When you launch a Smart Bidding campaign or make a big change to one, Google shows the bid strategy status Learning. During that time it tries different bids to work out which searches convert, so results swing more than usual. It usually lasts about a week, longer for campaigns with few conversions. The best thing to do is to leave bids, budgets and targets alone until it ends.
What restarts learning
| Change | Restarts learning? |
|---|---|
| Switching bid strategy (for example to target CPA) | Yes |
| Changing the target CPA or target ROAS a lot | Yes |
| Large budget changes | Often |
| Changing conversion actions or what counts as a conversion | Yes |
| Adding negative keywords | No, and you should keep doing it |
| Pausing a clearly bad keyword or ad | Usually small effect |
What to do during learning
- Read the search terms report and block searches that clearly don't fit.
- Check clicks by country and device for surprises.
- Check every landing page loads and the conversion tag fires.
- Wait: judge the campaign on at least 7 days of data, not on day one.
Distribute's checks after launch run on days 1, 3, 7 and 14. Until day 7 they say to leave bids and budgets alone, apart from blocking bad searches. See Google Ads conversion tracking.
Distribute researches your site, plans campaigns, writes and grades every ad, and manages budgets and rules across Google, Meta and TikTok. Nothing goes live or changes without your approval. 5% of the ad spend it manages. You pay the platforms directly from your own account.
Questions people ask
How long is the Google Ads learning phase?
Usually about a week. Campaigns with few conversions can take longer, because Google needs enough conversion data to learn from.
Can I add negative keywords during learning?
Yes. Blocking searches that clearly don't fit doesn't restart learning and saves money.
Why is my campaign stuck in learning?
Usually too few conversions. Make sure the conversion tag fires, and consider a conversion that happens more often until the campaign has enough data.