Google Ads Lead Forms: The Checks That Decide Whether Yours Is Working

2026-09-10

The short answer

A Google Ads lead form asset is a form that opens inside the ad. Someone taps the call to action, fills in their details on the results page or in the feed, and never lands on your site. Google's documentation describes lead forms as available on Search and Performance Max campaigns, with several delivery options: manual CSV export, an email notification per lead, a webhook, or the Ads API — and it notes that leads are stored for 60 days, with CSV downloads covering the last 30. That retention window is the first practical fact worth internalising, because until the lead reaches a system you control, the record you are relying on is Google's copy.

Setup is the easy part, and the pages that already rank cover it well. What they do not cover is the diagnostic consequence: when the form moves into the ad, your website leaves the measurement path. Page views, scroll behaviour, and a thank-you page URL are no longer available as evidence, and whatever the site journey was doing to shape who reached the form is no longer doing it. Both changes affect what your numbers mean.

The one thing that actually changed

A lead form asset removes a handoff between tools and creates a different one. Before, the path crossed from ad platform to your site, and your site recorded the completion. Now the path stays inside Google until the lead is delivered somewhere else — an inbox, a spreadsheet, a webhook endpoint, a CRM. Every point where the domain, platform, or tool changes needs a verified event assigned to it. That new delivery boundary is exactly such a point, and it is easy to leave uninstrumented because the form feels like it lives in one place.

Second, the depth of what you have proved changed. Name the deepest signal you actually hold — interest, click, lead, booked call, payment, repeat purchase — and treat everything past it as unproven. An in-ad form submission sits at 'lead'. It is a real rung, and it is not evidence that this person will take a call, pay, or stay. Skip this framing only if you already have profitable repeat purchase from this source; then depth is not your open question.

Work the stages in order

When there is no obvious constraint, there is a sequence that stops you diagnosing the wrong layer: viability, then measurement readiness, then offer and message, then traffic, then page and funnel, then lifecycle, then economics. Each stage is a precursor to the ones after it. A failure at measurement makes any conclusion you draw about creative or economics unreliable, which is why the order is not a preference. Stop at the first stage that fails.

The sections below follow that order, covering the stages an in-ad form specifically changes. Each check carries its own boundary — read those, because a check that does not apply to you is not the same as a check you can ignore.

Measurement readiness: is the lead data real?

Submit the form yourself, then follow it. These are two separate checks and they fail for different reasons. First, on a real device, complete the form as a stranger would and confirm Google records the submission. If the event never fires, the number you have been optimising against was not recording that action. Skip this only if the event has been verified since the last change to the form, the campaign, or the delivery method.

Second, go and find that same lead where it is supposed to arrive — the inbox, the webhook log, the CRM record. A submission Google recorded correctly can still fail on the way out, so non-arrival in your CRM tells you the delivery path broke, not that the submission event is broken. Work out which of the two it is before you touch anything.

Instrument the delivery boundary. Draw the actual path: ad, form, Google's lead store, delivery mechanism, your CRM. Assign a verified event to each point where the tool changes. Email notifications, webhooks, API pulls and manual CSV exports are all boundaries. If manual export is your only delivery route, a lead that nobody downloads before the retention window closes leaves no first-party record at all — where email, webhook, API or an integration is also running, that lead may already sit in an inbox or a CRM regardless of whether anyone touched the CSV. Skip if the whole path runs on one platform with no crossings and completion is verified.

Reconcile against the system that actually receives the outcome. Decide which system is authoritative for a lead — if accepted leads are recorded in your CRM, that is the one — and check Google's reported conversion count against it for the same period. If someone says there are no leads, or the two counts disagree, that disagreement is the finding, not the lead volume. Skip if platform numbers and first-party records already agree within a margin you understand.

Define the event before you quote a rate. Write down what counts as a conversion here and what population it is divided by. 'Conversion' can mean form opened, form submitted, submitted-and-verified, or delivered-to-CRM. Two of those numbers can sit in the same dashboard and describe different things, and if your account records several conversion actions, segment by action before you read the total as one thing. If you are trying to work out whether a figure is decision-ready before acting on it, our guide to cost per lead as a decision-ready number walks through the same discipline. Skip if every metric in play already has a written definition with an explicit denominator.

Do not average the form with the landing page. If you are running an in-ad form and a website form at the same time, report them separately by source. A combined cost per lead is fine for operations and useless as a test result: blend two acquisition routes into one number and that number cannot tell you which one moved. Skip if there genuinely is only one source, one version, and one offer in the period.

