How Google's ML Auto-Credit Model Replaced Manual LSA Disputes

For years, managing Google Local Services Ads (LSA) included a daily chore: opening each questionable lead, deciding whether it was a spam call, a wrong number, or a genuinely mismatched job, and filing a dispute for a credit. That workflow is gone. Around July and August of 2024, Google retired the manual dispute button for most advertisers and moved lead-quality credits onto a machine-learning system that assesses leads automatically. If you still manage LSA the way you did in 2023, you are working against a process that no longer exists.

This article walks through what changed, how the current credit lifecycle works, and where advertiser judgment still matters even though the dispute click is gone.

What the old model did

Under the legacy system, a charged lead could be disputed within a window. You clicked into the lead, selected a reason — spam, no service-area match, geo mismatch, booking never happened — and Google reviewed the request, often within a few business days. The advertiser was, in effect, an early line of quality control, and a diligent operator could reclaim a meaningful slice of spend simply by being thorough.

The downside was obvious to Google: manual disputes were noisy, inconsistent, and gameable. Two advertisers could receive the identical junk lead and one would dispute it while the other never bothered. The credit you received depended more on your diligence than on the objective quality of the lead.

What the ML auto-credit model does now

The current system flips the burden. Instead of asking you to flag bad leads, Google evaluates leads on their own signals and issues credits automatically when a lead is judged invalid. In practice there are two distinct clocks to understand:

• Assessment — roughly 72 hours. After a lead arrives, the system takes a few days to assess whether it qualifies as invalid. Nothing you do speeds this up; the model is gathering signal.
• Credit — within about 30 days. When a lead is deemed creditable, the credit lands on your account within roughly a monthly billing cycle rather than instantly.

The practical consequence is that your charged spend and your net spend can look quite different at different points in a month. A lead you were charged for on the 3rd might be silently credited around the end of the assessment window, with the money returning weeks later.

The Rate this lead survey

The manual dispute button was replaced, but advertiser input did not disappear entirely. Google now surfaces a lightweight Rate this lead survey that lets you provide feedback on quality. This is not a dispute in the old sense — you are not filing a formal claim and waiting for adjudication. Instead, your rating becomes one more signal the model can weigh, both for that lead and, over time, for the quality patterns on your account. Treat it as teaching the system, not as pulling a lever that guarantees a refund.

What is and is not creditable

The most important shift for budgeting is understanding what the model will not credit. Two categories advertisers routinely expect to recover are, as a rule, not creditable:

• Job-type mismatch. A caller who wanted a service adjacent to yours but not the exact job you booked is generally still counted as valid. The person reached the right kind of business.
• Geographic mismatch inside your served area. If a caller is within a service area you have told Google you cover, a too-far-to-drive objection typically will not earn a credit.

There are also excluded verticals. Lead-quality credits do not operate the same way for healthcare and tax categories, where Google restricts the flow of lead data. If you run LSA in those spaces, do not build a recovery strategy around credits the system is not designed to issue.

How much can realistically come back

It is tempting to assume automation means more money returned. In practice, third-party estimates put recoverable spend under the current model in the neighborhood of 6 to 7 percent of LSA cost. That is not nothing — on a serious monthly budget it is real dollars — but it is a modest correction, not a windfall. Any tool or agency promising to dispute your way to double-digit recovery is describing a world that no longer exists. The honest framing is that credits shave the edges; the bigger lever is preventing bad spend through better targeting, faster response, and tighter service-area definitions.

What advertiser judgment still controls

Even with automation doing the crediting, several decisions remain firmly in your hands and matter more than any dispute ever did:

• Service-area accuracy. Because geo mismatch inside your area will not be credited, drawing your area correctly is the difference between paying for leads you can serve and ones you cannot.
• Consistent lead rating. Feeding the survey honestly gives the model better signal about your account.
• Reconciling charged vs. credited. Because credits arrive on a delay, you need a way to track them so you measure net cost per lead, not gross.

The move to ML auto-credit is best understood as Google taking quality control in-house and asking advertisers to influence rather than adjudicate. The winners under this model are operators who stop chasing refunds and start engineering fewer bad leads.

Frequently asked questions