How to choose a Google Ads bidding strategy
You moved a campaign to Target CPA on a Monday. By Wednesday impressions had halved, cost per conversion was worse than before, and the strategy status said “Learning”. So you switched it back, which started the whole thing again.
Pick the strategy that matches the goal you can actually measure, then leave it alone long enough to learn. With no conversion tracking, use Maximize clicks. With tracking but low or unstable volume, use Maximize conversions and no target. Once you’re getting a steady 30 or more conversions a month, add a Target CPA. If your conversions are worth different amounts, move to Maximize conversion value, then Target ROAS once the value data is trustworthy. Target impression share is for brand defence, not performance. That ladder is the whole decision.
Most bidding problems aren’t the wrong strategy. They’re a target set before the data supported it, changed again before the strategy finished learning.
Which Google Ads bidding strategy should you use?
Google’s own guidance organises the choice by the goal you’re buying, not by how automated the strategy is. Six strategies, six goals:
| Strategy | Optimises for | Use it when |
|---|---|---|
| Maximize clicks | Clicks | You have no conversion tracking, or you’re buying traffic to find out what converts |
| Maximize conversions | Conversions | Tracking works, but volume is low or lumpy |
| Target CPA | Conversions at an average cost | Volume is steady and your CPA has stopped moving around |
| Maximize conversion value | Conversion value | Your conversions are worth materially different amounts |
| Target ROAS | Return on ad spend | Value tracking is accurate and there’s enough volume to hold a target |
| Target impression share | Where and how often you show | Brand defence. Not a performance strategy |
The ladder matters more than the list. Each rung needs the rung below it to be working. Target ROAS on top of conversion values you half-trust will optimise confidently towards the wrong thing, and it will look fine while it does it.
One honest caveat on Maximize clicks: it buys clicks, and it’s good at it. If your landing page converts badly, it will find you a lot of expensive proof.
The precondition for everything above the bottom rung is conversion tracking you’d defend in a meeting. Not “there’s a conversion action in the account”, but: it fires once per real outcome, it isn’t counting a thank-you page reload as two sales, and the actions you care about are the ones marked to include in the “Conversions” column. Four of the six strategies bid towards that number. If it’s inflated by duplicate fires or thin form-starts, the strategy will hit your target exactly and you’ll still lose money, which is the hardest version of this problem to diagnose because every dashboard says it’s working.
What changed about the names in June 2026?
Starting June 2026, Google relabelled two strategies. “Maximize conversions with a Target CPA” is now just Target CPA, and “Maximize conversion value with a Target ROAS” is now Target ROAS. Google is explicit that “the underlying bidding behavior remains exactly the same” and that no account changes are needed.
This is worth knowing mainly because of what you’ll read elsewhere. A lot of PPC writing from 2023 to 2025 treats “Maximize conversions with a target” and “Target CPA” as two different settings with different behaviour. In the current UI they’re one setting with one name. If a guide walks you through switching between them, it’s out of date.
How long is the bid strategy learning period?
Google says it “can take up to 3 weeks or 1-2 conversion cycles for the bid strategy to calibrate to the new objective”, and that it can be faster when there’s more conversion data to work with.
“One to two conversion cycles” is the more useful half of that sentence, because it’s the part that depends on your business. Your conversion cycle is how long it takes a click to turn into a conversion. If people convert the same day, a cycle is a day and calibration is quick. If your average click converts eleven days later, one to two cycles is eleven to twenty-two days, and judging the strategy on week one is judging a number that doesn’t exist yet.
Google names three things that set the duration:
- How many conversions the campaign, ad group, keyword or product gets
- How long your conversion cycles are
- Which bid strategy you picked
The practical rule is dull and it works: while the status says Learning, don’t touch the target. Google’s own advice on the Learning status is to “continue using your account as usual” but to be aware that “key metrics may vary during this time, so you may not want to measure performance until the learning period is over”.
What does “Limited” mean on a bid strategy?
Bid strategies carry a status: Inactive, Active, Learning, Limited, or one of two Misconfigured states. Four of those are self-explanatory. Limited is the one people misread, because it’s really four different problems wearing one label.
