Performance Max for Lead Generation in 2026: A Practical Setup
Nearly every Performance Max guide is written for retailers: feed, margin, labels, ROAS. But a huge share of advertisers sell appointments, quotes and demos, not products. For them Performance Max for lead generation plays by different rules — there is no feed, sales feedback is thin, and the algorithm will happily deliver a hundred cheap form fills that your sales team cannot do anything with.
Here is a working setup for Performance Max for lead generation in 2026: which conversion you should actually optimise toward, how to keep junk out, which settings matter, and when the honest answer is that PMax is the wrong campaign type for you right now.
Why lead gen breaks more often than e-commerce
Retail gives the algorithm hard feedback: a purchase, a value, sometimes a return. Lead gen gives it “someone submitted a form.” That is enough to start optimising — and the system will faithfully find people who enjoy submitting forms. The trouble is that a submitted form and money are two different events, sometimes barely related.
Three failure modes follow:
- Cheap leads, no revenue. The dashboard looks great, sales is furious. The algorithm optimised for exactly what you gave it.
- Spam and bot submissions. PMax reaches surfaces where forms get filled reflexively or automatically.
- Brand cannibalisation. Left unrestricted, PMax will scoop up your own branded demand and take credit for it.
All three are prevented before launch, not patched afterwards.
Step 1. Decide what counts as a conversion
This is the decision that defines the campaign, and it is made in your funnel, not in the Google Ads interface. Name the event that sits closest to revenue while still happening often enough for the system to learn from.
| Event | Volume | Link to revenue | When to make it primary |
|---|---|---|---|
| Form submission | High | Weak | Only at launch, before deeper data exists |
| Qualified lead | Medium | Good | The default choice for most businesses |
| Meeting or demo held | Low | Strong | When volume allows and the cycle is short |
| Closed deal | Very low | Direct | Long cycles: use as a value source and secondary goal |
Google’s own guidance for value-based strategies is to pick a goal close to a closed deal that has at least 15 conversions in the last 30 days. If your “qualified lead” fires five times a month, move one step up the funnel for your primary goal and keep the deeper action as a secondary one for observation.
Which actions belong in the primary bucket and which should stay secondary is the single highest-leverage setting in a lead gen account — covered in detail in primary and secondary conversions in Google Ads.
Step 2. Send quality back, not just volume
If Google only ever learns “a form was submitted,” it will scale form submissions. If it learns what a lead was worth, it starts hunting for people who look like buyers.
- Separate conversion actions per funnel stage. At minimum “lead” and “qualified lead,” with the latter as primary once volume supports it.
- Offline conversion import from your CRM. Push “qualified” and “closed won” back using GCLID or hashed customer data — the full mechanics are in offline conversion import in Google Ads.
- Values instead of counts. Assign each lead a predicted value — lead score, segment average order value, close probability × margin — and switch to Maximise conversion value, with a target ROAS once data is stable.
None of this works on a dirty base signal. Without enhanced conversions you lose attribution coverage, and without auditing duplicate conversions you may be scaling an inflated number and wondering why real CPA is twice the reported one.
Step 3. Build the campaign without a feed
A feedless PMax campaign is an asset group plus audience signals. What matters at this stage:
- One asset group, one offer. Don’t mix services with different economics — the system will shift budget to whatever is cheapest per conversion, not most profitable.
- Maximum variety inside the group. Headlines covering different motivations (price, speed, guarantee, expertise), multiple descriptions, images in every required ratio, logos.
- Upload your own video. If you don’t, the system generates one from your assets. A plain 15-second phone-shot clip with captions almost always beats it. For feed-only setups you can ask your account team to opt out of auto-generated video entirely.
- Treat Ad Strength as a hint. Google reports that advertisers reaching “Excellent” see roughly 6% more conversions on average — a correlation worth chasing with real assets, not with filler headlines.
- Audience signals are a suggestion, not targeting. Upload customer lists and CRM segments — see Customer Match and first-party data. PMax will still reach beyond the signal, and that is by design.
- Lead form assets: optional, with a caveat. They lift volume and cut cost per lead, and they reliably lower lead quality. If you use them, add qualifying questions.
If you also run Search, decide upfront who gets which query — the conflict is mapped out in campaign cannibalisation between PMax, Search and Shopping. Short version: exact-match keywords in Search win over PMax, broad ones do not.
Step 4. The settings that actually move lead quality
- Turn Final URL Expansion off at launch, or the system will send traffic to your blog, careers page and company story. Context and exceptions: final URL expansion in PMax and AI Max.
- Set location targeting to “Presence” rather than “Presence or interest.” For service businesses this is the difference between local enquiries and enquiries from another continent.
