Primary vs Secondary Conversions in Google Ads: Which Actions Your Bidding Actually Learns From

Here is a version of a conversation that happens a lot. An account owner opens Google Ads and sees 340 conversions last month at a cost per conversion that looks great. They check the CRM and find eleven actual enquiries. Nobody lied and nothing is technically broken. The tag fires, the column fills in, the bid strategy is working hard. It is just working hard on the wrong thing, because somewhere in the conversion settings a page view, a button click or a scroll event was marked as a conversion that bidding should learn from.

The setting that decides this is not hidden, but it is easy to leave alone. Every conversion action in your account is either primary or secondary. That one label determines whether Google’s bidding systems treat the action as the thing to buy more of, or as a number to look at on the side. Get it right and Smart Bidding spends toward what you actually care about. Get it wrong and the algorithm does its job perfectly against a target you never meant to set.

This post walks through how conversion goals, primary and secondary actions, account-default goals and custom goals fit together, where accounts typically go wrong, how conversion counting (one versus every) interacts with all of it, and how to audit and change your setup without knocking campaigns sideways. I have leaned on Google’s own help documentation for how the mechanics work, and on practitioner write-ups for the “what goes wrong in the real world” parts. Where a claim comes from a single agency blog rather than Google, I will say so.

Smart Bidding only reads one column

Start with the fact that everything else hangs off. Google’s documentation on account-default goals says that Smart Bidding, including Target CPA and Target ROAS, considers only the conversions reported in the “Conversions” column when it optimizes toward your target. Not the “All conversions” column. Not whatever you happen to be watching in Google Analytics. One column.

That makes the Conversions column the most consequential number in the account, and it also explains why two accounts with identical tracking can behave very differently. The column is not a neutral record of everything that happened. It is a filtered view, and the filter is made of your primary and secondary settings plus the goals each campaign is using.

Google’s help documentation describes the “Include in ‘Conversions'” setting in these terms: it lets you decide whether individual conversion actions appear in your Conversions and Conversion value reporting columns, and those columns feed bid strategies like Target CPA, Target ROAS and enhanced CPC, so a strategy will only optimize based on the conversions you chose to include. In the current interface, the same decision is expressed through the primary and secondary labels. Different wording, same lever.

If you have already read about why Google Ads conversion numbers often don’t add up, you will recognise that a lot of the confusion lives here. People compare the Conversions column to their backend and conclude the tracking is broken, when the real issue is that the column was never measuring what they assumed. Tracking accuracy and goal configuration are two separate problems, and they need separate checks.

Primary and secondary, in plain terms

Google’s definition, in its help page on primary and secondary conversion actions, is short. Primary actions are reported in the Conversions column and used for bidding, as long as the goal they belong to is used for bidding. Secondary actions are for observation: they show up in the “All conversions” column, and they are not used for bidding even if the goal containing them is.

The practical reading is:

• Primary means “this is a result I want the bid strategy to chase.”
• Secondary means “I want to see this happen, but do not optimize toward it.”

A good test for any given action is to ask what would happen if the algorithm found you twice as many of them at the same cost. If you would be delighted, it might deserve to be primary. If you would be suspicious, because more of that action does not obviously mean more revenue, it should probably be secondary.

Here is how that test plays out for common actions:

• Completed purchase: primary, nearly always. It is the business outcome.
• Qualified lead form submission: primary, if the form itself is a reasonable proxy for a real enquiry.
• Phone call over a meaningful length: usually primary for local service businesses.
• Add to cart: secondary in most accounts. Plenty of people add to cart and never buy, and the algorithm cannot tell the difference between a buyer and a browser from this signal alone.
• Page view of a pricing page, scroll depth, video play, outbound click: secondary, and sometimes not worth tracking as a conversion action at all.
• Newsletter sign-up: depends entirely on whether a newsletter subscriber is worth money to you. For a publisher, probably primary. For a B2B software company whose subscribers rarely convert, it is a distraction.

One detail from Google’s documentation is worth knowing. It says that primary conversion actions not used for optimization may still be used to enhance predictions. So the line between “drives bidding” and “ignored” is not completely clean. Secondary is for observation, but a primary action in a goal that a campaign is not bidding on can still add information. I would not build a strategy on that nuance, but it is a reminder that these systems are not just a simple on/off switch per column.

Three layers: goals, account defaults and custom goals

The label on each action is only the first layer. Two more decide which campaigns actually use it.

Conversion goals. Google groups conversion actions into goals by category, such as purchase, lead, sign-up and so on. Google’s page on updating your conversion goals makes a point that is easy to skim past: conversion goals are used for campaign optimization and bidding, so edits to them can change how campaigns perform. A goal is not just a reporting folder.

