The Custom Audience Signal That Makes Performance Max Cheaper

The Custom Audience Signal That Makes Performance Max Cheaper

PMAX audience signals are the fastest way to shorten the learning phase and cut early CPA. Most accounts leave them half-configured and pay for it. Here's the setup that works.

Chris Beechey

• 10 min read
Google Ads Strategy

Most Performance Max campaigns launch with the audience signal half-built. Someone drops in a customer match list, ticks a couple of interests that vaguely relate to the product, calls it done and hits publish. Then everyone stares at the dashboard for six weeks while CPA sits 40% above target and Google tells them the campaign is "still learning".

The learning phase is real. But most of the time, the reason it's dragging on isn't Google's algorithm being slow. It's the audience signal being weak. And there's one part of the signal, custom segments, that does most of the useful work. The rest is nice to have.

If you want Performance Max cheaper, faster, this is where I'd spend the next 30 minutes.

What Performance Max audience signals actually do

Audience signals in Performance Max aren't targeting. Google's own documentation is clear about this: PMAX can and will show your ads to people well outside the signal if it thinks they're going to convert. What the signal does is tell the algorithm where to start looking.

That starting point matters more than most people realise. In the first two to four weeks of a campaign, PMAX is spending your budget exploring lookalike patterns from whoever you fed it. Feed it a broad or vague signal and it'll explore a broad or vague slice of the internet. Feed it a precise signal, and it'll narrow in on people who look like actual buyers within days instead of weeks.

Every day you shave off the exploration phase is a day you're not paying to teach the algorithm things you already knew.

The custom segment trick

Audience signals let you plug in three ingredients: your data (customer match, website visitors), interests and demographics, and custom segments. Custom segments are the one that matters.

A custom segment lets you tell Google "the kind of person I want to reach is someone who searches for these things, browses these URLs, or uses these apps." That's exactly the signal PMAX needs to skip half the exploration.

You build one under Tools, then Audience manager, then Custom segments, then New custom segment. Then you feed it two things:

  • Search terms the person would actually type into Google. Not the terms you'd bid on in a Search campaign, though there's overlap. Think about what someone researches when they're a week or two out from buying.
  • URLs of pages that those people are already visiting. This is where the trick lives. Competitor product pages, category pages, industry publication articles, review sites, "best X for Y" listicles. Google will read those pages, work out who visits them, and use that pattern as the starting point.

You can also add app package IDs and place-of-interest URLs, but URLs and search terms do 90% of the lifting.

What to put in the URLs field

This is where most people either give up (they can't think of URLs) or over-do it (they paste 50 URLs and dilute the signal). The sweet spot is 5 to 10 URLs, chosen carefully.

Include:

  • Two or three direct competitor product pages (not the homepage, the page that shows the actual product a buyer is comparing).
  • One or two industry publications your target buyers actually read. For B2B this is often trade press. For ecom it might be a "best of" comparison site or a niche blog with a loyal audience.
  • A problem-aware content URL that people visit before they know your product exists. If you sell CRM software, that might be a "how to keep track of leads in a spreadsheet" article. If you sell shutters, it might be a Houzz gallery or a home renovation forum thread.
  • The pricing page or category page from one big-name competitor that ranks for your generic terms.

Skip:

  • Your own site. Website visitors already feeds into "your data". Doubling up dilutes the signal.
  • Generic megasites like Amazon or Wikipedia. They tell Google nothing specific.
  • Any URL you've never actually visited yourself. If you can't picture your buyer reading that page, Google's not going to make it work.

What to put in the search terms field

Same logic: fewer, sharper terms beat a long list. 10 to 20 keywords is plenty.

Include:

  • Buyer-intent phrases ("best CRM for consultants", "plantation shutters Melbourne cost", "compare accounting software").
  • Category terms ("timber shutters", "cloud accounting software", "custom kitchen cabinets").
  • Problem-aware terms that come earlier in the journey ("how to track leads", "how much do shutters cost").

