Google AI Mode bidding requires a fundamentally different approach to Smart Bidding targets — the auction mechanics, intent signals, and conversion paths in conversational search don't behave like standard text ad environments. If you're running campaigns that now appear in AI Mode placements, applying your existing tCPA or tROAS targets without adjustment is one of the fastest ways to overpay for low-quality traffic or starve your budget before the auction even fires. This guide walks you through exactly how to configure, calibrate, and monitor Smart Bidding for AI Mode — from initial signal setup through steady-state optimisation.

Understanding Google AI Mode Bidding and Why Standard Benchmarks Don't Transfer

AI Mode surfaces ads inside a conversational, generative search experience where users are asking multi-part, exploratory questions rather than issuing discrete keyword queries. The practical consequence for bidding is significant: a single user session can span several turns of conversation, each potentially triggering a new auction. The intent behind those auctions is richer and more contextual than a traditional keyword match, but the conversion window is also longer and less predictable.

Your existing tCPA benchmarks were almost certainly derived from a world where one query equals one auction signal. In AI Mode, the system is weighing a chain of conversational signals, page context from cited sources, and the user's apparent stage in a decision journey. Industry practitioners report that conversion rates in AI Mode placements tend to be higher per session than per individual impression — but the cost-per-impression is also elevated because the placement competes on richer contextual relevance, not just keyword match quality.

"Applying a tROAS target from a standard Shopping or Search campaign directly into an AI Mode environment is like setting a speed limit based on road signs from a different country — the units look the same, but the underlying conditions are completely different."

For a broader foundation before you configure any targets, review the full Google AI Mode ads guide, which covers placement mechanics, ad formats, and eligibility criteria. The bidding configuration you'll set here is downstream of those fundamentals — getting them wrong upstream makes target-setting an exercise in false precision.

Prerequisites: What You Need Before Touching Your Targets

Rushing into target changes without the right data infrastructure guarantees poor outcomes. Before adjusting a single bid strategy, confirm all of the following are in place.

  • Minimum conversion volume: Smart Bidding requires statistical confidence. For tCPA, most practitioners recommend at least 30–50 conversions in the past 30 days at the campaign level before tightening targets. For tROAS, the threshold is higher — aim for 50+ conversions with reliable revenue values attached.
  • Conversion value tracking: If you're targeting ROAS in AI Mode, every conversion action must have a dynamic or static value assigned. Missing values cause the algorithm to anchor on the wrong signal and produce erratic pacing.
  • Segmented campaign structure: AI Mode placements should ideally be isolated or clearly tagged so you can read their performance separately. The right AI Mode campaign structure makes this measurable rather than guesswork.
  • At least 14 days of AI Mode impression data: You need a baseline to understand your current effective CPA and ROAS before setting targets — otherwise you're setting targets in a vacuum.
  • Auto-tagging enabled and GA4 connected: Cross-channel attribution for multi-turn conversational sessions requires clean data handoffs between Google Ads and your analytics environment.

Google's official AI Mode ads rollout page confirms the product is scaling globally, including in markets like Singapore, Malaysia, and the Philippines, though market-specific performance benchmarks haven't been publicly disclosed — which makes your own account-level data the only reliable reference point for initial target-setting.

Step 1 — Audit and Segment Your Conversion Actions

The first concrete action is a thorough conversion action audit. AI Mode auctions interact with Smart Bidding differently depending on which conversion actions are marked as "primary" versus "secondary" — and many accounts have accumulated legacy actions that actively pollute the signal.

  • Identify and demote micro-conversions: Page views, scroll depth, and time-on-site events should be set to "secondary" or removed from the Smart Bidding signal pool. In AI Mode, users often browse longer before converting — rewarding those micro-signals distorts your effective CPA downward on paper while actual acquisition costs rise.
  • Assign accurate conversion windows: Conversational search users often convert hours or days after the initial AI Mode session. Extend your conversion window to at least 30 days for lead-gen and 7 days for e-commerce, then review attribution reports to validate where conversions are actually clustering.
  • Use data-driven attribution exclusively: Last-click attribution systematically undercounts AI Mode's contribution because the placement frequently functions as a discovery touchpoint rather than a closing touchpoint. Data-driven attribution spreads credit more accurately across the multi-turn journey.
  • Create a conversion action label for AI Mode: Tag your confirmed AI Mode conversion paths distinctly in your reporting. This isn't a bidding configuration — it's a measurement hygiene step that makes every subsequent optimisation decision more reliable.
  • Validate revenue values in Test mode: Before going live with tROAS targets, run a 48-hour Test mode check to confirm revenue values are firing correctly on conversion events. Corrupted values are the single most common cause of erratic ROAS performance in new bidding environments.