Account-default goals. Turning on the account-default setting for a goal means its primary conversion actions are tracked as conversions in reporting and used for bidding across all your campaigns, except those where you have set campaign-specific goals. Google also says that adding a goal as an account-level default adds it to every campaign using the account-default settings, and removing it does the reverse. Campaigns on other goal combinations are not affected either way.

So an account-default change is a wide change. One checkbox, and dozens of campaigns can be re-pointed at a different definition of success at the same moment.

Campaign-specific goals and custom goals. A campaign can opt out of the account defaults and choose its own goals. Within that, a custom goal lets you bundle specific actions together. Here is the exception that surprises people: a secondary action inside a custom goal is used for bidding when that custom goal is applied in the campaign settings. Google also notes that you can bid toward only one custom goal per campaign. And there are edge cases: the store sales (direct) action cannot be used for bidding, even in a custom goal.

Google’s own recommendation, stated in its help material, is that to get the most out of its AI and cross-campaign learning within the account, you should optimize all your campaigns toward the same set of goals. That is a recommendation from the company that sells the bidding product, so apply a little judgment. But it points at something true in practice: the more your campaigns each have a bespoke goal setup, the more places there are for one to drift without anyone noticing.

The two-condition rule

Google spells out that for a campaign to use a conversion action for bidding, two conditions must both hold:

• The conversion action is set to primary.
• The campaign is using the goal that contains that action.

Both. Miss either and the action is not steering that campaign’s bidding. This is the thing worth printing out, because almost every confusing situation I can think of in this area comes from someone checking only one of the two.

For example: you correct an action to primary and nothing changes, because the campaign you are looking at uses a custom goal that does not include it. Or you demote an action to secondary and are surprised it is still affecting a campaign, because that campaign’s custom goal explicitly includes it, and secondary actions inside a custom goal count for bidding.

When something looks off, ask the two questions in order. First, what is the label on the action? Second, which goal is this specific campaign using, and does that goal contain the action? You can answer the second one from the campaign’s settings, and you can see the goal setup across many campaigns at once from the conversions summary, where Google says you can manage conversion goals in bulk.

How accounts end up with the wrong primary

Nobody sits down and decides to optimize for page views. It happens by accident, and the accidents are fairly predictable.

The initial setup imported everything. If Google Analytics events were imported into Google Ads, or if a tag manager template created several actions in one go, many of them may have landed as primary by default. People then add the “real” conversion later and never go back to demote the rest. A practitioner write-up on this subject describes accounts that count page views as conversions and also run Smart Bidding on conversions, which compounds the error, because the bid strategy now has a clear and wrong instruction.

A thank-you page trigger that fires too broadly. A common version: a tag manager trigger meant to fire on the confirmation page fires on every page load, or on a URL pattern that matches far more than intended. Each visitor registers as a conversion. The conversion rate looks astonishing for a week, and then someone checks sales.

Double-firing. One practitioner source points out that firing the same conversion from both a hardcoded gtag.js snippet and Google Tag Manager makes every conversion count twice. This is frustratingly common after a site migration or an agency handover, where the old snippet is still in the template and the new one is in the container.

Soft actions promoted to primary. Another guide makes the sharp observation that setting actions like add to cart as primary means the system finds more of those rather than more buyers. That is the logic of the whole exercise stated plainly. Smart Bidding is not trying to guess your intent. It is trying to find people likely to do the thing you labelled primary.

New actions defaulting to the account level. A Search Engine Land report, which is now a little old and worth checking against current documentation, said Google changed how goals created through the API default to account-default status: a newly created goal would only become an account-level default if every other goal in that category already was. The detail may have moved on, but the lesson holds. When tools or integrations create conversion actions on your behalf, check what they did to your defaults afterward. Which brings up what to check before connecting any tool to your Google Ads account, since connected tools are one of the quiet ways new conversion actions appear.

Nobody owns the setup. Accounts that have passed through three freelancers and an agency often contain conversion actions whose purpose nobody remembers. They are active, they are labelled primary, and removing them feels risky. So they stay.

Conversion counting: one or every?

Primary or secondary decides whether an action drives bidding. The counting setting decides how many times it can count per ad interaction. Google’s guidance on conversion counting options is straightforward: “one” is best for leads, where only one conversion per ad click likely adds value, and “every” is best for sales, where each conversion likely adds value. The help page’s own example is a person who fills out several forms: the advertiser wants unique leads, so the lead action counts one, while the sale action counts every.

