Primary and Secondary Conversions in Google Ads: What Your Algorithm Actually Learns
There is a familiar kind of account: the Conversions column shows a healthy number, and the CRM shows three closed deals for the same month. The cause is rarely “bad traffic”. It is usually that the algorithm was told to learn the wrong thing. It is doing exactly what was asked — maximising the events you flagged as meaningful. If those events include a phone-number click, a contact-page view and a callback request, Smart Bidding will diligently hunt for people who click, view and request. Not people who buy.
Primary and secondary conversions are the switch that decides what enters the learning process and what enters reporting. It looks like a minor toggle in a list of conversion actions, and it quietly controls where your bids go. This article covers how the action-goal-campaign chain works, what actually changes when you flip that toggle, and how to build a signal that optimises toward a qualified lead rather than an imitation of one.
The algorithm does not optimise for “sales”. It optimises for whatever you put in the Conversions column. The gap between those two things is your real lead-quality problem.
Three levels you have to keep separate
Nearly all confusion comes from three different entities being blurred together in the interface. Let us pull them apart.
| Level | What it is | Where it is configured |
|---|---|---|
| Conversion action | A specific event: purchase, form submit, call longer than 60 seconds | Goals → Summary → the individual action |
| Conversion goal | A group of actions with the same meaning: Purchase, Lead, Book appointment | Goals → Summary, actions grouped inside a goal |
| Campaign setting | Which goals a given campaign uses: account-default goals or its own set | Campaign settings → Conversion goals |
The primary/secondary switch sits at the level of an action inside a goal. And here is the detail that breaks half of all setups: if a campaign uses its own conversion goals rather than the account defaults, your account-level toggle does not affect it. Before arguing with the interface, open the campaign settings and check what it is actually following.
What genuinely changes when you flip the switch
| Aspect | Primary | Secondary |
|---|---|---|
| Used for bidding | Yes | No — unless the action sits inside a custom goal the campaign bids toward |
| Conversions column | Included | Excluded |
| All conversions column | Included | Included |
| Affects cost per conversion and conversion rate | Directly | No |
| Role | Optimisation target | Observation and diagnostics |
That gives a simple working definition. A primary conversion is the thing you are paying for clicks to get. A secondary conversion is the thing you look at to understand why you are not getting enough primaries. Micro-events — pricing page views, brochure downloads, add-to-cart — should almost always be secondary: valuable as funnel diagnostics, destructive as optimisation targets. Their correct use is covered in our piece on micro-conversions and funnel optimisation.
The exception everyone forgets
A secondary action can still influence bidding — when it is part of a custom goal and a campaign is set to optimise toward that goal. There is no contradiction on paper, but in practice this is the number one source of “inexplicable” behaviour: the account setting says one thing and the campaign does another. Hygiene rule: if you use custom goals, keep an explicit list of which campaign bids toward which goal.
How many primary actions should you have?
The most common mistake is marking everything primary “so the algorithm has more data”. It achieves the opposite. When a single goal contains a $500 purchase and a phone-number click, the system treats them as equivalent events unless they carry different values. The outcome is predictable: it finds as many of the cheap event as it can.
A working configuration looks like this:
- One core primary action — the thing you consider money: purchase, qualified lead, confirmed booking.
- At most one or two additional primaries — only if they are economically equivalent and carry correct values.
- Everything else secondary. It still counts in All conversions and remains available for analysis.
If the core action does not produce enough volume for the algorithm (as a directional benchmark, fewer than ~30 conversions per 30 days per campaign for target CPA), the answer is not to promote junk actions to primary. It is to consolidate campaigns, simplify structure, or temporarily optimise toward a higher-funnel but still meaningful event. Why data volume matters so much is covered in our article on the Smart Bidding learning period.
Telling a junk lead from a real one, in the data
“Bad leads” is not one category. Before fixing anything, work out which of four types you are dealing with — they have different causes and different cures.
| Type | How it looks | Root cause | Fix |
|---|---|---|---|
| Bot spam | Gibberish form fields, bursts of submissions within a minute | Unprotected form | CAPTCHA, honeypot field, server-side validation |
| Wrong intent | The person wanted a job, a freebie, or a competitor | Broad matching without negatives | Search terms work and negative keywords |
| Unqualified | Right topic, wrong budget, region or segment | Offer and landing page do not filter | Qualifying questions in the form, pricing on the page |
| Too early | Genuine interest, wrong timing | Upper-funnel events set as primary | Move to secondary, optimise toward a lower-funnel event |
The first type is fixed on the site, the second through negative keywords and search term analysis, the third in the offer and the form, and only the fourth in conversion configuration. Until you have separated those four streams, any primary/secondary tinkering is cosmetic. The traffic-side layer of protection is covered in our guide to traffic anti-fraud.
