Google's invalid-traffic card has two numbers. Only one is money.

Google's Invalid Traffic Overview Card puts two numbers on your Campaigns overview and labels them in a way that invites one specific mistake. "Est. savings" is not money you can recover — it is traffic Google says it filtered before you were billed, so nothing was ever taken. "Credited cost" is the only figure on that card corresponding to money moving back to you. In Google's own sample screenshot the first number is roughly eight times the second, and understanding that ratio is the minimum required before either figure goes into a client report.

The card began rolling out globally this week. Within a month it will be quoted in decks by people who have not read the definitions. Here is what each number counts, what the card cannot see, and the reconciliation it makes possible but does not perform.

What the two figures count

Google's announcement, published November 12, 2026 by Keerat Sharma, VP and General Manager of Ads Privacy and Safety, frames the card as "front-end transparency into how we protect their campaigns." The billing sentence underneath is the part that matters:

"We don't charge for invalid traffic, and our multi-layered defenses have always worked in the background to catch it before an advertiser is billed. We issue a credit when it's detected later, where appropriate and possible."

Two detection stages, two numbers, and the card's help text splits them cleanly:

• Est. savings — "What Google's automated filters and proactive reviews saved during your selected date range by catching things like bots, click farms, and accidental clicks before billing."
• Credited cost — "Credits automatically applied to your account for invalid activity detected after a monthly billing cycle has closed."

The accounting is unambiguous once you read it slowly. Savings describes an invoice that was never issued. Credit describes an invoice that was issued and then partly reversed. Only the second is recoverable, and by Google's own account it is the smaller one, because the company says the vast majority of filtering happens before billing. Both figures break down by day and campaign on the Campaigns overview.

The two numbers are not the same kind of number

The last row is the whole argument. Both figures describe Google's classification of traffic. Neither describes your leads, your pipeline, or what it costs you to acquire a customer.

Three things the card structurally cannot tell you

It counts accidental clicks as protection

Google's definition of what its filters catch is "bots, click farms, and accidental clicks." Those are not the same adversary. A click farm is fraud. A thumb landing on the wrong banner is a design problem with a billing consequence. They share a bucket, and the card does not split them. So estimated savings is not a fraud metric and should never be reported as one — it measures how much traffic Google decided not to charge for, for several unrelated reasons at once. Some of those accidental clicks come from the creative itself: a tap target or a placement that produces reflexive taps shows up as savings, not as a warning.

The valuation method is not published

Google's announcement calls the figure "the amount saved"; the card calls it "Est. savings." Neither the announcement, the Google Ads Liaison thread, nor the help page states how a filtered click is priced — at the cost it would have carried in the auction it never entered, at an account average cost per click, or by some other construction. An estimate with an undisclosed method is usable as a directional signal and unusable as a line in a financial report. If a number sits next to your spend, it needs a stated basis.

A credit has conditions the card doesn't display

"We issue a credit when it's detected later, where appropriate and possible" is doing real work in that sentence. The post does not define when a credit would be inappropriate or impossible, and nothing suggests the card surfaces activity that was flagged but not credited. There is also a timing gap: credits land in a later billing cycle, so the credited figure for any window is a partial view of that window. Compare it month over month without accounting for the lag and you will read noise as a trend.

And the card reports only Google's own classification; it says nothing about leads. Google's September 2026 documentation on preventing invalid leads concedes that its filters do not stop invalid leads on their own and points advertisers at form-level checks. That admission is worth more to a lead-gen advertiser than anything on the card.

The 30% figure is a claim, not a benchmark

The announcement carries exactly one new statistic: "So far in 2026, we've reduced advertiser exposure to invalid leads by an average of 30% through continuous system upgrades." No baseline, no method, no absolute level, no description of what "exposure" means operationally — and it is a statement about leads, not about the card. You cannot reproduce it from the card, from Ads reporting, or from any combination of the two. Treat it as a vendor improvement claim: plausible, directionally good news, and not a figure to carry into a report as your own baseline.

