Search Themes in Performance Max 2026: How to Use Them Without Breaking Your Search Campaigns
Performance Max spent years as a black box: you handed over assets, a feed and a budget, and guessed where the search traffic actually went. Search themes in Performance Max shifted that balance. They are the one direct lever an advertiser has to tell the system “look over here as well” — and the fastest way to break a well-structured account by pulling traffic away from Search or diluting an asset group into meaninglessness.
This guide covers how search themes in Performance Max behave in 2026: how they differ from keywords, how many to use, how they interact with keyword priority rules, what search category reporting will and will not tell you, and how to prove whether themes added incremental conversions or merely moved them between campaigns.
What search themes in Performance Max are — and are not
A search theme is a phrase you add to an asset group to describe user intent. Technically it is an additional input to the system’s query matching, not a targeting condition.
Three differences from keywords that drive every downstream decision:
| Property | Keyword in Search | Search theme in PMax |
|---|---|---|
| Role | Entry condition for the auction | A hint that expands where the system looks |
| Match types | Exact, phrase, broad | None; behaviour resembles broad |
| Bidding | Can be set per keyword | Controlled entirely by campaign strategy |
| Reporting | Full search terms report | Aggregated search categories |
| Exclusions | Negatives at every level | Campaign- and account-level negatives only |
The practical takeaway: a theme is not a way to lock in a phrase. If you need hard control over a specific query, that query belongs in a Search campaign.
A search theme answers “where else should the system look”, not “where must I appear”. Expecting exact-match behaviour is the single biggest source of disappointment with PMax.
How themes interact with your Search campaigns
The priority logic in 2026 works like this: when a user’s query exactly matches an eligible keyword in your Search campaign, Search wins. In every other case, normal ranking decides, and PMax can take the query.
Search themes strengthen PMax specifically on those non-exact matches. Configure them carelessly and you get a reshuffling effect: total conversions stay flat while reporting shifts toward PMax, where attribution looks more generous. That is textbook PMax and Search cannibalisation, and adding themes makes it sharper.
Do this before you add a single theme:
- Confirm your priority commercial queries are covered by exact match in Search and those keywords are actually eligible — not “low search volume”, not disapproved.
- Record a baseline: conversions, CPA and impression share across Search for two to four weeks.
- Only then add themes, and watch the sum of both campaign types rather than PMax alone.
How many themes, and which ones
The limit is 25 search themes per asset group. That is a ceiling, not a target. Filling it almost always dilutes the group, because one set of assets ends up serving wildly different intents.
A practical starting range — a directional benchmark, not a rule — is 5 to 12 themes per asset group, all within one semantic cluster. If you genuinely need 25, you need two or three asset groups instead.
Themes that tend to earn their place:
- Queries missing from your Search coverage. Cross-reference the PMax search category report against your search terms report; converting language you have no keyword for is a strong candidate.
- New products and seasonal lines with no campaign history yet. The theme substitutes for data the system has not accumulated.
- Customer vocabulary that differs from your internal product naming.
- Regional and colloquial variants if the business is location-based.
Themes that waste budget:
- Brand terms — they arrive anyway, and you risk pulling cheap brand traffic out of Search and muddying its attribution.
- Very broad phrases (“services”, “buy online”) that push discovery toward irrelevant queries.
- Duplicates of terms already performing on exact match in Search.
- Themes with no supporting headlines, images or landing page. The promise in the theme must match what the user actually lands on.
Setting them up
- Open the PMax campaign → Asset groups → the group → Edit.
- In the audience signal block, find Search themes.
- Add phrases one at a time. Each should express one complete intent, roughly two to five words — not comma-separated lists inside a single theme.
- Save and leave the campaign alone for at least two weeks. Anything shorter measures transition noise.
Note that search themes and audience signals both live at asset group level but are entirely different mechanisms with different effects. How the second one works, and why it is not targeting, is covered in our guide to optimised targeting and audience signals.
Where good themes come from: five sources
Inventing themes at a whiteboard is the weakest approach — you end up reproducing internal product vocabulary that customers never use. Sources that produce themes worth adding:
- Your Search campaigns’ search terms report. Filter for terms with conversions that have no exact-match keyword behind them. That is a ready-made list of demand you are currently serving blind.
- The PMax search category report itself. Categories with conversions confirm real demand; categories with impressions and zero conversions are negative keyword material, not theme material.
- On-site search. People search your site with the same words they use on Google. An export from analytics surfaces phrasing no keyword planner will give you.
