Quick Answer: Audience signals change how fast Performance Max finds its audience, not how good the campaign eventually gets. So the thing to measure is learning speed in the first two to four weeks, not final CPA. Give the signal a full two weeks to process, then check whether the segments converting in your audience insights overlap with the ones you seeded. If they never overlap, the signal was aimed wrong or the algorithm walked past it.
Everyone tells you how to set the signal. Nobody tells you how to check it.
Go looking for advice on audience signals and you will find a hundred versions of the same post. Seed a customer list. Add a few in-market audiences. Skip affinity audiences unless you cannot find anything better. Keep search themes tight.
All setup. Every last piece of it.
Then you fill in the asset group, launch, and three weeks later you have mediocre results you cannot explain. Did PMax fail, or did you set it up wrong? There is no report labeled "your audience signal worked." So you guess, and you carry the guess into the next campaign, and the next.
The advertisers who suffer most here are the careful ones. They spend an hour building a custom segment out of their best search keywords and competitor product URLs, then spend the following month quietly wondering whether the algorithm looked at any of it.
First, know what a signal can and cannot change
This is where most of the confusion starts.
An audience signal is not targeting. It does not fence off who sees your ads the way an audience does in a Search campaign. Google treats it as directional input, and the system routinely serves outside the segments you provide when it finds people it likes better.
What the signal actually buys you is time. A campaign with strong signals locates its converting audience sooner, which means fewer wasted weeks at the front end. A campaign with weak signals usually gets there anyway, provided you have enough conversion volume and decent assets. The ceiling is roughly the same. The road to it is shorter.
That single fact reorganizes the whole question. If you are comparing month-six CPA between a seeded asset group and a bare one, you are measuring the wrong window and you will conclude signals do nothing. Look at weeks one through four instead.
Step 1: Wait out the processing window
Google says a new signal can take up to two weeks to fully process.
Most people evaluate on day four, see nothing, and decide the field is decorative. Do not judge anything before day fourteen. If you changed the signal mid-flight, the clock restarts.
Step 2: Compare your seeded segments against what converted
Open the campaign, go to the Insights tab, and pull your audience insights. This report shows the segments Google associates with your conversions.
Now put that list next to what you actually seeded. Your customer list, your remarketing list, the in-market audiences you picked, any demographics you set.
Three things you might see:
Real overlap. The segments you fed it are showing up among your converters. The signal pointed somewhere useful.
Partial overlap with surprises. Google found adjacent segments you did not name. This is normal and usually fine, but the surprises are worth reading closely. They often describe your buyer better than your own guess did.
No overlap at all, two weeks in. Your signal described a customer you do not have. That is not a reason to delete the field. It is a reason to reseed it with the segments the report is showing you.
Step 3: Check your search themes against search terms insights
Search themes are the other half of the asset group signal and they get judged the same lazy way.
Pull the search terms insights for the campaign and compare the categories against the themes you wrote. If your themes are transactional and the report is full of research-stage queries, the themes are not steering the search side. Tighten them toward buying language and cut anything that reads like someone gathering information.
Practitioners who run a lot of PMax tend to write themes with the same discipline they apply to broad match keywords, pitched at the level of the product category instead of the individual SKU.
Step 4: Run the only test that actually proves it
Everything above is inference. If you want proof, you have to build it.
Duplicate the asset group. Same budget, same search themes, same assets. Seed one with your full audience signal and leave the other one bare. Let both run until each has enough conversions to be worth reading.
Then compare the thing signals actually affect. How long did each one take to settle into a stable CPA? Not which ended lower. Which got there first.
This costs real budget, so run it once, in an account you manage regularly. The answer tells you how much signal work is justified across everything else on your plate.
The rule that saves you an hour per asset group
Here is the payoff, and it is the part nobody in these discussions gets to.
If two weeks pass and you see no segment overlap and no difference in how quickly the seeded group settled, stop rebuilding elaborate custom segments. You are duplicating work you already did in search themes and getting credit for neither.
Put that hour into assets and theme specificity instead. Those inputs have measurable effects you can see in the asset performance report the same week.
That is not a claim that custom segments are worthless. It is a claim that you can now find out, in your account, instead of picking a side in an argument between two strangers who both manage accounts that look nothing like yours.
Where this usually goes wrong upstream
Most audience signal problems are downstream of a weaker problem. If your search themes came out of Keyword Planner suggestions, the themes are describing Google's idea of your market rather than your buyer's actual language, and no amount of audience seeding fixes that.
If that sounds like your account, our free course Stop Flying Blind walks through building a keyword and theme list from buyer language and real intent signals rather than Planner estimates. It is the input that makes everything in this post easier to read, because you are checking your themes against a list you built deliberately.
FAQ
I have fewer than 10 conversions. Can I use purchase data as a signal?
Not usefully. Customer lists generally need around 1,000 members before Google will match enough of them to matter, and a handful of purchases will not seed anything. Start with in-market audiences and a tight set of search themes, then add your data segments once the list is large enough to be worth uploading.
Are custom segments the same thing as custom audiences?
Yes. The naming changed and old advice still uses both terms, which is most of why this comes up. You build them in Audience Manager under custom segments.
If I leave search themes and audience signals blank, what happens?
The campaign leans on your feed and your assets to work out who to show ads to. It still runs. It generally takes longer to find its footing, which is the same learning-speed cost described above, and on a small budget that delay is expensive.
Should I add affinity audiences?
Most experienced PMax advertisers reach for in-market audiences, life events, and customer lists first, and treat affinity as the fallback when nothing more specific exists for the category. If you sell something narrow enough that no in-market segment fits, affinity is better than an empty field.
Does changing the audience signal restart the learning phase?
Changing an asset group signal restarts the two-week processing window for that signal. Make your changes in one pass rather than adjusting the field weekly, or you will never see a clean two-week read on any version of it.