Google does not list a budget change as a trigger for the Learning status, so when results fall after scaling, look at what the extra budget bought and at any strategy or structure change made at the same time. The ranked causes are a budget jump into costlier auctions, a bid strategy change that restarted Learning, new campaigns that split the conversion data, too few conversions for Smart Bidding, and saturation on your core keywords.
The four questions and five causes below are the same playbook the free campaign troubleshooter runs. Here each cause also names the PaidSync tool that checks it in your own account and the change you can preview before anything is applied.
Answer yes or no to each. Every cause below says which answer points to it.
Points here when you answer Yes to question 1.
Google does not list a budget change among the triggers for the Learning status, which are a new strategy, a settings change and a composition change. A large budget increase still changes what the bidder buys: the extra money goes to auctions it used to skip, which tend to cost more for each conversion. Step back to the last budget that performed, then raise it in stages and read results after at least one conversion cycle at each step.
get_change_history, get_campaign_performance, get_budget_pacing
Ask your assistant: “When did the budget change, and how did spend, conversions and cost per conversion move in the two weeks either side?”
update_campaign_budget defaults to dry_run and returns the daily budget it would set before it is applied. Budget updates also need confirm_destructive, and a change of more than 50 percent needs confirm_large_change.
Points here when you answer Yes to question 2.
Google puts a bid strategy into Learning after a new strategy or a settings change, and says calibration can take up to around 50 conversion events or 3 conversion cycles. Changing only the target CPA or target ROAS does not trigger Learning. If you switched strategies while scaling and the old one performed, switch back and hold it steady.
get_change_history, get_campaign_details
Ask your assistant: “Did the bid strategy or its settings change in the last 14 days, and what was the strategy before?”
update_campaign defaults to dry_run and returns the switch back to the prior bid strategy before it is applied.
Points here when you answer Yes to question 3.
Each new campaign starts with less conversion history, and adding campaigns to or removing them from a shared bid strategy is a composition change that restarts Learning. If you split one winning campaign into several, each now learns from a fraction of the data. Consolidate until each campaign records conversions steadily.
list_campaigns, get_campaign_performance, list_bidding_strategies
Ask your assistant: “List campaigns created in the last 30 days with their conversions, next to the campaign they were split from.”
pause_campaign defaults to dry_run and names the campaign it would pause. The live pause also needs confirm_destructive.
Points here when you answer No to question 4.
Google says calibration can take up to around 50 conversion events or 3 conversion cycles, so a campaign with only a handful of conversions a month takes a long time to get there. Consolidate low-volume campaigns, or run them on Manual CPC and keep Smart Bidding for the campaigns with steady volume.
get_campaign_performance, list_campaigns
Ask your assistant: “How many conversions did each Smart Bidding campaign record in each of the last 4 weeks?”
update_campaign with bidding_strategy set to MANUAL_CPC defaults to dry_run and shows the change for each low-volume campaign before it is applied.
Points here when the causes above are ruled out, with answers No to question 1, No to question 2, No to question 3, Yes to question 4.
More spend on the same keywords buys the searches you were skipping, and those tend to convert less. Check Search impression share on your core keywords. If you already win most of the eligible impressions, extra budget goes to weaker queries. Broaden to new keyword themes that match your buyer, such as problem queries and alternative-to queries, rather than pushing more money into the same head terms.
get_impression_share_analysis, get_search_terms_report, research_keywords
Ask your assistant: “What is Search impression share on my core keywords, and which new keyword themes have search volume in my market?”
add_keywords defaults to dry_run and lists the keywords it would add for the new themes before anything changes.
In the PaidSync workspace, by default, every write is approval-gated: it is held as a card before it runs, and an applied card gets a one-tap undo. On every path, Google Ads creates and most other creates are previewed with dry_run, destructive operations need confirm_destructive, and applied changes are logged, on a best-effort basis, with a 7-day before-and-after outcome. When your own assistant calls PaidSync over MCP, no card is written, so whether a person confirms first is set in your assistant.
The full model, including what is not built yet, is on the security page.
Run this check in Claude, Claude Code, Cursor or ChatGPT. Connect Google Ads once, then paste any of the questions above to your assistant.
Not sure which symptom fits? The free Google Ads audit tool scores wasted spend, structure, Quality Score and Performance Max, and the campaign troubleshooter runs its playbooks without an account.
Google lists four things that put a bid strategy into Learning: a new or reactivated strategy, a change to its settings, a composition change such as adding or removing campaigns, ad groups or keywords, and an ad group target change on Shopping. A budget change is not on the list. A large increase still changes which auctions the bidder buys.
Google does not publish a safe percentage. Raise it in stages you can compare, and read results after at least one conversion cycle at each step, which is the minimum window Google suggests before evaluating bid strategy performance.
No. Google says changing a target won't trigger a Learning status and won't reset what Smart Bidding has learned. Switching to a different bid strategy, or changing its settings, does.
Each new campaign starts with less conversion history, and adding campaigns to or removing them from a shared bid strategy is a composition change that restarts Learning. Several small campaigns each learn from a fraction of the data one campaign had.
In the PaidSync workspace, by default, every write is approval-gated: it is held as a card before it runs, and an applied card gets a one-tap undo. On every path, Google Ads creates and most other creates are previewed with dry_run, destructive operations need confirm_destructive, and applied changes are logged, on a best-effort basis, with a 7-day before-and-after outcome. When your own assistant calls PaidSync over MCP, no card is written, so whether a person confirms first is set in your assistant.