- Negative keywords at campaign and account level. They work in PMax, but search terms reporting lags, so don’t expect immediate feedback.
- Brand exclusions if you don’t want PMax harvesting your own brand demand.
- Placement exclusions. The placement report routinely surfaces mobile games and junk apps; for PMax these lists often need to be applied at account or manager level — see placement exclusions and brand safety and inventory modes.
- Form protection. reCAPTCHA, server-side validation, blocking disposable domains and obviously fake numbers. Submissions you would never pay for should never become conversions in the first place.
- Qualifying questions. Budget, timeline, location, project type — one or two questions typically cut raw volume by 20–40% and raise the qualified share. Those numbers are directional; measure your own.
Step 5. Budget, bidding and the learning period
- No value data, thin volume → Maximise conversions with no target. Let it gather data.
- Stable CPA → add a tCPA close to your maximum acceptable cost per lead. A target set too low does not make leads cheaper; it starves impressions.
- Lead values available → Maximise conversion value, adding target ROAS once the data holds. Keep the target realistic — an inflated tROAS kills volume.
Size the budget from the goal, not the other way around. A daily budget that cannot produce one to two target conversions per day turns learning into a permanent state. If your target CPL is $50 and your budget is $30 a day, the campaign will never stabilise.
On learning itself: Google advises leaving the campaign alone for one to two weeks, and up to six weeks for complex setups or low conversion volume. Which edits reset the process is listed in the Smart Bidding learning period. A practical habit: keep a change log and apply edits in weekly batches instead of daily nudges.
What you can actually afford per lead is a spreadsheet question, not an interface one — the method is in media buying unit economics.
Step 6. The weekly review
- Qualified lead share of total volume — the campaign’s real KPI. If it falls while volume rises, you are scaling junk.
- Channel reporting. How much went to Search, video and display: PMax channel reporting shows where the budget actually went.
- Search terms and placements — feed your negative and exclusion lists.
- Landing pages. Cheap leads with poor qualification usually point at the page, not the traffic: the landing pages report.
- Assets. Swap the weakest performers every three to four weeks without rebuilding the campaign. To compare two creative sets properly, use asset experiments.
Structure: one campaign or several
The first question after launch is always whether to split. There is no universal answer, but there is a working rule: split by economics and by data, not by preference. A separate campaign is justified when the acceptable cost per lead differs materially or the conversion goals differ. Otherwise splitting just dilutes volume and stretches the learning period.
| Situation | Call | Why |
|---|---|---|
| Two services with 3× different CPL | Separate campaigns | One strategy averages the targets and starves the expensive service |
| One service, five cities, same economics | Single campaign | More data, faster learning |
| Markets with different lead costs | Separate campaigns | Budget otherwise drifts to the cheapest market |
| B2B and B2C segments | Separate campaigns | Different conversions, creative and sales cycles |
| Under 30 conversions per month in total | Single campaign | Splitting guarantees permanent learning mode |
Inside a campaign, split with asset groups — by offer, not by keyword theme. PMax has no keywords, and recreating a Search campaign structure here buys you nothing.
A simple lead value model
You do not need sophisticated scoring to move to value-based bidding. One table, built in an hour with your sales team, is enough: for each lead type, the close rate and the average margin.
The formula is lead value = close probability × average deal margin. That number is what you pass to Google as conversion value.
| Lead type | Close rate | Average margin | Lead value |
|---|---|---|---|
| Quote request with stated budget | 18% | $1,200 | $216 |
| Quote request, no budget | 6% | $900 | $54 |
| Price list download | 2% | $800 | $16 |
| Call longer than 60 seconds | 12% | $1,000 | $120 |
The figures are illustrative — run your own. But the effect is consistent: an algorithm that sees $216 next to $16 stops chasing count and starts hunting for the first group. Even a crude model beats treating every submission as equally valuable.
One caveat: revisit the coefficients quarterly, and avoid changing them abruptly mid-learning. To the campaign that is the same shock as switching bid strategies.
What a six-week rebuild looks like
A typical sequence when the campaign runs but lead quality is poor:
- Week 1. Conversion audit: what is counted, whether duplicates exist, which actions are primary. Demote everything that isn’t a real business action. Enable enhanced conversions.
- Week 2. Form protection, qualifying questions, lead status captured in the CRM. In parallel: turn off final URL expansion and fix location settings.
- Weeks 3–4. Accumulate status data. Don’t touch the campaign. Keep a list of pending edits without applying them.
- Week 5. Connect offline conversion import, switch the primary conversion to qualified lead, apply the batched edits.
- Week 6. New learning period. Compare cost per qualified lead against the baseline, not raw CPL.