- Support and sales conversations. A customer’s opening sentence in chat is very close to a finished theme — especially valuable for services with opaque category names.
- Marketplace questions and reviews, if you sell there too: buyers describe the job, not the product.
Run the candidate list through two filters. First: does the asset group contain a headline and an image that speak to this intent? Second: does the landing page answer this query? A theme with no supporting creative is a paid click with a predictable bounce.
How to phrase a theme
Phrasing controls how loosely the system interprets your hint:
| Phrasing | How it is read | Verdict |
|---|---|---|
| “repair” | Extremely broad, any adjacent category | Poor |
| “washing machine repair” | The category as a whole | Workable |
| “same day washing machine repair at home” | A narrow, specific intent | Good for a dedicated group |
| “repair, washers, urgent” | A comma list — dilutes itself | Poor |
The sweet spot is two to five words expressing one complete intent, with no lists. Overly long themes (seven words and up) underperform too: the system rarely finds a close match and the hint simply never fires.
Reading the reporting you do get
There is no full search terms report in PMax. There is a search categories report that buckets queries into themes and shows impressions, clicks and conversions per bucket. It is aggregated, but it is enough for three jobs:
- Find waste. High-volume categories with zero conversions are your negative keyword candidates at campaign level.
- Find new demand. Converting categories absent from Search are candidates for both new Search campaigns and new themes.
- Validate a theme. If a category matching your new theme did not grow after you added it, the theme was redundant — the system was already finding that demand.
To compare quickly, build a single tracked metric using custom columns and the report editor: PMax conversions as a share of PMax + Search conversions, by week. That one number catches cannibalisation faster than any dashboard.
Negative keywords: your only brake
In PMax, negatives apply at campaign and account level. That is the throttle when themes pull in junk. In practice:
- Maintain an account-level junk list: “free”, “diy”, “jobs”, “salary”, “reviews of employer”, competitor brands you do not want.
- Keep a campaign-level list per offer: models you do not stock, cities outside your delivery zone, adjacent services you do not provide.
- Do not build a ten-thousand-line list “just in case” — you will cut profitable long tail. Add negatives from the category report, not from imagination.
The selection methodology is the same one described in our piece on the search terms report and negative keywords; only the level of application changes.
Structure: one campaign or several
A recurring question is whether to split themes across asset groups or across campaigns. A workable decision table:
| Situation | Do this | Why |
|---|---|---|
| Different intents, same budget and goal | Separate asset groups | Learning data stays pooled |
| Different target CPA/ROAS | Separate campaigns | Goals are set at campaign level |
| Different markets with different economics | Separate campaigns | Needs independent budget control |
| Different product lines, same economics | Asset groups plus feed segmentation | Easier to manage and read |
Splitting campaigns “for control” is a common trap. Each campaign needs its own conversion volume to learn. If a split leaves a campaign with only a handful of conversions per month, you have traded stability for a false sense of control. Why data volume matters is covered in the Smart Bidding learning period.
Proving lift instead of assuming it
You add themes, PMax conversions rise. That proves nothing on its own — those conversions may have migrated from Search or from organic. A defensible test:
- Baseline. Two to four unchanged weeks, recording total conversions across all brand campaigns, not just PMax.
- Experiment. A 50/50 split through Google Ads experiments.
- Decision metric. Incremental conversions and blended CPA at account level, never PMax share.
- Duration. At least one full conversion cycle plus a week. For long sales cycles, align with your conversion window or you will undercount delayed sales.
- No parallel edits. Freeze budgets, feed and assets for the duration.
If there is no incremental lift but PMax share went up, the themes acted as a redistribution pump. That is not a reason to switch PMax off; it is a reason to restore exact-match coverage in Search and tighten the themes.
What redistribution looks like in numbers
An illustrative example, because the trap is easier to see with figures. One account, Search plus PMax, shared budget, 30 days before and after adding 18 themes.
| Metric | Before | After | Reading |
|---|---|---|---|
| PMax conversions | 210 | 318 | +51% — looks excellent |
| Search conversions | 340 | 241 | −29% — rarely looked at |
| Total conversions | 550 | 559 | +1.6% — inside the noise band |
| Total spend | $40,000 | $42,500 | +6.2% |
| Blended CPA | $72.70 | $76.00 | 4.5% worse |
The PMax view reports a 51% gain while the business pays more for the same number of sales. The trap is that PMax tends to claim the conversion under prevailing attribution rules, so the dashboard looks convincing. The only defence is to read totals and blended CPA rather than per-campaign numbers.
