its impact shows up later through Search and PMax.

Brand converts cheaply because those people already chose you; blended reporting lets that mask weak non-brand performance. Separation gives honest ROAS per segment, distinct budgets, and the ability to test brand incrementality — pause brand and watch whether organic recovers those conversions.

Auction Insights plus the lost-IS columns: lost to budget means the campaign is capped — fix budget or efficiency; lost to rank means bids or quality — fix Quality Score components or bidding. Also check whether a competitor entered aggressively. The split tells you which lever to pull.

Raise budgets gradually (learning resets punish big jumps), loosen targets stepwise rather than at once, expand keywords and audiences from proven search-term and asset data, and validate each step against the incrementality question. Scaling is a sequence of controlled experiments, not a budget switch.

First tracking — tag firing, conversion counting changes, GA4 import duplication. Then change history — bids, budgets, targets, auto-applied recommendations. Then the auction — Auction Insights for a new competitor, search-term mix shifts. Then landing pages — breakage, speed, out-of-stock. Nine times out of ten it’s tracking or an unnoticed change, not the market.

An MCC for access and cross-account tools, Google Ads Editor for bulk changes, scripts for monitoring (budget pacing, anomaly alerts, stock-based pausing), and Looker Studio dashboards per client so reporting is live rather than manual. Automation handles the repetitive layer; human review handles judgment.

Import offline outcomes — qualified, opportunity, closed — back into Google Ads via enhanced conversions for leads, assign values by stage, and bid to value (tROAS) instead of raw CPA. Add qualifying friction where needed. Otherwise Smart Bidding happily fills your CRM with cheap junk.