Google Ads Bidding Strategies: Maximize Conversions & Target CPA Mastery

30 conversions. That is the magic number Google wants before Target CPA bidding stops guessing. Maximize Conversions has no hard floor, but the algorithm still needs volume to find its footing. Every bidding guide repeats those numbers like scripture.

Nobody asks the question that actually decides whether the strategy works: were those 30 conversions real?

I have managed accounts where Smart Bidding hit every milestone on paper. Conversion volume climbed. Target CPA held steady.

The dashboard looked like a win. Then I cross-checked the conversion records against a clean session log and found that somewhere between 10 and 40 percent of those "conversions" were bot form fills, click-fraud landing-page hits, and recycled spam emails. The algorithm did not learn to find customers. It learned to find whatever was cheap and abundant. In 2026, what is cheap and abundant is bots.

This is not a bidding-strategy post. It is a post about what your bidding strategy is being fed. DataCops exists because the fix is architectural - you filter the conversion signal at the source, before it ever reaches Google, instead of hoping Google's click filter catches it after the fact.

See the Google Conversion API, fraud traffic validation, and Google Ads click fraud.

Quick stuff people keep asking

What is the difference between Maximize Conversions and Target CPA? Maximize Conversions spends your full budget chasing the most conversions it can get, with no cost ceiling. Target CPA spends toward a cost-per-acquisition you set, throttling volume to protect that number. Maximize Conversions is volume-first.

Target CPA is efficiency-first. Same underlying machine learning, different objective.

How many conversions do I need for Target CPA bidding? Google's working guidance is 30 conversions in the last 30 days, per campaign. Below that, the algorithm has too few examples to model who converts. But 30 is a floor for quantity, not quality.

Thirty clean conversions and 30 bot conversions look identical to the requirement and produce wildly different results.

When should I switch from Maximize Conversions to Target CPA? When you have stable conversion volume above 30 a month and you actually know your real CPA target. Run Maximize Conversions first to build data and discover your natural cost-per-conversion, then move to Target CPA to enforce it. Switching too early just hands the algorithm a target it cannot hit.

Why is my Google Ads Smart Bidding not working? Three usual suspects. Budget too low to escape the learning phase. Conversion volume too thin.

Or - the one nobody checks - the conversion data is contaminated, so the algorithm is optimizing toward traffic that never buys. The first two show up in the interface. The third does not.

Does bot traffic affect Google Ads Smart Bidding? Yes, and worse than people think. Smart Bidding is a prediction engine trained on your conversion history. If bots are completing your forms, the algorithm treats bot-like signals - certain IP ranges, device patterns, times of day - as conversion predictors.

It then bids up to find more of that traffic. It does exactly what you asked. You asked the wrong thing.

How long does the Smart Bidding learning phase last? Usually 1 to 2 weeks, sometimes up to 4 if conversion volume is low or you made a big change. The learning phase is the algorithm building its model. If the data going in is dirty, a longer learning phase does not help - it just builds a more confident wrong model.

Should I use Target CPA or Target ROAS for ecommerce? Target ROAS if you have reliable revenue values per conversion and order values vary a lot. Target CPA if your order values are fairly uniform or your revenue tracking is shaky. For lead gen, Target CPA almost always - there is no purchase value to optimize against.

Either way, the rule underneath is the same: the strategy is only as good as the conversion data feeding it.

How does conversion data quality affect bidding performance? It is the whole game. Smart Bidding does not optimize toward your goals. It optimizes toward your recorded conversions.

If those records are 30 percent garbage, the algorithm spends 30 percent of its intelligence getting better at buying garbage.

The conversion record Google trusts is not the conversion record you think it is

Here is the number that should bother you. Industry estimates put invalid traffic on paid search somewhere between 9 and 40 percent depending on vertical, geography, and how aggressively you are targeted. Google's own automated click filtering is good at the obvious stuff - repeated clicks from one IP, known data-center ranges, simple scripts.

On sophisticated invalid traffic, residential-proxy bots and human click farms, independent testing suggests it catches only a fraction, often quoted in the 5 to 15 percent range.

So picture the pipeline. A bot or a click-farm worker hits your ad. Google's filter does not catch it.

The click is billed. The visitor lands on your page and - because modern bots are built to look human - fills out your lead form or triggers your "conversion" event. Google Ads records a conversion.

Smart Bidding files it as a training example: this kind of visitor converts.

Multiply that by weeks. The algorithm now has a model where bot-shaped traffic is a positive signal. It starts bidding more aggressively for the placements, times, and audiences where that traffic lives, because that is where conversions are "cheap." Your Target CPA drops.

Your conversion count rises. Your reporting glows.

And your actual revenue does not move. Because none of those conversions were people.

This is the Layer 5 failure, and it is the cruelest one because it is invisible by design. Layers 3 and 4 - scripts blocked by ad blockers, bots padding your analytics - at least leave a gap you might notice. Layer 5 does the opposite.