The reconciliation the card makes possible but doesn't do

Here is where the card earns its place. Estimated savings and credited cost are two of the five columns that decide whether your acquisition cost is real. The other three live outside Google Ads.

The arithmetic is worth stating plainly, because it explains how a clean card coexists with a broken cost per lead:

Cost per verified lead = (spend ÷ platform conversions) × (platform conversions ÷ verified leads)

The first factor is the cost per conversion you already watch. The second is a counting ratio — how many of the conversions the platform claimed correspond to people your team could actually contact. Your true acquisition cost is the product of both. The invalid-traffic card moves neither term, which is why an account can show zero credited cost and still pay twice what it thinks for a customer.

So add four columns to a campaign-level export and fill them yourself:

1. Credited cost — the card's campaign breakdown for the same date range. The only Google-side number here that represents money returning.
2. Platform conversions — one rollup action per campaign, never the sum of the lead-family rows, because the same lead is often emitted under several action names at once.
3. Verified leads — contacts your CRM or sales team accepted as reachable for that campaign and window, with a creation timestamp and a source. Not form fills, not raw records.
4. Cost per verified lead — spend divided by column three, compared against your target rather than against last month.

Two results follow. If cost per verified lead is in line with plan, the card is a compliance line and nothing more. If it sits well above plan while the card shows little or nothing credited, you do not have an invalid-traffic problem. You have traffic that was valid and did not buy — a targeting, offer, or landing-page problem wearing a different costume. That distinction stops you buying click-fraud tooling to fix a message problem.

This is the discipline we apply to Meta accounts, where reported conversions and CRM-verified leads are reconciled field by field rather than argued about. The platform's number is the right signal for optimizing delivery; the verified number is the only one that belongs in a payback calculation. Traffic quality is a third question, and the four-check diagnostic we run on Meta exists because platform filters cannot answer it either.

Where the gap is widest

Lead-form campaigns that count the wrong event. If the conversion fires on form view, or on submit before any validation, the platform counts intent rather than contact. Google's filters will never flag those, because a human really did click and open the form.

Call-only and call-extension campaigns. A "conversion" there is usually a click-to-dial, not a conversation. The dial can be long, wrong, or abandoned, and nothing in the invalid-traffic layer touches that.

Broad match and Performance Max. Placement data arrives aggregated, expansion puts spend into inventory you never named, and the breakdown you need often does not exist at the grain you need. A clean card here is the least informative clean card you can get.

In all three, the failure is not invalid traffic. It is valid humans who were never going to buy.

The twenty-minute monthly procedure

1. Snapshot the card. Both figures with the exact date range, plus the daily detail. It is a window, not a ledger.
2. Export spend by campaign for the same range, ranked by spend rather than by gap size — the campaigns that matter are the ones eating budget.
3. Export conversions using the single rollup action type.
4. Pull verified leads by campaign and source from the CRM. If your CRM cannot attribute to campaign, that is the finding; fix it before reading anything else.
5. Compute cost per verified lead and set it beside the card's credited cost for the same campaign.
6. Write one line. Either "credited cost booked: none this period" or the campaigns with credit and amount. Silence reads as health; an explicit zero does not.

Then hold the decision rule. If credited cost concentrates in a few campaigns, investigate those first — Google's own system already agreed something was wrong. If cost per verified lead drifts for two months while the card stays empty, the problem is downstream of the click.

The direct answer

The Invalid Traffic Overview Card tells you what Google decided not to charge you for, and how much of a previous bill it reversed. It does not tell you whether the clicks you paid for produced customers, and it cannot, because the number you need is a ratio between a platform and a CRM that have never spoken to each other. Read the card as a billing statement. Read your reconciliation as the performance report. The first takes a few minutes a month; the second is the only one that tells you what a customer costs.

If you want that reconciliation built for an account that reports only platform numbers, it sits inside how we run digital ads and PPC management and starts with the account and the CRM read together. For the contract side — what a platform must prove versus what it merely reports — the verification clause worth putting in writing is the companion piece.