We audited 47 Performance Max accounts. Here's the lead-quality leak in 91 percent of them.

We pulled 47 Performance Max accounts in Q1 2026. 43 of them had the same leak. The cost-per-lead in those accounts ranged from $61 to $812, and the share of unqualified leads ranged from 22 percent to 71 percent. The leak was the same in 91 percent of cases, and it had nothing to do with bad creative, broad audiences, or the placements everyone obsesses about.

The leak is the input. PMax is not broken. It is optimizing for exactly what these accounts told it to optimize for. Once you see that, the fix order changes completely, and so does the math on what you can recover.

The audit pool

We looked at 47 US accounts running Performance Max as a primary or secondary acquisition channel. The mix: 22 service businesses (medspas, home services, professional services), 14 B2B SaaS accounts under $90k MRR, and 11 e-commerce shops doing under $200k/month. Monthly Google Ads spend in the sample ranged from $4k to $90k.

Across the pool, monthly junk-lead rate (defined as forms submitted by people who could not, would not, or did not exist as a real prospect) averaged 47 percent. That is roughly half of every form fill these businesses paid for going to a contact who never had a chance of buying. Some accounts were as clean as 22 percent. Several were over 70 percent.

Two patterns predicted everything else. We will get to the second one in a moment. Here is the first.

The 91 percent leak: every form fill counted the same

In 43 of 47 accounts, the primary conversion goal was a flat form submission with no dollar value attached. A newsletter signup counted as 1. A contact form counted as 1. A 30-minute demo booking counted as 1. A "free quote" tire-kicker counted as 1.

This is the leak. PMax is a value optimizer. When you give it a flat target, it does not know that one of those events is worth 200 times more than another, so it picks the cheapest path to any of them. Cheap form fills come from the lowest-intent traffic on the open web, which means low-quality video placements, autoplay carousels on news sites, and mobile gaming apps where someone tapped the screen by accident.

The fix is not switching bid strategies. It is teaching the algorithm what "good" actually means. Assign explicit conversion values: newsletter $1, contact form $5, demo booked $30, sales-qualified call $100, closed deal $500. Then move PMax to Maximize Conversion Value, not Maximize Conversions. Google's own guidance says you need at least 15 of those events in a 30-day window before value-based bidding stabilizes, so do not panic-revert in week two.

If your account is structured around lead generation more broadly, this is also where most of the budget mistakes in our Google Ads strategy for service businesses walkthrough get amplified. PMax compounds them. A flat conversion goal in Search wastes some money. A flat conversion goal in PMax wastes a multiple of it.

The 81 percent silent miss: no offline data flowing back

In 38 of 47 accounts, no offline conversion data was flowing from the CRM back into Google Ads. Google saw the front door (form submitted) and never saw what happened on the other side (lead disqualified, demo no-show, opportunity lost, deal closed). The algorithm was learning from a feedback loop that ended at the form.

That is the actual death loop. PMax doubles down on whatever audience cluster produced the most form fills last week, and if those form fills were 70 percent garbage, it will go find more of the same garbage next week. The longer it runs, the worse it gets, because Google's machine learning is performing exactly as designed on noisy data.

The fix is mechanical. Wire your CRM stages back to Google through the offline conversion API. The cheap stack: HubSpot, Salesforce, or Pipedrive on the source side, Zapier or Make as the connector, and the Google Ads offline conversion endpoint as the destination. Send back the GCLID with the stage update (SQL, opportunity, closed-won) and a dollar value. Within 60 days the algorithm starts re-weighting, and across the accounts where we shipped this in Q1, median CPL on the qualified side dropped 34 percent by month three. None of that requires changing the ads.

If you are unsure which CRM events should be uploaded as conversions, that is a lead qualification framework question, not a Google Ads question. Decide what an SQL is, then teach Google.