It does not leave a gap. It fills your dashboard with green. The contaminated conversion data trains the bidding model, the model gets confident, and confidence on bad data is worse than uncertainty on good data.

Garbage in, garbage optimized, garbage out.

I watched a B2B lead-gen account live this exact story. Target CPA was set at $45. Within six weeks the reported CPA was $31 and lead volume was up 40 percent.

The client was thrilled. The sales team was not - connect rates had collapsed, and reps were burning hours on phone numbers that rang nowhere and emails that bounced. We pulled the lead list and fingerprinted it.

A large block of the "new" leads traced back to a handful of device fingerprints and a tight cluster of IPs. The algorithm had found a cheap, repeatable source of form fills and optimized straight into it. The $31 CPA was real.

The leads were not.

If you want to see how brazen this gets, look at what a fraud honeypot turns up. The team at PillarlabAI ran a deliberate signup trap and pulled 3,000 signups. When they fingerprinted the cohort, 77 percent were fraudulent - and 650 of those accounts traced back to a single device fingerprint.

One device. Six hundred and fifty identities. If that device had been clicking a Google ad and completing a conversion event each time, Smart Bidding would have seen 650 conversions and concluded that whatever that traffic looked like was the most valuable audience you had.

The 30-conversion requirement assumes every conversion is a signal. In 2026 a meaningful share of them are noise wearing a signal's clothes.

Why "just use Google's conversion filtering" does not fix this

The instinct is to assume Google handles it. Google has invalid-click detection, you get the occasional credit, surely the conversion data is clean. It is not, for a structural reason.

Google filters at the click layer, on its own infrastructure, using signals it can see. It does not have full visibility into what happens on your site after the click - the session behavior, the form-fill timing, the device consistency across your funnel, the IP reputation correlated to your own first-party history. It cannot, because that data lives on your domain, not Google's.

So the only place the conversion signal can actually be cleaned is before it leaves your infrastructure. That is the architectural point. A first-party setup that runs on your own subdomain sees the full session, scores the visitor against an IP intelligence database, and decides - at ingestion, before the conversion event is forwarded - whether this is a human worth training the algorithm on.

Identity intelligence at signup catches the recycled-email, single-fingerprint cluster the moment it forms.

That is the model DataCops is built on. Bot filtering happens at ingestion, against a 361.8 billion-plus IP database covering residential, data-center, VPN, proxy, and Tor. Anonymous session analytics flow unconditionally - you keep your full traffic picture.

But the conversion signal that gets sent onward to Google via CAPI is the filtered one. The algorithm trains on humans. To be straight about it: the shared-CAPI piece is still in verification, and DataCops is a newer brand than the incumbents, with SOC 2 Type II in progress.

“

The architecture is the part that matters here, and the architecture is sound.

The contrast with the default setup is the whole story. Default setup: every form fill becomes a conversion, gets sent to Google, trains the model. Filtered setup: the form fill gets scored first, and the bot one never becomes a training example.

Decision guide

You just launched and have under 30 conversions a month. Start on Maximize Conversions. Do not touch Target CPA yet. Get volume and a real CPA reading first.

You have stable volume and know your target CPA. Move to Target CPA. Set the target at or slightly above your proven Maximize Conversions CPA, not your wish number.

Ecommerce with variable order values and reliable revenue tracking. Target ROAS. It optimizes for money, not transaction count.

Lead gen of any kind. Target CPA. There is no purchase value to feed ROAS, and lead gen is exactly where bot form fills do the most damage.

Your CPA is dropping but revenue or sales-team feedback is flat. Stop celebrating. Audit the conversion records before you scale spend. You are likely training on contaminated data.

Smart Bidding stuck in the learning phase. Check budget and volume first. If both are fine, check whether your conversion source is mixing in invalid traffic - a noisy signal slows learning.

You are about to scale a winning campaign. Validate conversion quality before you raise the budget. Scaling a campaign trained on bots just buys you more bots, faster.

You are not optimizing your bidding. You are optimizing your data.

The mistake I see, over and over, is treating the bidding strategy as the lever. People A/B test Target CPA against Maximize Conversions, tweak the target by a few dollars, argue about portfolio versus campaign-level bidding - and never once question whether the conversions underneath are real.

Smart Bidding is not magic and it is not broken. It is an obedient optimizer. It will get ruthlessly good at finding more of whatever you told it was a conversion.

If 30 percent of what you told it was a conversion came from a bot, you have not bought a bidding strategy. You have bought an automated system for finding bots cheaply, and you are paying Google for the privilege.

So go pull your last 200 conversions. Not the count - the records. Look at the IPs, the device fingerprints, the form-fill timestamps, the email domains. How many of them could you actually call a customer?

If you do not know, your bidding algorithm has been answering that question for you. And it has been answering it wrong.