The fix in this scenario: remove the broadest themes — typically three to five of them account for most of the leakage — confirm priority queries are covered by exact match in Search, and re-measure. If total conversions hold while spend falls, you have taken control back.
Seven mistakes that show up repeatedly
| Mistake | Consequence | Fix |
|---|---|---|
| 25 themes “to cover everything” | Diluted group, irrelevant impressions | One cluster, 5–12 themes |
| Brand terms as themes | Brand traffic drains out of Search | Remove brand from themes |
| Themes with no matching assets | Low CTR, poor conversion rate | Align headlines and landing page |
| No campaign-level negatives | Spend on irrelevant categories | Build the list from the category report |
| Judging by PMax share | Illusory growth | Measure account totals |
| Edits every two or three days | Campaign never exits learning | Two-week quiet window |
| Themes instead of feed hygiene | Serving products with no stock or price | Fix the feed first |
Themes, feed and assets are one system
A search theme expands demand, but the whole asset group has to serve it. If a theme pulls in a query with no matching product in the feed, no matching headline and no relevant landing page, you have bought a click that could never convert.
A fifteen-minute coherence check:
- List the themes in the asset group.
- For each, note whether a headline speaks directly to that intent — not “Reliable Service”, but language matching the theme.
- Note whether an image shows that specific product or situation.
- Check the group’s final URL: does the user land on a page about the theme, or on a generic homepage?
- For retail campaigns, confirm the feed actually contains in-stock, priced products in that category.
Any unchecked row is either a reason to remove the theme or a reason to strengthen the group. There is no third option: leaving a theme in place and hoping the system “figures it out” is the most expensive choice available.
One note on retail specifically: the feed outweighs themes. Disapproved products, missing GTINs and stale prices cannot be compensated for by any amount of theme tuning. Product segmentation for different goals belongs in custom labels in Merchant Center, not in an attempt to describe your catalogue in words.
A sane rollout order
- Fix conversion tracking: one primary goal, everything else secondary. Otherwise you optimise noise.
- Audit the feed and landing pages: availability, pricing, policy compliance.
- Build asset groups around semantic clusters, not one catch-all group.
- Add audience signals — customer lists, site visitors, similar segments. Customer Match and first-party data does the heavy lifting here.
- Let the campaign accumulate data untouched.
- Only now add 5–12 search themes that fill genuine gaps.
- After two to four weeks: category report, negatives, theme adjustments.
This order saves money. Themes added to a campaign with broken tracking and a weak feed simply accelerate the burn.
Further reading
Search themes are about where you appear. The adjacent layers are what the ad says and who sees it: ad customizers and business data feeds and demographic targeting and audience exclusions.
If you are scaling media buying and need dependable account infrastructure, see Google Ads agency accounts.
FAQ
Do search themes replace keywords?
No. Themes are an expansion signal; keywords are an auction entry condition with full control and reporting. In most accounts Search and PMax run alongside each other rather than one replacing the other.
How many search themes should an asset group have?
Five to twelve is a sensible working range against a limit of 25. Needing more usually means the group mixes unrelated intents and should be split.
Will PMax steal traffic from my Search campaigns?
When a query exactly matches an eligible keyword, Search keeps priority. Everything else is contestable, and search themes make PMax more competitive in that space.
Can I exclude queries in Performance Max?
Yes, with negative keywords at campaign and account level. There are no asset-group-level negatives, so plan your structure with that in mind.
Can I see the full list of PMax search queries?
No. You get aggregated search categories with metrics. That is enough to find waste and new demand, but not for phrase-level analysis.
Do themes affect Shopping and YouTube placements?
Themes influence the search side of inventory. Product and video placements are driven far more by the feed, assets and audience signals.
How long before results are readable?
Two weeks without changes is the practical minimum, and long sales cycles need a full conversion window. Earlier conclusions are almost always transition noise.
Should I add themes if PMax already converts well?
Not necessarily. If the category report shows no gaps and Search covers your core demand, themes add little. Start with diagnosis, not with filling the field.
Can I bid differently on a search theme?
No. Bidding is entirely campaign-level. If different intents need different targets, separate them into different campaigns.
Should the same theme appear in several asset groups?
Avoid it. You create internal competition and make attribution harder to read. Keep clusters non-overlapping.
CPA went up after I added themes. What now?
Check the category report for new irrelevant clusters, add campaign-level negatives, and remove the broadest themes. If two weeks of that does not help, revert and reintroduce themes one at a time.
Should I change audience signals at the same time?
No. Change one lever per measurement cycle, or you will never know which one moved the numbers.