Offer and message: does the person know what they agreed to?

Compare the two routes as you have actually built them rather than in the abstract. An in-ad form can carry ten questions; a landing page can carry two. What matters is the ask and the explanation your specific configuration presents. Two checks follow.

Name the actual next step in the call to action and the form description. Say what happens after they submit — who contacts them, how, and about what. Anything vaguer asks people to agree to something they cannot name. Skip if your call to action already names the specific next action and nothing later-stage is introduced prematurely.

Check that the commitment grows gradually. Write down the full ask at each step: money, time, effort, preparation, trust. A one-tap form followed by a demand for a sales call, a payment, or a long programme is a sharp jump, and conversion can drop hard at one specific transition. Skip if each step's ask is already proportionate to what the prospect has received so far.

Make every question earn its place. Google's forms allow custom questions, so qualifying questions are one control available to you. The requirement is tighter than adding a few: for each question, name the decision it supports, and match the answer format to how the answer behaves in reality — questions that force a single choice where several are true corrupt the result you were trying to protect. Skip if every question already changes scoring, qualification, education, or the recommended action. If quality is your live complaint, diagnose it before you start bolting on filters — see how to diagnose lead quality first.

Traffic: the ad or the people?

If form completions look healthy and sales do not follow, investigate who is arriving before rewriting the creative. Strong engagement with weak downstream conversion is the pattern that points at the composition of the traffic — a geography change, an audience change, cheap inventory bought because it was cheap. That tells you where to look, not what to conclude; the creative, the offer, and how leads are handled may all still be part of it. Skip if engagement and downstream conversion moved together, because then the ad and the traffic are telling you the same story. The same ordering logic applies across platforms, which is why our walkthrough of diagnosing ads that are not converting in stage order reaches for the same check.

What the campaign is being pointed at

Once you have looked at who is arriving, look at what you asked the campaign to buy. Open the settings and read the optimisation event. You are trading two things against each other: how deep the outcome is, and how much signal it receives. Pick the deepest real business event that still gets enough volume to learn from.

A form submission is shallow by design. If optimisation is pointed at that shallow event and lead quality is poor, that trade-off is where to look. If it is pointed at a deep event so sparse that there is barely anything to learn from, that is the other failure mode. And if the optimisation event is not firing at all, treat platform optimisation and every conclusion that depends on that event as unreliable — spend, clicks, your CRM records and your payments are separate evidence and still stand. Skip if your event is deep, firing, and receiving volume you can learn from.

Where the real outcome is too rare at your current volume, using a form submission as a proxy is a legitimate concession — but keep it labelled as a proxy and track the real outcome separately and continuously. The failure is not using a proxy; it is forgetting it was one.

Lifecycle: what happens in the hours after

Two things worth separating here.

Do not add a nurture sequence as an automatic rescue. Before inserting an email step between the form and the offer, name the reason this buyer needs time. If you cannot, you may be adding another place to lose people rather than fixing the thing that was losing them. Skip if the purchase genuinely requires consideration and the sequence serves that consideration. Response speed is a separate question with its own evidence — worth checking against whether response time is actually your constraint rather than assuming it is.

Attribute revenue back to the enquiry that produced it. If buying takes weeks, link each sale to the form submission that started it and report immature cohorts separately from matured ones. If the cohort has not matured, you are reading a half-finished result and calling it a verdict. Skip if purchase is effectively immediate; then same-period comparison is valid.

Before you decide the architecture is right

The choice between an in-ad form and a landing page is an architecture decision, and it should come from what your own traffic has actually done — has cold traffic here ever purchased, and does this capture route attract people who buy? Convention is not evidence. If your current architecture is converting inside economics you have modelled, architecture is not your present constraint, and the work belongs somewhere else on this list.

One gate before any of it: whenever something is about to launch or relaunch, check budget type and amount, dates, projected spend, the optimisation event, the audience, the destination or form, and the tracking — then re-read the live settings immediately after publishing. That is a pre-launch check, not a diagnosis, and it is cheap.

If you work these checks and none of them isolates a cause, the honest conclusion is that the evidence in front of you is not sufficient to diagnose it yet. Insufficient volume, execution problems, timing, and interactions between factors can all survive every check on this list. That points at more first-party investigation, not at a bolder guess.

If you want a second read on where your own campaign is losing people, GrowthHackr's Campaign Leak Audit walks your setup through this kind of stage-by-stage diagnosis and costs nothing.