Teaching the algorithm quality instead of volume
The whole point of this architecture is to send Google not “a lead happened” but “the lead turned out to be good”. There are three maturity levels.
Level 1: split actions by quality
The simplest version: create two conversion actions — Lead (secondary) and Qualified lead (primary). The second fires not on form submit but after a check: by a salesperson, by a scoring rule, by automated phone validation. Even that crude fork changes who the system goes looking for.
Level 2: import offline conversions
When the deal closes in a CRM days or weeks later, that status has to travel back to Google Ads. The mechanics and requirements are detailed in our guide to offline conversion import. The binding constraint is identity: without a correctly captured click ID or user-provided data, nothing joins up. The companion topic is enhanced conversions, which raise match rates where the click identifier gets lost.
Level 3: value instead of count
The top tier is sending money rather than a unit: different values for different lead types, expected margin, close probability from a scoring model. At that point the algorithm stops counting leads and starts counting revenue. The tooling and standard patterns are in our article on conversion value and value rules.
Each level demands cleaner data than the one before it. Starting at level three without fixing level one is a reliable way to train the algorithm on beautifully formatted garbage.
Configuration in eight steps
- Export every conversion action. A typical running account has 8–20 of them and nobody remembers half.
- Tag each with its source and creation date. Actions from GA4, from the tag, from imports, duplicates of the same event — all of it needs separating.
- Find duplicates. The same purchase arriving through both the tag and a GA4 import doubles conversions and halves your apparent CPA.
- Identify the single money action. Ask what you are paying for clicks to get. That is your primary conversion.
- Move everything else to secondary. Delete nothing — secondaries keep counting in All conversions.
- Audit each campaign’s settings. If a campaign uses its own goals, either align them or return it to account defaults.
- Check conversion windows. Actions with the same meaning must share the same window, otherwise the numbers are not comparable — details in our breakdown of conversion windows and lag.
- Give the system time. Changing the conversion set used for bidding is a high-impact edit: budget for a learning period plus a full conversion lag before drawing conclusions.
How long to wait and how to judge the result
Changing the primary action is not cosmetic — it changes the objective function. In the first days after the switch the numbers almost always look worse: fewer conversions in the column (because they are now counted more strictly) and a higher CPA. That is expected, and it is not a reason to roll back after three days.
| Horizon | What to look at | What not to do |
|---|---|---|
| Days 1–7 | Tracking correctness: events arriving, no duplicates | Do not assess performance at all |
| Days 8–21 | Strategy exiting learning, volume stabilising | Do not jerk targets or budgets around |
| Days 22–45 | CPA on the new action, share of qualified leads in CRM | Do not compare to “before” without adjusting for a different conversion definition |
| After 45 days | Cost per closed deal and revenue per dollar spent | Do not revert to the old set without a test |
The only correct way to compare before and after is not to place two historical periods side by side but to run an experiment where half the traffic keeps the old goal set.
Seven mistakes we see most often
Everything marked primary. The classic. The algorithm receives a mixed bag and optimises toward the cheapest event in it.
Double-counting purchases. The tag and an analytics import record the same event. Apparent CPA is half the real one and scaling decisions rest on phantom numbers.
A campaign on forgotten custom goals. The account setting was changed, the campaign still follows an old goal, and nobody understands why.
Different conversion windows on equivalent actions. “Lead” on a 30-day window and “Qualified lead” on 90 days are incomparable by construction.
Changing the conversion set alongside budget and strategy. Three edits on the same day and the effect can never be attributed.
Optimising toward calls of any length. A 60–90 second threshold filters accidental and misdialled calls better than any negative keyword list.
Importing CRM statuses without checking match rates. If 40% of leads never join to a click, the algorithm learns from a biased sample — worse than not importing at all.
A 40-minute checklist
- Export all conversion actions with source, primary/secondary status and conversion window.
- Identify and remove duplicates.
- Keep one money action as primary; move the rest to secondary.
- Check every campaign for custom conversion goals.
- Align conversion windows inside each goal.
- Reconcile monthly Google Ads conversions against CRM leads — a gap above 20% needs an explanation.
- Check the offline conversion match rate if imports are running.
- Record the change date and leave the account alone for two weeks.
Reconciling Google Ads with your CRM
Configuring primary and secondary conversions only makes sense once you know how closely the interface numbers track reality. A regular reconciliation takes half an hour a month and usually returns more than any bid tuning.
- Take one complete closed month — not the current one, so conversions have time to back-fill.
