Skip to content
whitead

Smart Bidding Learning Phase

glossary google ads updated 2026.05.06 7 min read
// contents · 9 sections
  1. What the learning phase is
  2. How thresholds and resets work
  3. What changed in 2025-26
  4. Common misconceptions
  5. Google says learning takes 7 days. Why do my campaigns swing for 3 weeks?
  6. Will pausing a campaign for the weekend reset learning?
  7. Can I shorten learning by importing historical conversions?
  8. Related concepts
  9. Sources

The Smart Bidding learning phase is the window after a launch or material change when Google's automated bidding has not yet collected enough conversion data to optimise reliably. Decisions made during this window — pausing, slashing budgets, swapping strategies — are the single most common cause of Smart Bidding "failure" in audits.

What the learning phase is

Smart Bidding refers to Google Ads' conversion-aware automated bid strategies — Target CPA (tCPA), Target ROAS (tROAS), Maximize Conversions, and Maximize Conversion Value. Each strategy uses a machine-learning model that predicts conversion probability (and, for value strategies, conversion value) for every auction, then sets a bid based on your target. The ML model needs recent conversion signals scoped to your account, campaign type, audience, geo, device, and time-of-day patterns. When those signals are missing or stale, Google enters a "learning" status visible on the campaign's bid strategy report (Google Ads Help: About Smart Bidding, 2025-09-12).

The status appears in Campaigns → Bid strategies column → values are Learning, Eligible, Limited, or Misconfigured. Performance during Learning is expected to be more volatile than steady-state — both up and down.

How thresholds and resets work

Google's published data thresholds for steady-state performance:

Strategy Minimum recent conversions Source
tCPA / Maximize Conversions ≥30 conv in trailing 30 days per campaign Google Ads Help, 2025-09-12
tROAS / Maximize Conversion Value ≥50 conv in trailing 30 days per campaign Google Ads Help, 2025-09-12
PMax with value Google's published Smart Bidding tROAS guidance for Search references ≥50 conv/30d; PMax with value bidding mirrors this in operator practice but Google does not publish a separate threshold Google Ads Help: About PMax, 2025-10-21

Below those thresholds, the strategy still runs but with wider exploration and less stable cost-per-acquisition. Above them, the model has enough signal to keep prediction error low.

Resets trigger a new learning phase. Confirmed reset triggers per Google's bid strategy reference:

  1. Switching bid strategy type (e.g., Maximize Conversions → tCPA)
  2. Changing the target by material amounts (tCPA target $50 → $65) — Google does not publish a precise threshold; operator consensus is in the 20-30% range
  3. Material budget changes on shared-budget campaigns (Google does not publish a precise threshold; operator consensus is in the 20-30% range)
  4. Modifying or removing the conversion action the strategy optimises against
  5. Adding or removing major audience segments or location targets
  6. Restructuring ad groups inside the campaign

Edits that do not reset learning: creative additions, negative keyword additions, small bid adjustments on segments, label changes.

What changed in 2025-26

Three shifts are worth knowing.

First, Consent Mode v2 enforcement (March 2024 in EEA, expanded globally through 2025) means a portion of conversions now arrive as modeled rather than observed. Smart Bidding consumes both, but accounts that did not deploy Consent Mode v2 saw measured conversion volume drop 10-25% in EEA traffic (PPC Land, 2025-06-18) — pushing some campaigns below the 30/50 conversion threshold and back into prolonged learning.

Second, Smart Bidding Exploration rolled out in 2024-25 as an explicit setting on Maximize Conversion Value campaigns. It allows the model to deliberately explore lower-confidence auctions, lengthening the perceived volatility window but improving long-tail discovery (Google Ads & Commerce blog, 2024-05-09).

Third, the AI Max for Search opt-in (GA May 2025) reuses the Smart Bidding model but adds keyword/URL/asset expansion — re-triggering learning on opt-in even though no bid strategy changed.

Methodology note. Whitead's audit detector for premature Smart Bidding intervention compares the campaign's bid-strategy modification timestamps against its 14-day forward performance window. We flag a campaign as "interrupted learning" when (a) the bid strategy or target was last modified within the trailing 14 days, AND (b) a second modification happened during that window, AND (c) CPA/ROAS volatility is within the model's expected exploration range (±25% from trailing 60-day mean). The detector does not flag a single change after a long stable period — only repeated mid-flight edits. We also surface a "below-threshold" warning when trailing-30d conversion count for the campaign is <30 (tCPA) or <50 (tROAS), regardless of edit history. The fix-priority logic ranks "do nothing for 14 days" above any creative/structural change, because the second-order cost of restarting the model exceeds most first-order issues we observe at audit time.

Common misconceptions

Google says learning takes 7 days. Why do my campaigns swing for 3 weeks?

Google's "typically up to 7 days" estimate (Google Ads Help, 2025-09-12) describes the status badge, not stabilised performance. Third-party measurement frequently shows materially longer convergence:

  • Optmyzr's 2025 study of 1,800+ accounts found median tCPA stabilisation at 14 days post-change; the 75th percentile took 21 days (Optmyzr, 2025-08-14).
  • Search Engine Land's 2026 review of B2B lead-gen accounts (where conversion volume is often near the 30/30 floor) found CPA still drifting at day 21 in 40% of cases studied (Search Engine Land, 2026-01-22).

Both observations are consistent with Google's published model — the badge clears as soon as the model has enough samples to predict; stabilised outcome requires those predictions to also be calibrated against your specific funnel. The practical guidance: trust Google's badge to know when learning is technically over, but use the 14-day floor before judging performance, and 21 days for low-volume accounts.

Will pausing a campaign for the weekend reset learning?

No, pausing does not reset. The model state is preserved; on resume the strategy continues with the same accumulated signal. What does erode is the 30-day rolling conversion window — extended pauses (≥7 days) shrink the conversion count toward the threshold and may push a previously-eligible campaign back into Limited status when it resumes (Google Ads Help: About Smart Bidding, 2025-09-12).

Can I shorten learning by importing historical conversions?

Partially. Offline conversion import (OCI) backfills via GCLID/GBRAID/WBRAID up to 90 days; those conversions count toward the 30-day rolling threshold once attributed. This can pull a campaign across the 30/50 floor faster than waiting for new conversions, but it does not bypass the calibration phase — the model still needs recent signal to predict current auction dynamics (Google Ads Help: About OCI, 2025-07-30).

The learning phase only matters if your conversion signal is clean — see Conversion Tracking for the foundation, Enhanced Conversions for Leads for B2B/lead-gen accounts pushing past the 30/30 threshold, and Smart Bidding for the parent strategy family. For attribution shifts that interact with learning resets, see Last Click → DDA Migration. When conversion volume is the bottleneck, Search Budget Limited often explains why a campaign cannot reach the threshold.

Sources

← Back to Wiki