Skip:

  • Your own brand and any variations. That teaches PMAX nothing it doesn't already know from customer match.
  • Head terms so broad they're meaningless ("marketing", "software", "renovation"). Custom segments built on broad terms behave like no signal at all.

Two real examples

I helped an ecom client launch a new PMAX campaign for a mid-priced product line earlier this year. First attempt used the default setup: customer match, website visitors, and a couple of Google-suggested affinity audiences. Three weeks in, CPA was sitting at $124 against a target of $70, and Google kept flashing the "still learning" notice.

We paused, rebuilt the audience signal with two custom segments (one built around competitor product URLs, one around "best of" review site articles plus buyer-intent keywords), and relaunched. By week two CPA was at $58 and PMAX moved out of learning. Same product, same creative, same feed. The only variable was the audience signal.

Second example, B2B lead gen, mid-market SaaS. Their original audience signal was customer match plus "career and education" interests, which was pointing at nobody in particular. Added a custom segment with URLs of three industry publications their buyers actually read, plus branded search terms of two direct competitors. Cost per lead settled 40% lower within six weeks and lead quality (measured by qualified-to-SQL rate) went up, not down. That second bit matters, because plenty of "cheaper" audience signals also drop lead quality off a cliff.

Take those numbers with the usual grain of salt: audience signal strength varies by vertical, and I've seen accounts where the signal makes almost no difference because PMAX had enough conversion data to figure it out on its own. Bigger accounts with more history need less signal. Smaller or newer accounts, more.

What NOT to put in the audience signal

Three things to leave out, in order of how often I see them cause problems:

  • Everyone who's ever been to your site, unrefined. Website visitors as a raw pool includes bounces, competitors, students doing research, and random tyre-kickers. Segment it: past 30 days, high-value pages, cart abandoners. Or leave it out and rely on custom segments instead.
  • Broad affinity or in-market audiences Google suggests based on your website. The suggestions are usually two levels too broad ("business services", "shoppers"). They dilute more precise custom segments if you stack them together.
  • "Similar audiences" that Google auto-generates. These were phased out for most use cases in 2023 and what's left is a shadow of the old lookalike segments. Not harmful, just not helpful.

If you're staring at the audience signal builder and you're not sure whether to include something, leave it out. A tight signal beats a big one every time.

My take

The default "let PMAX figure it out" setup is Google's favourite way to spend your budget slowly. Custom segments are the only part of the audience signal that reliably earns their keep, and they're the part everyone skips because they take 20 minutes of thought rather than two minutes of clicking checkboxes.

If you're running a PMAX campaign that's stuck in learning, or spending above target for the first month, the audience signal is the first thing I'd rebuild before touching bids, budgets, creative or the feed. It's the cheapest, fastest lever in the campaign, and the one Google is quietly hoping you won't pull. Now that I've written this up, you know where the lever is.

While you're in there, check your PMAX brand exclusions too, since the brand cannibalisation trap I wrote up earlier this month usually hits the same accounts that are running weak audience signals.

Key Takeaways

  • Performance Max audience signals aren't targeting, they're a starting hint for the algorithm. The stronger the hint, the shorter the learning phase.
  • Custom segments do 90% of the useful work in the audience signal. Customer match is helpful, interests are usually noise.
  • Build custom segments with 5 to 10 carefully chosen URLs (competitor product pages, industry publications, problem-aware content) plus 10 to 20 buyer-intent search terms.
  • Leave out your own brand, generic megasites like Amazon, and unrefined website visitors pools.
  • Rebuild the audience signal before touching bids, budgets or creative if PMAX is stuck in learning or missing CPA targets in month one.
  • See Google's own documentation on audience signals for the current field-by-field click path.

If your PMAX campaign has been dragging its heels on cost, the audience signal is where I'd look first. Get in touch if you want a second pair of eyes on the setup, or grab a full Google Ads audit for a deeper look across the whole account.

← Back to Blog