Step 2 — Set Initial tCPA and tROAS Targets for Conversational Auctions

Once your conversion data is clean and segmented, you're ready to set initial targets. The key principle here is to start permissive and tighten deliberately — not to start aggressive and wonder why delivery collapses.

Scenario Recommended Starting Target Rationale
No prior AI Mode data tCPA = 1.3–1.5× your current Search tCPA Allows the algorithm to explore AI Mode auction dynamics without starving delivery
14–30 days of AI Mode data tCPA = observed AI Mode CPA × 1.1–1.2 Slight buffer above observed performance to maintain volume during learning
No prior AI Mode ROAS data tROAS = 70–80% of your current Search tROAS Accounts for lower purchase intent in exploratory conversational queries
30+ days of AI Mode conversion data tROAS = observed AI Mode ROAS × 0.9 Slight discount to maintain impression share while tightening efficiency

Apply these targets at the campaign level first, not the ad group level. AI Mode auction signals operate across broader contextual patterns than ad group keyword clusters — granular target fragmentation at the ad group level typically reduces the signal pool and extends the Smart Bidding learning period unnecessarily.

Avoid making more than one significant target change per 7-day window. Each change resets or extends the learning period, and stacking adjustments produces compounding uncertainty that can take 3–4 weeks to resolve rather than the standard 1–2 weeks.

Step 3 — Layer Audience and Context Signals for AI Mode

Smart Bidding in AI Mode responds to audience signals with higher sensitivity than in standard Search, because conversational query context amplifies the value of knowing who is asking. Layering the right signals directly improves bid accuracy without requiring you to change your targets.

  • Add Customer Match lists as observation segments: Upload your existing customer lists — purchasers, lapsed customers, high-LTV segments — as observation audiences. The algorithm will learn that users resembling these profiles in AI Mode auctions convert at different rates and adjust bids accordingly.
  • Layer in-market audiences relevant to your category: These are particularly effective in AI Mode because in-market classification is itself derived from behavioural signals that overlap heavily with the intent patterns expressed in conversational search.
  • Use remarketing lists with extended membership durations: AI Mode is frequently used for research and comparison. A 90-day remarketing window captures users who may have started their journey weeks before returning to a conversational search to finalise a decision.
  • Apply device bid adjustments cautiously: Industry data suggests AI Mode skews toward desktop and tablet usage compared to standard mobile Search. Monitor your device breakdown in the first 30 days and apply modest positive adjustments to desktop if the data confirms higher conversion rates there.
  • Do not use audience targeting in "Targeting" mode initially: Stay in "Observation" mode for the first 60 days. Forcing audience targeting restricts reach precisely when you need broad signal collection to teach the algorithm what AI Mode conversion looks like for your account.

Step 4 — Configure Budget Pacing to Match AI Mode's Demand Patterns

AI Mode query volume does not follow the same intraday distribution as standard Search. Because conversational queries require more user deliberation, volume tends to cluster differently — many practitioners observe heavier concentration in mid-morning and early evening windows rather than the peak lunch-hour patterns typical of mobile Search.

  • Use Standard delivery, not Accelerated: Google retired accelerated delivery for most campaign types, but if you're managing accounts with any legacy configurations, confirm Standard delivery is active. Standard delivery is essential for Smart Bidding to manage pacing against AI Mode's irregular demand curve.
  • Set daily budgets at 2–3× your target CPA: Budget constraints that kick in before noon are the most common cause of AI Mode underperformance. The algorithm needs headroom to participate in high-value auctions that occur outside standard peak windows.
  • Avoid shared budgets across AI Mode and standard Search campaigns: Budget competition between campaign types with different auction economics produces unpredictable spend distributions. Keep AI Mode campaigns on dedicated budgets until you have 60+ days of stable performance data.
  • Review the "Budget lost impression share" metric weekly: If this metric exceeds 15% consistently, your budget ceiling is constraining Smart Bidding's ability to compete in AI Mode auctions — raise the daily budget before tightening your CPA or ROAS target.
  • Model seasonal budget uplifts separately for AI Mode: Historical seasonal patterns from standard Search campaigns are only partially applicable. AI Mode query volume during high-intent periods (product launches, peak retail seasons) may spike independently of your standard keyword volumes.

"Budget starvation and over-tight targets are two sides of the same constraint — both tell the algorithm it cannot win the auctions it needs to learn from."