Junk placements come from three sources, 80 percent of the time

Open the placements report inside any PMax campaign that has been running 60 days. Sort by impressions. The same three patterns show up in account after account:

1. Mobile gaming apps with names ending in "Run," "Crush," or "Idle." Tap-through traffic, near-zero intent.
2. Made-for-advertising sites: thin content domains that exist only to sell ad space. The "12 most surprising tax tips" listicles you click and immediately regret.
3. Low-budget YouTube channels with autoplay viewers under 13 (legally, you should not be advertising to them anyway).

In our 47 audited accounts, those three categories accounted for over 80 percent of placements with more than 500 impressions and zero conversions. Pulling the placements report monthly, filtering by that exact threshold, and pasting the offending domains into account-level placement exclusions takes about 20 minutes. Most accounts had never done it once.

Google does not love that you do this. The placements report is buried. The exclusion list caps at 65,000 entries (you will not hit it). Push past the friction and your effective CPM on the actual web drops fast.

The form is doing more damage than the ads

A one-field email capture form is a noise floor. Anyone with a phone, a thumb, and a half-second of curiosity submits one. We saw single-field forms in 19 of the 47 accounts.

Switching to a three-step form, where the first question is qualifying (budget range, company size, problem urgency) and the email field comes last, cut junk submissions roughly in half across the accounts that adopted it. Add reCAPTCHA Enterprise on top, blacklist the obvious throwaway email domains for B2B accounts (gmail, yahoo, hotmail, mailinator), and add a "what is your monthly ad spend" or "what is your annual revenue" question that filters the bottom-tier inquiries before they hit your CRM.

The named tool everyone forgets: Google's own Lead Form asset inside PMax supports qualifying questions natively. You do not need a third-party form builder for the first two steps. You do need to actually use the qualifying-question slots, and most agencies set up the asset, take a screenshot for the deck, and never come back.

The branded-search cannibalization nobody mentions

PMax will steal your branded search traffic and report it back to you as a PMax conversion win. We found this happening in 31 of 47 accounts. People are typing your company name into Google, getting served a PMax shopping or display ad, clicking, converting, and being attributed to PMax instead of branded search.

The CPL math is the giveaway. If your branded-search clicks cost $0.80 and your cold PMax clicks cost $40, lumping them together makes PMax look better than it is and starves your real cold-traffic spend of attribution clarity. Branded conversions are not incremental. You would have closed those people anyway.

The fix is two minutes. Add brand exclusions in Google Ads under "Account-level brand exclusions" and exclude your own brand terms (and any obvious misspellings). PMax stops eating your branded inventory. Your branded search campaign goes back to looking like the cheap, high-converting workhorse it actually is. Your PMax CPL gets honest, which is uncomfortable for a week and useful forever after. Without that honesty, your attribution model cannot tell you anything real.

The day-one fix order, ranked by recovered CPL

If we walk into a stalled PMax account tomorrow, this is the sequence and roughly what each step is worth in our 47-account sample:

1. Set conversion values per goal and switch to Maximize Conversion Value. Biggest single lever. Recovers 30 to 50 percent of the CPL gap on average.
2. Wire offline conversions back from your CRM. Compounds the first fix. Median 34 percent qualified-CPL drop by month three.
3. Run the placement report and add account-level exclusions. Fast win. Reclaims 8 to 15 percent of media efficiency.
4. Tighten the form: multi-step, reCAPTCHA Enterprise, qualifying questions. Cuts junk submissions roughly in half, which improves the data feeding fix #2.
5. Add brand exclusions. Two-minute job that makes every other number you see honest.

The order matters. People skip to fix #3 because the placements report is the most visible and feels like progress. Doing it first, without the value-weighted goals and offline conversions in place, only delays the inevitable: the algorithm will find new junk to optimize toward.

Give it 30 to 60 days after the first two changes ship. Performance Max is not a knob. It is a learner with no taste and a generous budget. Tell it what good looks like, send the bill back home, and it gets out of its own way.

If you want a performance marketing agency to run this audit on your account before your next quarterly review, that is the work we do. Same checklist, same order, same 30-60 day window.