PERFORMANCE MAX SEARCH TERM INSIGHTS REPORT GOOGLE ADS

The complaint I heard for two years was that PMax hid the queries. It didn't, quite it showed them grouped, which is worse, because grouped data looks like an answer. Then the raw terms arrived in 2025, and the complaint became silence: people opened the report, exported it, and went back to guessing.

That's the failure mode I want to attack. The search terms report in performance max google ads is not a smaller version of the Search one. It answers a different question, from a different sample, and it rewards a different kind of reading. Treat it like a Search query report and you'll make three specific mistakes I'll name below.

Two reports, not one: insights vs raw terms

People conflate them constantly, so start here. The categories view lives in Insights: search categories are auto-generated groupings of the search terms driving traffic to your campaigns, based on customer intent and product and service attributes. The raw view is separate under 'Insights and reports' in the Campaigns menu, from the drop-down you choose 'Search terms and landing pages for Performance Max' Google Ads Help on the PMax search terms report.

The distinction that actually matters

It's not resolution, it's coverage. Google says the categories take all search terms into account, including those not exposed in the search terms report due to privacy reasons. So the aggregate view sees traffic the granular view legally cannot show you. Two reports over overlapping-but-unequal populations, which is exactly why performance data is processed differently between the two, which can result in minor differences. Stop reconciling them. They aren't wrong; they're answering different questions.

I've sat in more than one meeting where a client's analyst had spent a day building a variance report between the two views, and the honest conclusion is that the variance is the design, not a bug. The moment you accept that, both reports get more useful, because you stop asking each of them for the other's job.

Which one I open first

Categories, always and I mean it as a diagnosis, not a preference. It's the only place the whole demand shape is represented. Raw terms come second, to answer a question the categories already made me ask. Reverse the order and you'll spend a morning reading individual queries with no idea whether they're 2% of the campaign or 40%.

The other reason for the order is psychological, and I'd rather name it than pretend I'm above it. Raw queries are compelling in a way aggregates aren't one absurd search term will hijack a meeting for ten minutes regardless of whether it cost eleven cents. Opening with the categories forces the room to argue about proportions first, which is the only argument worth having.

How much do you really see, compared with classic Search

Here's where the honest accounting matters, because the answer is: less than Search, and less than you feel like you're seeing.

Both reports have holes, in different places

Classic Search hides low-volume queries for privacy that's old news, and it's why a Search query report never sums to campaign cost either. PMax inherits that, then adds its own gaps. In the raw export, some clicks have no search term attached at all. The category labels are built from a rolling recent window, so a label can name a term that stopped serving weeks ago. And the historical floor is March 2023 no earlier data exists, no matter how old the account is.

That last one bites in a specific way nobody plans for: any pre-2023 seasonal comparison is simply unavailable, so the first Black Friday you analyse with this report has no baseline and the second one has exactly one. Structural questions need years, and this report is still growing into its own history.

The comparison people get backwards

The instinct is "PMax shows me less, so I trust it less." Wrong direction. In Search you're seeing a censored sample of queries matched to keywords you chose. In PMax you're seeing a censored sample of queries matched to nothing you chose the model inferred them from your feed, your assets, your landing pages and your signals. The Search report tells you how your keywords are performing. The PMax report tells you what the model thinks you sell. That second thing is far more useful and almost nobody reads it that way.

Put it in reviewing terms. A Search query report grades your own homework: you wrote the keywords, the report says how they did. The PMax report is closer to reading a stranger's summary of your business back to you. When the summary is wrong, the interesting information isn't in the queries it's that your inputs produced that summary. No amount of exclusion work fixes a bad summary; it just deletes the evidence of it.

What's missing that hurts most

No asset group breakdown, and depending on view no cost per term. Anyone who's built a habit around a query report is used to sorting by spend, and it isn't there. Google's guidance is explicit about the reason: prioritise conversion data over clicks and impressions, since some conversions may occur without direct clicks or impressions, which can lead to CTRs or CPMs that don't reflect the true value. You're being told to stop reading it like a Search report, and the missing columns enforce it.

Which is irritating and mostly correct. In a campaign that serves across YouTube and Display, per-query CTR is a comparison between things that aren't alike, and a "bad" CTR on a term may say more about the surface than the query. Losing the column costs a habit; keeping it would have cost accuracy.

The API caveat, briefly

Anyone planning to automate this should know it doesn't behave. PMax terms aren't in the standard search term view; they live in a per-campaign insight resource that must be queried one campaign at a time. Budget more engineering than you expect, and see what the channel report already tells you about surfaces before you build anything.

How I read a query cluster

The categories are the analytical unit. Reading them well is a specific skill, and it's mostly about resisting the obvious inference.

Share first, performance second

My first pass ignores every metric except relative size. Which categories are big? The performance max search term insights report google ads produces is a demand map before it's a performance report, and the shape of the map is the finding. A category you didn't know existed taking 15% of your clicks is more interesting than your best-converting one, which you could have named from memory.

The three shapes I look for

Expected and big: fine, note it, move on. Unexpected and big: the model found demand you weren't pricing, and the question is whether that's opportunity or drift. Expected and absent: the most alarming one, and the one people never notice you can't spot a missing row by scanning present rows. I keep last quarter's category list open next to this quarter's for exactly this reason.