- Limited by inventory. There isn’t enough matching traffic. Widen targeting or add keywords.
- Limited by bid limits. You’ve capped max or min bids somewhere and the strategy can’t move. Loosen the caps.
- Limited by budget. Your average daily budget is below what the current settings could spend.
- Limited by bidding strategy. A manual or semi-automated strategy is holding it back.
The two Misconfigured states are worth a look before anything else, because both silently break the thing you’re optimising for. One fires when Maximize-type strategies share a budget with other strategies. The other fires when a conversion-based strategy doesn’t have conversion actions properly set up and included in the “Conversions” column. A Target CPA campaign with a misconfigured conversion action isn’t bidding badly. It’s bidding towards nothing.
“Limited by budget” is where bidding and budgeting stop being separate settings. Google defines it as your average daily budget being “lower than the recommended amount to capture all available impressions and clicks for your current settings”. Raising a Target CPA on a budget-limited campaign usually does nothing useful, because the constraint isn’t the target. If your budget maths is shaky, start with how the daily and monthly limits actually work rather than tuning bids on top of a broken budget.
What changed on 17 August 2026?
Google published a notice on the budget-limited status that’s directly relevant if you’ve touched bidding in the past fortnight:
Starting August 17, 2026, Google is updating its bidding systems to deliver more consistent and predictable performance for campaigns that are limited by budget. If your campaigns use Target CPA or Target ROAS, these changes may cause temporary performance and traffic fluctuations.
The update covers Target CPA, Target ROAS and Target CPC (Demand Gen only), across Search, Shopping, Performance Max, Demand Gen and Travel campaigns, plus Display & Video 360 and Search Ads 360. Google says it won’t automatically adjust your daily budgets or your bid targets.
So if a budget-limited Target CPA campaign got strange in the last week or two, the cause may not be anything you did. The correct response to a documented, temporary fluctuation is to wait it out. The expensive response is to read it as a targeting problem, change three settings, restart learning, and lose another three weeks working out which change did what.
When to switch, and when to leave it alone
Switching strategy is cheap to do and expensive to do often. A few rules that survive contact with real accounts:
One change at a time. Budget and target are the two levers that both restart learning. Move one, wait, then consider the other. Changing both in the same week means you’ll never know which one mattered.
Move targets in steps, not jumps. A target is a constraint on an auction, not a dial for spend. Large moves produce large fluctuations, which look like evidence and aren’t.
Judge on volume, not on days. Google suggests measuring Target CPA performance over periods containing at least 30 conversions. Seven days is not a measurement period. Thirty conversions is.
Accept the spread. Google’s wording on Target CPA is that “some conversions may cost more than your target and some may cost less”, with the average being what it aims at. A single expensive conversion is not a broken strategy.
Don’t switch to escape a status. Learning is not a problem to solve. It’s the strategy doing the thing you asked for.
Where an agent helps, and where it makes things worse
Bidding is a tempting thing to hand to an AI agent, because it looks like a loop: read yesterday’s numbers, nudge the target, repeat. That loop is precisely the failure mode above, run faster. An agent checking performance every morning and adjusting a Target CPA is changing the target inside the learning window, every day, forever. The campaign never calibrates and the reporting never means anything.
The genuinely useful agent job here is the read. Pulling google_campaigns_list and google_stats across every campaign, flagging which strategies are Learning, which are Limited and for which of the four reasons, and which conversion actions aren’t feeding the strategy they’re supposed to feed, is tedious work that an agent does well and a person does on a Friday if at all.
Writes are where you want the brakes on: a change per campaign per fortnight, a human deciding the target, and a record of what moved and when. If you can’t answer “who raised this target and on what day” from something other than memory, you can’t tell a bidding problem from a bidding change. That’s the same argument as keeping spend controls outside the prompt, applied one layer up.
Bidding is a good thing to read automatically and a bad thing to change automatically. FlyWheel gives your AI agent one MCP surface across Reddit, Google Ads, Meta, and X, with every tool call logged with args, status, latency and actor, and new campaigns shipped paused by default. Get started with FlyWheel.