The trap in this plan is judging it in week three. When you switch the primary conversion, reported CPL almost always rises — you started counting a rarer event. React to that rise by changing settings and you are back where you started.
In lead generation, Performance Max optimises for exactly the event you called a conversion. Fixing lead quality is not a fight with the algorithm — it is correcting what you handed it.
If there’s no CRM to export from
Common in small businesses: enquiries land in an inbox and a messenger, and there is no system of record. Full offline conversion import is off the table, but you can still send quality signals in a simplified form.
- Different forms, different conversions. “Request a quote” and “subscribe to updates” should never be the same event.
- Call duration as a quality proxy. A call over 60 seconds is almost always worth more than a 20-second one, and that is configurable in call tracking without any CRM.
- A qualifying question as the splitter. “Budget under $5,000” and “budget over $30,000” can fire as separate conversion actions with different values.
- Weekly manual marking. Even a spreadsheet where a salesperson marks “qualified / not qualified” gives you, after a month, the basis for a GCLID-based conversion upload.
None of this is elegant, but any of it beats an algorithm that sees every enquiry as identical.
Five mistakes that show up again and again
- Cutting tCPA when leads get expensive. That throttles impressions and resets learning. The right response is to check whether traffic quality slipped and work the exclusion lists.
- Swapping assets every other day. The campaign never exits learning and lives there permanently.
- Counting every micro-action as a conversion. Phone clicks, scroll depth, contact page views — all of it teaches the system to find people who scroll, not people who buy.
- One campaign for five services with different margins. Budget migrates to whatever is cheapest per conversion.
- Running PMax instead of Search rather than alongside it. Precise commercial intent is cheaper and more predictable to buy in Search.
When Performance Max for lead generation is the wrong campaign type
- Under 15–30 conversions a month account-wide. Start with exact-match Search; statistics are cheaper to accumulate there.
- No way to send quality back. If your CRM cannot export even a lead status, PMax will optimise toward spam.
- A very narrow geography or a B2B market of 200 companies. Broad reach works against you.
- No team to work leads quickly. Volume arrives in waves, and unworked leads destroy the economics regardless of your settings.
Launch checklist
- Pick a primary conversion close to revenue; everything else secondary.
- Enable enhanced conversions and audit for duplicates.
- Connect CRM import for at least the qualified status.
- Turn off final URL expansion; set location targeting to Presence.
- Upload your own video and a full asset set.
- Add audience signals from first-party data.
- Protect the form and add qualifying questions.
- Choose the bid strategy by data availability; budget for one to two target conversions a day.
- Leave it alone for two weeks, then review in batched changes.
If the bottleneck is account infrastructure rather than campaign settings, that is a separate track: Google Ads agency accounts and PPC Rebels services. Today’s companion pieces: free product listings for retailers and DV360 or Google Ads for teams hitting a reach ceiling.
FAQ: Performance Max for lead generation
Can PMax run without a product feed?
Yes. For lead generation the campaign is built on an asset group and audience signals. A feed only adds retail surfaces, which service businesses don’t need.
Which conversion should be primary if deals close in two months?
The one closest to revenue that still gets enough volume — usually the qualified lead. Closed deals come back as a value source and a secondary conversion.
How many conversions does PMax need to be stable?
Google’s benchmark for value-based strategies is at least 15 conversions on the chosen goal in 30 days. Below that, move your goal up the funnel or stay on Search.
How do I stop spam submissions?
Three layers: form protection (reCAPTCHA, server-side validation), qualifying questions, and — most importantly — not counting those submissions as conversions so the algorithm never learns from them.
Should I use lead form assets?
They raise volume and lower CPL, and they usually lower quality compared with a landing page form. Test them separately and always with qualifying questions.
Why is PMax eating my brand searches?
Because brand conversions are cheap and the algorithm takes the cheapest path. Use brand exclusions, or run a dedicated branded Search campaign with tight match types.
Leads are cheap but nothing closes — what do I change?
The optimisation target. Send qualification status or lead value back to Google. While it only sees “form submitted,” that is all it will deliver.
How much control do I have over placements?
Partial: placement and brand exclusions, negative keywords, content suitability and location settings. Full placement control is not something PMax offers — that is the trade for automation.
When can I judge results?
No earlier than two weeks, and up to six when conversion volume is low. Decisions made on the first three days are almost always wrong.
Should Search and PMax run at the same time?
Yes, with a clear split: exact and phrase keywords in Search, PMax for reach and adjacent demand. Otherwise they compete for the same queries.
Do I need a separate tCPA per service?
If the economics differ, yes — and ideally separate campaigns. One strategy spanning multiple funnel stages or margin profiles performs poorly.