The absent category is worth dwelling on, because the diagnosis branches sharply. Either the demand isn't there, which is a market fact and not your problem to fix or the demand is there and the model isn't reaching it, which is an input problem sitting in your feed or your assets. The report won't distinguish those for you. A quick look at what the keyword planner says about volume for that theme usually settles it in five minutes.

The subcategory is where the intent lives

Top-level categories are almost always too coarse to act on. A category can look mediocre in aggregate while containing one subcategory that converts at triple the rest and three that convert at zero. The aggregate is the average of a good business and a bad one. Every real decision I've made from this report was made at subcategory level, on the entity breakdown that shows which campaigns and asset groups the category appeared in.

The entity column is the part people skim, and it's often the punchline. When a single category shows up across three asset groups, that's not a query insight it's a structure problem wearing a query costume, and it means the groups aren't as distinct as the naming convention suggests.

The trap: category ≠ audience

A category is a description of queries, not a segment you can target. You cannot bid on it, exclude it, or budget against it. It's evidence, and the action always happens somewhere else in the feed, the assets, or a negative list. Confusing the two produces the meeting where somebody asks to "turn off" a category and the room agrees before anyone checks that the button exists.

What I actually do with it: negatives

The obvious use, and the smaller one. Google frames it as an enhanced negative keyword strategy: identify irrelevant search terms, add them as negatives to prevent your ads showing for those searches. True and worth doing but it's the floor.

Where the exclusion has to be applied

Coverage isn't uniform and this is the operational detail that catches people. For Search and Shopping inventory, apply negative keywords at the account or campaign level; for display and video inventory, use excluded content keywords in the Content Suitability Centre. One report, two exclusion mechanisms, depending on where the query fired. Add a negative keyword and assume you've handled it everywhere, and you'll come back in a month to find the same theme still spending on the video side. It's the same asymmetry that runs through the overlap between PMax and Search: reporting and control don't share a boundary.

The three I always cut, and the many I don't

Competitor brands, adjacent industries that share vocabulary with mine, and job-seeker or how-to queries where the intent is unambiguously not commercial. That's a short list on purpose. PMax works the funnel by design, and a chunk of what looks like waste is upper-funnel inventory doing its job at a delay. I've watched accounts negative their way to a clean report and a smaller business.

The self-check I apply before adding anything: can I explain why this query can never convert, without using the word "probably"? Adjacent-industry terms pass a query about a product I don't sell is a fact. Low-intent research terms usually fail, because "hasn't converted yet in a last-click window" and "cannot convert" are different claims and only one of them justifies a permanent exclusion.

The clue that a bad cluster isn't a keyword problem

Here's the inference I'd most want people to take from this article. A cluster of genuinely irrelevant queries is rarely a targeting accident. It's the model reading your inputs and concluding, reasonably, that you sell that thing. When irrelevant traffic clusters, my first suspect isn't the negative list it's a product title, a description, or a landing page that says something I didn't intend it to say. Negatives treat the symptom, and the symptom regrows.

The tell is the shape. Scattered nonsense across unrelated themes is ordinary matching noise and negatives handle it. A tight, coherent cluster of wrong-but-consistent queries is an input problem: something in your account is describing a business you're not in, clearly enough that the model built a whole category around it. Exclude that cluster and you've hidden the mirror without changing the face.

The bigger payoff: feed and assets

The report's real value runs the other way. Google puts it plainly in the benefits refined assets: discover new relevant search terms performing well and use this to create new text assets geared towards your highest-performing search terms and the categories view goes further, recommending you tailor creatives, landing pages, and Merchant Center feed descriptions to incorporate top search categories surfaced or trending.

In e-commerce, titles are the keywords

This is the whole game. PMax has no keywords; the product title is the closest functional equivalent, and it's the input the model matches queries against. So a converting subcategory whose vocabulary appears nowhere in my titles is a direct instruction: the language exists, the demand exists, my feed isn't speaking it. Rewriting titles toward that vocabulary moves more than any negative list I've ever built which is why the return on optimising the feed compounds where negatives just cap losses.

The discipline is to take the customer's word, not the catalogue's. Manufacturers name products; customers describe them, and the two vocabularies drift apart quietly over years. A converting subcategory is a free, evidenced list of the words your customers actually use the kind of research people pay for, sitting in a report they closed after ninety seconds.

For lead gen, the same logic on assets

No feed, same principle: headlines, descriptions, search themes and landing page copy are the input. Take the converting subcategory's actual phrasing into headlines not paraphrased, the phrasing. And search themes are the one place you can push proactively toward a cluster rather than away from one, which is why how the asset groups are structured determines whether the finding can be acted on at all. If every product sits in one group, a subcategory-level insight has nowhere to go.

My cadence, and the honest ceiling

Monthly, excluding the last two weeks for conversion lag, and I write one sentence per significant category explaining why it's there. The sentence is the artifact twelve of them is a year of history and the only way to tell drift from design. The ceiling: this report shows you what the model heard, not what it will do next. You get evidence about your own inputs and a modest set of defensive controls. That's less than the Search report and, read properly, considerably more useful.