The setting is per conversion action, not account-wide. So a lead account might legitimately mix a “one” lead action with an “every” purchase action.

Why does this matter for bidding? Because an over-counted lead is a lie told to the algorithm. If a person submits a form, lands on a thank-you page, refreshes it and triggers the tag again, and the action is set to count every conversion, one real lead looks like two or three. Smart Bidding then learns that whatever sort of click produced that particular person is worth more than it is.

There is a recent data point here, with caveats. An agency audit published in August 2026 on thedoctorads.com examined settings across a single agency management account (a July 2026 snapshot) and reported that 285 of 558 lead-type conversion actions were set to count every conversion, and that 79 of the 98 accounts holding any lead action had at least one set that way. The author attributes this partly to a default, claiming Google’s default for website conversion actions is “every.” I could not verify that default claim against Google’s own documentation, so treat it as one agency’s account of what they saw. The audit also measures configuration only, not conversion volumes or outcomes. It is one agency’s book of business, not a study of the whole platform. But the underlying check costs you two minutes: open each lead-type action and confirm the count setting is what you intended.

Counting interacts with the other issues. A “page view” primary action set to count every conversion is a nearly unbounded inflation machine. A properly defined purchase action set to count every conversion is fine. A form submission set to count every conversion on a site where the thank-you page can be reloaded is a slow leak. If you are dealing with junk submissions on top of that, the problem compounds, which is covered in the post on spam and junk leads in Google Ads.

Performance Max and the “use everything” default

Performance Max adds a twist, because its conversion goal behavior is easy to assume instead of check. Google’s API documentation on campaign-level goals says that by default, Performance Max campaigns use all of the customer-level conversion goals unless you explicitly apply a subset, and that you can control this per campaign by marking each goal as biddable or not.

That has two consequences. First, a stray primary action in the account will quietly steer a Performance Max campaign unless you have scoped it. Second, if you add a new goal later, your PMax campaigns may pick it up without anyone deciding that they should.

A few Performance Max-specific goals also come with their own conditions, according to Google’s help pages. The new customer acquisition goal is available for Performance Max only, and to bid for new customers you need an audience segment with at least 1,000 active members in at least one network. Store visits and store sales cannot both be selected as the goal at the same time, and Google recommends a minimum of 10 business locations in campaigns with store visits or store sales goals. Where you track values, Google recommends maximize conversion value, and where you do not and care about all conversions equally, maximize conversions. If the new-customer angle is relevant to you, the post on the new customer acquisition goal goes into when it helps and how it can flatter ROAS.

If you are already trying to audit a Performance Max campaign when you cannot see inside it, the goal configuration is one of the few places where you do have full visibility. It is worth looking at first, before you draw conclusions from the reports.

A thought experiment, not a case study

The numbers below are invented to make the logic concrete. They are not from any real account.

Suppose a service business has two conversion actions, both primary. One is a quote request form. The other is a click on the phone number in the site header. The form is a real enquiry. The click happens on mobile every time someone taps the number, including people who immediately hang up, and people who tap by accident while scrolling. Say a hundred form submissions turn into thirty real jobs, and a hundred phone-number taps turn into maybe five.

To Smart Bidding, both are “a conversion.” If taps are cheap and plentiful, the system will find a lot of tappers, because that is the cheapest way to improve the number it was told to improve. The cost per conversion falls and the report looks healthier. Meanwhile the real jobs per pound spent drops. The fix is not clever. Demote the tap to secondary, or replace it with a call conversion that only counts calls of a minimum length, and keep the form as primary. The bid strategy now has a better question to answer.

The principle scales up. Whenever your primary actions have very different downstream values and you treat them as equal, the algorithm will gravitate to the cheap, low-value one. If the actions really do differ in value, either give them different conversion values or keep only the valuable one primary. The post on conversion value rules covers one way to express that difference when the same action is worth more in some contexts than others.

How to audit your own setup

You do not need special software to do this. You need an hour and a willingness to be slightly alarmed. Work through it in order.

Step 1: List every conversion action. Open the conversions summary and write down the full list: name, goal category, source (website tag, Analytics import, call, offline import), primary or secondary, count setting, and status. Google notes that removed actions are archived and not deleted, and that you can re-enable them, so the list may be longer than the default view suggests if you include inactive ones.

Step 2: Sort them by “would I pay for more of this?” Use the doubling test from earlier. Mark each as keep primary, demote, or investigate. The “investigate” pile is usually where the interesting surprises sit: actions nobody can explain, with suspiciously high volume.