- Export primary conversions from Google Ads for that month by conversion date rather than click date: CRM records events on the event date.
- Export CRM leads attributed to Google Ads for the same period.
- Compare the two numbers. A gap of 10–15% is normal: some leads arrive without correct tagging, some are removed as duplicates.
- A gap above 20% deserves a specific explanation, not a shrug about “attribution”.
| Symptom | Likely cause | Where to look |
|---|---|---|
| Google Ads materially higher than CRM | Double counting, or the conversion fires in the wrong place | Conversion action list, duplicates, tag firing conditions |
| CRM higher than Google Ads | Lost tagging: redirects, iframed forms, untracked calls | URL parameters, tracking templates, call tracking |
| Numbers match but no deals | Not a tracking problem — a qualification problem | Offer, form, target queries |
| A sharp break from a specific date | A site change: deploy, new form, CMS migration | Release log, tag change date |
If you lose the click identifier across redirects and hand-offs, tracking parameters deserve their own pass — the mechanics are in our article on GCLID, GBRAID and tracking templates.
Three standard configurations by business type
There is no universal conversion set — the right choice depends on your sales cycle and on where money actually changes hands.
Short-cycle e-commerce
Primary: purchase with order value passed. Secondary: add to cart, begin checkout, newsletter signup. The difficulty here is not picking the action — it is getting value right. Returns, discounts and shipping costs must be reflected, otherwise the algorithm optimises gross revenue and delivers unprofitable orders.
Long-cycle lead generation
Primary: a qualified lead imported from the CRM days later. Secondary: form submission, calls under a minute, content downloads. The critical parameter is the conversion window — it must cover your typical qualification time, or a large share of deals will never join back to a click.
Services driven by calls and bookings
Primary: a call longer than 60–90 seconds, or a confirmed booking. Secondary: phone number clicks, contact page views, direction requests. The call-duration threshold matters most here: without it you optimise toward misdials and cold callers.
| Business type | Primary action | Directional conversion window | Main risk |
|---|---|---|---|
| E-commerce | Purchase with value | 7–30 days | Value that ignores returns |
| B2B lead gen | CRM-qualified lead | 60–90 days | Low import match rate |
| Local services | Call of 60+ seconds | 7–30 days | Optimising toward short calls |
| Subscription | Paid subscription | 30–60 days | Trials that never convert to paid |
Treat those figures as orientation rather than policy: the correct window comes from your own conversion back-fill curve, not from a table.
Where PPC Rebels fits
Conversion configuration is half the job; the other half is running it on an account that behaves predictably. If you need a working account with history, look at our Google Ads agency accounts. If you want to work through measurement and bidding systematically, we run Google Ads training. Closely related pieces are Google Ads brand lists and Google Ads automated rules — they are what gives you an early warning when tracking breaks and conversions stop arriving.
FAQ
Primary versus secondary conversion in one sentence?
Primary participates in bidding and appears in the Conversions column. Secondary does not affect bidding and shows only in All conversions.
Should I delete secondary actions I do not bid on?
No. They cost nothing and they are useful diagnostics — they show where the funnel breaks. Deleting them removes data for no benefit.
How many primary actions are optimal?
One, if volume allows. Two or three only when they are economically equivalent and carry correct values.
I set an action to secondary and bidding did not change — why?
Most likely the campaign uses its own conversion goal set rather than the account defaults. Check that specific campaign’s settings.
Can a secondary action affect bids?
Yes — when it belongs to a custom goal the campaign optimises toward. That is the one documented exception.
What if my core action produces too little data to learn on?
Consolidate structure, merge campaigns, or temporarily optimise toward a meaningful higher-funnel event. Promoting junk actions to primary is the worst available option.
How do I handle online purchases and phone calls together?
Both belong in primary if both produce revenue. The condition is that they carry different, correct values — otherwise the algorithm chases the cheaper one.
Does flipping primary/secondary change historical data?
Reporting is recalculated against the current setting, so the historical Conversions column can shift. Always record the switch date, otherwise period comparisons will be invalid.
Should add-to-cart ever be primary?
Only as a temporary measure under severe data scarcity, with a planned return to purchase. Over any real time horizon it degrades traffic quality.
How long should I wait after changing the primary action?
At minimum the learning period plus a full conversion lag — usually two to four weeks. First-week conclusions are almost always wrong.
How do I know the algorithm is learning from junk?
Google Ads conversions rise while CRM deals stay flat or fall. That divergence is the most reliable indicator that signal and reality have parted ways.
Do I need offline conversions for this?
Not strictly, but they are what closes the gap between “a lead” and “money”. Without them, your accuracy ceiling is the quality of your online qualification.
