Setting Up a Conversion That Measures Something Worth Measuring
You tracked pageviews, marked one as a conversion, and the report told you nothing useful. That is the expected outcome, not a setup error. A pageview records that a browser requested an address. It is not evidence that a person wanted anything, and no amount of correct configuration turns it into evidence.
The setup is the easy half of this job and it is the half every guide covers. The hard half is deciding what to measure, and it happens before you open the interface. This article is about that half.
We have run no advertising campaigns and hold no ad accounts, so nothing here is a result from anyone's account. It is a way of choosing, and the choosing is where the value is.
The direct answer
Measure the shallowest action that only a genuinely interested person would take. Shallow enough to happen often enough to be useful for optimization. Deep enough that taking it means something. If an action can be taken by accident, by a bot, or by someone who is lost, it is not a conversion, whatever you have labelled it.
The rest of this is how to find that action on your own site and how to check that you found it.
Why a pageview is not a conversion
A pageview can be produced by accident. Someone lands on the wrong page and leaves.
It can be produced by something that is not a person. Automated traffic reaches ordinary pages routinely.
It can be produced by the same person twice. They come back, they refresh, they open it in two tabs.
And it carries no intent. Arriving on a pricing page and reading a pricing page and being ready to buy are three different things and a pageview cannot distinguish them.
There is one narrow exception, and it is the reason the pageview approach persists: a page nobody can reach except by completing something. A confirmation page after a form submission, a receipt page after a purchase. Those are not really pageviews of a page, they are the visible evidence of a completed action. That is a legitimate conversion and it is a different thing from marking your pricing page as one.
Start from the money, not from the site
Most bad conversion setups happen because somebody opened the analytics interface and looked for something to track. Do it the other way round.
Write down what a good outcome is for this business, in a sentence, without using an analytics word. "Somebody calls us and we quote them." "Somebody buys." "Somebody books a demo and turns up to it."
Then ask which visible thing on the website happens on the way to that sentence. Usually there are two or three: the form goes in, the phone number gets tapped, the calendar booking completes.
Then ask which of those you can actually observe. A phone call to a number printed on the page is not observable by the website. A tap on a phone link on a mobile device is. Those are different measurements and treating the second as if it were the first is one of the more common quiet errors in this whole subject.
What you now have is a candidate. It came from the business outcome rather than from the interface, which is the entire difference between a measurement that is worth optimizing toward and one that is not.
The depth problem, and how to resolve it
The instinct is to measure the deepest action available, because it is the most meaningful. On a small account that instinct is wrong, and it is wrong for a mechanical reason.
Automated bidding learns from conversions. Few conversions means little to learn from, and a strategy optimizing toward an action that happens rarely spends a long time not knowing what it is doing. A shallower action that happens more often gives the system something to work with.
But shallow enough and you are optimizing toward noise. If the action does not distinguish interested people from everybody else, you have taught the system to find people who take a meaningless action, and it will.
The resolution is to measure both and be explicit about which one you are optimizing toward. Measure the deep action because it is what the business cares about. Measure a shallower one if you need volume for bidding. And then hold in your head, and preferably in a document, which of them is the target and which is the ground truth. Confusing the two is how a campaign comes to look excellent and produce nothing.
The four tests a candidate action has to pass
Before you set anything up, put your candidate through these.
Can it happen by accident? If a stray tap or a mis-click produces it, it will be produced by stray taps and mis-clicks, at a rate you cannot estimate and will not notice.
Can it happen more than once for the same intent? Somebody who submits the form, gets an error, and submits it again has one intent and two events. Decide now whether you count that as one or two, and know which way you have configured it, because the number in your report depends entirely on that choice and it is not usually visible in the report itself.
Would you be pleased if this number doubled next month? This is the most useful of the four and it takes five seconds. If doubling it would not obviously be good news, it is the wrong thing to measure. Twice as many people reaching the pricing page is not obviously good news. Twice as many completed bookings is.
Can you tell, later, which of these turned into real business? Not necessarily in the analytics tool. In your inbox, your CRM, your phone log. If there is no way to trace any of these back to an actual outcome, you are measuring a proxy and you will never know how good a proxy it is.
Now set it up, and where to get the steps
The mechanics are the easy part and this article deliberately does not walk through them. The interface, the menu names and the terminology in Google Analytics change, and a set of steps written today would be wrong steps by the time somebody follows them. Take the steps from Google's own current Analytics documentation on the day you do it, and take the terminology from there too, because it has changed more than once and will change again.
Three things worth knowing that are not steps:
The action has to be observable to the tool before it can be marked as anything. If the event is not arriving, no amount of configuration downstream helps. That is the first thing to check when something does not appear.
Marking something as important in Analytics and having it count in your ads platform are two different acts. They live in different products with different rules and different counting behavior, and they are commonly assumed to be one action. They are not, and this is a common reason a conversion "does not show up."
Names are permanent in practice. Whatever you call the event, you will be reading that name for years and so will everybody else. Name it after the thing that happened rather than after the page it happened on, because pages get renamed and things that happened do not.
Verify it before you trust it, and there is only one way
Do the action yourself and check that it arrives. Fill in your own form. Tap your own phone link on an actual phone. Complete your own booking.
Then check three things: that it appeared, that it appeared once rather than three times, and that it did not also fire when you did something else. The third is the one people skip, and a conversion that fires on the wrong action is worse than no conversion, because it is confidently wrong and it will be optimized toward.
Then check it again in a month. Site changes break tracking silently. A form gets redesigned, a page gets rebuilt, a tag manager container gets tidied, and the event stops arriving with nothing in the interface to say so. The report does not show an error. It shows a lower number, and lower numbers get attributed to the market rather than to the tracking.
The failure this whole article is trying to prevent
A conversion that is easy to measure and means nothing gets set up because it was available. It produces a number. The number goes into a report and into an automated bid strategy. Both then work extremely hard to increase a number that does not correspond to anything the business wants.
That is a worse position than having no conversion tracking at all, because it is confident. Nobody questions a metric that is going up.
The check that catches it is question three above: would you be pleased if this doubled? Ask it before you configure anything, and ask it again about the metric you have been reporting for the last six months.
And it connects to what your budget is doing
The reason this matters beyond tidiness is that a conversion definition is what automated bidding optimizes against. A weak definition sends your money toward people who take a weak action, which is one of the specific ways a small budget disappears without anybody being able to say where it went. That mechanism, and the two others that sit alongside it, are what where a small Google Ads budget actually goes works through in order.
The short version
Decide what a good outcome is, in a sentence with no analytics words in it. Find the shallowest observable action on the way to it that only an interested person would take. Put it through the four tests, especially "would I be pleased if this doubled." Take the setup steps from Google's current documentation rather than from any article, including this one. Do the action yourself and confirm it arrives once and only when it should. Check again in a month.
The interface part takes ten minutes. The choosing part is the job, and it is the part that decides whether every number you report for the next year means anything.