Step 3: Check the volume against reality. For each primary action, compare the count in Google Ads to a source you trust: orders in your shop backend, leads in your CRM, calls in your phone system. They will not match exactly, because of attribution differences, consent settings and conversion lag. But they should be in the same neighbourhood. An action that reports ten times what your backend shows is a candidate for double-firing or a runaway trigger. Lag matters here: the post on conversion lag explains why the most recent couple of weeks will always look thinner than the truth, so compare periods that have had time to mature.

Step 4: Check the count setting for each lead action. Anything that is a lead, sign-up or enquiry should usually be “one,” per Google’s guidance. Anything that is a sale should usually be “every.” Look for exceptions and ask why they exist.

Step 5: Check the tags themselves. Use tag preview in your tag manager and walk through the conversion path yourself. Does the conversion fire once? Does it fire on the right page? Does it fire again on refresh? Is there both a hardcoded snippet and a container tag? If you use enhanced conversions, confirm they are working, because broken enhanced conversions fail quietly; see the post on enhanced conversions for how that tends to go.

Step 6: Check which goals each campaign actually uses. This is the second half of the two-condition rule, and it is where people stop too early. Open the goal settings for each campaign, or use the bulk “update conversion goals” option from the campaigns table, which Google describes in its help for campaign-specific goals. Note any campaign using a custom goal and read exactly what is inside it. Pay particular attention to Performance Max campaigns, given the use-everything default.

Step 7: Check your offline imports. If you import offline conversions, make sure they land in a goal that your campaigns actually bid on. An offline import that is set to secondary, or sits in a goal no campaign uses, is data you are paying to collect and not using. There is more on this in the post about offline conversion imports and Smart Bidding in 2026.

Step 8: Write it down. A one-page document listing each primary action, what it represents, and why it is primary will save the next person (or you in a year) from reconstructing the logic. This sounds trivial. It is probably the highest-value step on the list.

Changing goals without breaking your bidding

Once you have found something to fix, the question is how carefully to change it. Google is clear about two things. First, its help material says AI bidding models take time to adapt to changes in your conversion configuration, at both account and campaign level. Second, its guidance on Smart Bidding is to avoid frequent manual changes to budgets, targets or conversion goals during the learning period, because those adjustments can reset the learning window. The commonly quoted figure for Smart Bidding’s learning period is seven to fourteen days. I would treat that as a rough planning number rather than a guarantee. I could not find a Google page that states exactly how long a campaign will sit in learning after a goal change, so I will not pretend to know.

Given that, a reasonable approach is:

• Do the clean-up in one pass, not in a dribble. Making one small goal change a week means the system is perpetually adjusting. Better to decide what the right configuration is, make the changes together, and then leave it alone. The post on the Smart Bidding learning phase covers why editing mid-calibration keeps resetting things.
• Time the change away from your peak. If you run a seasonal business, do not re-point your bidding at a different goal a week before your busiest period. If you are heading into Q4, there is a case for fixing this early, or for waiting until after the rush, and not for doing it halfway through.
• Expect a short dip, and note the date. Removing a flattering primary action will make your reported conversions fall. That is the correct outcome, not a regression. Record the date of the change so that in a month you are comparing like with like. The change history will log the edit, which is useful when someone later asks why the numbers dropped.
• Watch the real outcome, not the column. After you change the primary actions, the Conversions column stops being comparable to its past self. Judge performance against your backend: orders, qualified leads, revenue.
• Recategorise before you delete. Google’s goal-updating guide describes moving a miscategorized action out of a category like “Page View” or “Other” into a meaningful one such as “Purchase” or “Sign-up” as a way to unlock bidding improvements, while warning that it can affect optimization and performance. Removal is possible too, with the action archived and re-enable-able, but demoting to secondary first gives you a reversible step and a chance to see what happens.

If you are changing the goal for a campaign that is on a different bid strategy, the interaction between strategy and goal also matters. A value-based strategy needs values attached to the primary actions; a count-based strategy treats every primary action as equal. The post comparing Target CPA, Target ROAS and Maximize Conversions is the place to sort that out, and if your account is thin on conversions, how much conversion data Smart Bidding needs may argue for a slightly higher-up-the-funnel primary action than you would otherwise choose.

The micro-conversion question

The honest counterargument to “only count real outcomes” is data volume. A business that gets four sales a month cannot feed a bid strategy much. In that situation, people reach for a micro-conversion, such as a high-intent page visit or a started checkout, to give the algorithm more signal.

There is a case for it, and there is a way to do it badly. The case: Smart Bidding works better with more events, and a well-chosen earlier step can be strongly correlated with the final outcome. The danger: the algorithm will optimize hard for the earlier step whether or not it correlates with revenue in your business. A cheap, plentiful action will win every time.