Common Mistakes to Avoid

Even experienced practitioners make predictable errors when moving into AI Mode bidding. These are the ones that cause the most damage.

  • Copying tROAS targets from Shopping campaigns: Shopping and AI Mode have different average order values, intent stages, and purchase cycle lengths. A target that works perfectly for a Shopping feed will typically be too aggressive for AI Mode's exploratory query environment.
  • Reacting to week-one data: Smart Bidding learning periods in novel auction environments are real. Pulling targets downward or pausing campaigns in the first 7–10 days based on early volatility destroys the learning data you need. Commit to a minimum 14-day observation period before any adjustments.
  • Using portfolio bid strategies across mismatched campaign types: Pooling AI Mode campaigns with standard Search in a portfolio strategy dilutes the AI Mode-specific signals. The aggregate data may look stable while your AI Mode performance drifts significantly in either direction.
  • Ignoring the auction insights report: AI Mode auctions involve a different competitive set than your standard keyword auctions. Competitors who don't appear in your standard auction insights may be aggressive in conversational search placements. Check this report monthly.
  • Setting target CPA below the minimum viable bid: If your tCPA target is set below what the system needs to win a single AI Mode impression, the campaign simply won't serve. This manifests as near-zero impressions with no "Limited" status warning — it's a silent failure mode that's easy to miss.

Expected Results and Timeline

Setting realistic expectations protects you from premature intervention. Here's what a healthy AI Mode bidding ramp typically looks like.

Timeline What to Expect Action Required
Days 1–7 High CPA variability, inconsistent impression share, possible learning period status Monitor only — no bid or budget changes
Days 8–14 Delivery stabilises, CPA begins trending toward target range Confirm conversion tracking is firing correctly; check budget lost impression share
Days 15–30 Algorithm has sufficient signal to make meaningful auction decisions First opportunity to adjust targets — small increments of 10–15% only
Days 31–60 Audience signal layering begins producing measurable bid differentiation Review audience performance segments; consider adding Customer Match if not yet active
Day 60+ Stable performance baseline established; targets can be optimised toward efficiency goals Begin systematic tROAS or tCPA tightening in 10% increments every 7–10 days

Accounts that follow this staged approach consistently report reaching stable, efficient performance faster than those that apply aggressive initial targets. The counterintuitive truth is that patience in the first 30 days buys you significantly more control in months two and three.

Frequently Asked Questions

What is the minimum conversion volume needed before setting a tCPA target for Google AI Mode?

Most Smart Bidding practitioners recommend at least 30–50 conversions in the past 30 days at the campaign level before applying a tCPA target. Below this threshold, the algorithm lacks the statistical signal to make reliable bid decisions, which typically produces erratic delivery and inflated costs. For tROAS, the threshold is higher — 50 or more conversions with accurate revenue values attached is a more reliable minimum for conversational search environments.

Should I use the same tROAS target for AI Mode as I use for standard Search campaigns?

No — directly transferring a tROAS target from standard Search to AI Mode almost always produces either delivery starvation or efficiency losses. AI Mode auctions involve users at earlier, more exploratory stages of the decision journey, which typically means lower immediate conversion rates per auction. Start at 70–80% of your standard Search tROAS target and adjust upward once you have 30 or more days of AI Mode-specific conversion data.

How long does the Smart Bidding learning period last in AI Mode campaigns?

The standard Smart Bidding learning period is 7–14 days, but in AI Mode environments the effective learning window is often 14–21 days due to lower initial query volume and longer conversion windows. Each significant change to your target, budget, or campaign structure restarts or extends this period. Avoid stacking multiple changes in the same week to prevent compounding learning delays.

Can I use Maximize Conversions without a tCPA target in AI Mode?

Yes, and for accounts with limited conversion history it's often the better starting point. Maximize Conversions without a target gives Smart Bidding full flexibility to gather signal across AI Mode auctions, which builds the data foundation needed to set a meaningful tCPA target later. The risk is that uncapped spend can be high in the early phase — set a daily budget ceiling that you're comfortable spending during the learning period before switching to this strategy.

Do audience bid adjustments work differently in Google AI Mode auctions?

Audience signals carry more weight in AI Mode than in standard keyword Search because conversational query context amplifies the relevance of knowing who the user is. Layering Customer Match, in-market, and remarketing audiences in Observation mode gives Smart Bidding richer signals to differentiate bids across users with different conversion probabilities. The key is to stay in Observation mode for the first 60 days rather than forcing audience targeting, which would restrict reach during the critical signal-collection phase.