If you do go this way, some guardrails help:

• Pick a step that is genuinely hard to do casually. Starting a checkout beats viewing a product page. Submitting a form beats opening one.
• Check that the micro-conversion rate is stable relative to the real outcome over a few months of data. If the ratio swings wildly, it is not a good proxy.
• Keep the real outcome in the account as a secondary action, so you can see the relationship and notice when it breaks.
• Plan to revisit. When volume grows, move the real outcome back to primary.

This is a judgment call, and I would not claim there is one right answer. What is wrong is doing it by accident, with no one having decided it, which is what most accounts with a page view counted as a conversion are doing.

What to look at every so often

A goal setup is not a one-time project. It drifts. A developer adds a tag. A plugin updates. Someone imports a new Analytics event. An integration creates an action. A campaign gets duplicated with a custom goal that no longer matches. None of these announces itself, and the first symptom is usually a bid strategy that quietly stops behaving the way it used to.

A light-touch routine works better than a heroic annual audit. Once a month, spend ten minutes checking:

• Has the list of primary actions changed since last time?
• Has any action’s volume jumped or collapsed relative to the previous month?
• Is the Google Ads conversion count still in the same neighbourhood as your backend?
• Have any new campaigns been created with their own goal settings?

If that feels like a lot to remember, the older post on how often you should actually check your Google Ads account puts the monitoring rhythm into perspective.

Where daily monitoring fits in

This is the kind of drift that is awkward to catch by looking at the interface once in a while, because nothing is red and nothing is flagged. Growera’s Google Ads manager reads your account every day, and a change in conversion volume on a primary action, or a jump in cost per conversion with no matching movement in real leads, is the type of pattern a daily check is built to notice and bring to your attention. It does not replace deciding what should be primary, which is a business decision that only you can make. What it can do is tell you sooner when the numbers stop behaving as they did, so you are not discovering a runaway trigger a month after it started. If you are curious about how the product works in more detail, the ERA AI page explains it, and the pricing page shows what it costs.

Common questions

Should I ever have more than one primary action? Yes, when they represent genuinely different valuable outcomes, such as purchases and qualified calls for a business that wins work both ways. One agency source suggests keeping just one primary conversion for purchases to avoid double or triple counting, and setting the others to secondary once the data is verified; that advice makes sense for a single purchase flow where several tags describe the same event. The issue is overlap, not the number itself.

If I demote an action to secondary, do I lose the data? No. Secondary actions still appear in the “All conversions” column, so you keep the reporting. What you lose is the bidding influence, which is the point.

Do I need to fix Analytics too? The two systems are separate. Fixing goals in Google Ads does not change what Analytics counts, and the reverse. If your numbers disagree, work out whether the difference is attribution, counting or tagging before you adjust anything. The post on auto-tagging and GCLID covers one common source of mismatch between clicks and sessions.

Will fixing this immediately improve performance? Not necessarily, and I would be wary of anyone who promises it will. What you get is that bidding now optimizes toward the outcome you actually want. Whether that produces better results depends on how bad the previous setup was, how much data you have, and a long list of other factors. The more reliable benefit is that your reporting becomes trustworthy, so later decisions are made on numbers that mean something.

Can an agency or a tool just handle this for me? They can help, but only you know what a real customer is worth in your business. Whoever manages the account should be able to tell you, in a sentence each, what every primary action is and why it is primary. If they cannot, that is information. If you are weighing that question more broadly, what Google Ads management costs compares the options.

Summary

The Conversions column is the one Smart Bidding reads, and it is built from your primary actions in the goals your campaigns use. Both halves matter: an action must be primary, and the campaign must be using the goal that holds it. Account-default goals apply broadly and change many campaigns at once, custom goals override them and can pull in secondary actions, and Performance Max defaults to using all customer-level goals unless you scope it.

Most accounts that have this wrong got there by accident: imported events, a trigger that fires too often, a duplicate tag, a soft action promoted to primary, or a tool that created actions nobody reviewed. The count setting adds a second layer, with Google’s guidance being “one” for leads and “every” for sales.

The audit is dull and worth doing. List every action, ask whether you would pay for more of it, compare volumes to your backend, check the count setting and the tags, then check which goals each campaign uses. Make the changes in one pass, note the date, expect the reported numbers to fall if you were inflating them, and judge results by real orders and leads, not by the column that just changed definition.

Then write down what each primary action is for. That page is what keeps the setup from drifting back into the same mess the next time someone adds a tag.