Why Performance Max drifts in Germany, and how I structure asset groups to steer it

Performance Max in Germany usually fails on structure, not on culture

An Irish shop has a Performance Max campaign that works at home. It copies the campaign, changes the location to Germany, runs the assets through a translation tool and waits. The budget is even in the same currency. Six weeks later the German return is half the Irish one, and the verdict in the office is that Germans are different.

In the accounts I have looked at, culture is rarely what broke it. A campaign built for a market with plenty of signal was moved into one where the machine sees much less, with every product and brand search mixed into one asset group. The old version of this article leaned on "German buyers are more literal". I could not find a study behind that, so it stays my experience and no more. The structural causes you can check in your own account this afternoon.

This article covers one campaign type in one country. How to run several countries from one account is in cross border PPC from an Irish base, and the general case for and against automated campaigns in the shift to black box campaigns.

Worked example: the shop I will use throughout

Shandon Ironware is an invented shop in Cork. It sells cast iron frying pans, Dutch ovens and enamelled cookware online, at shandonironware.example, with a German section at shandonironware.example/de. The owner is Siobhán. Nobody real, no client, every number made up so that every step can be shown.

Her Google Ads account is in euro. The Irish Performance Max campaign spends €9,000 a month and is doing grand. In spring she copied it for Germany: one campaign, €3,000 a month, one asset group with every product, translated headlines, Irish images, no videos. German shoppers pay in euro too, so there is no exchange rate anywhere in the sums below, and all reporting is in euro.

What Performance Max controls, and what it only suggests

Performance Max is one goal based campaign that serves across Search, YouTube, Display, Discover, Gmail and Maps, and Shopping when a feed is attached.¹ You give it a goal, a budget, creative and some hints. It decides where and to whom to show what.

Inside the campaign sit asset groups. An asset group is "made of images, logos, headlines, descriptions, videos, sitelinks, assets, and audience signals", and Google's own advice is to build each one around "a single theme or audience".² A campaign needs at least one asset group and can hold up to 100, and an asset group cannot be shared between campaigns.³ For a shop, an asset group can also carry a listing group, the slice of the product feed it is allowed to advertise.

The asset group has hard limits: up to 15 headlines of 30 characters, at least one of 15 or fewer, five descriptions of 90 characters, four to 20 landscape and square images each, and up to 15 videos. If you add no video, Google may generate one from your other assets.⁴ The business name and logo now live in the campaign's brand guidelines rather than in each asset group.²

Now the part that matters most. Some settings are hard rules, and some are suggestions.

The hard rules sit on the campaign: budget and location. Google's advice is a separate campaign whenever you need a different budget or location targeting.⁵ An asset group cannot have its own country, so the old article's "asset groups for Germany, Austria and Switzerland" would all serve everywhere the campaign targets. Each country gets its own campaign.

The suggestions sit on the asset group. Audience signals are lists and segments you hand the machine as a starting point: site visitors, customer lists, custom segments, interests, demographics. Google states that "Performance Max may show ads to relevant audiences outside of your signals" when they are likely to convert.⁶ Signals are applied at asset group level and can take up to two weeks to be fully absorbed.⁷ Search themes are the same idea for Search: up to 50 per asset group, optional, additive to what the machine finds on its own, and carrying the same priority as phrase and broad match keywords in your Search campaigns.⁸ An exact match keyword in a Search campaign that matches the query exactly takes priority over Performance Max.¹

So an asset group can only enforce what it contains: its products, its landing pages and its words. It cannot enforce who sees it.

Worked example: Siobhán's German campaign as it was

Siobhán's single German asset group contained 62 products from all three ranges, one final URL (the German home page), the translated Irish headlines, eight Irish lifestyle photos, the Shandon Ironware logo, no video, and one audience signal: her Irish customer list, which matched almost nobody in Germany.

So before a single impression, Google was free to send any searcher to any of 62 products with words written for another market, YouTube would run videos Google built from Irish photos, and the only audience hint was empty. None of that is culture. It is a campaign with nothing to steer by.

Consent shrinks the signal before the first asset group is built

Performance Max learns from conversions it can see. In Germany it sees fewer of them, by law.

Section 25 of the German Telecommunications and Digital Services Data Protection Act (TDDDG) allows storing or reading information on a visitor's device only with their consent, on clear and comprehensive information, with two narrow exceptions: transmitting a message, and what is strictly necessary for a service the visitor explicitly asked for.⁹ Advertising measurement is neither. The full legal picture is in cookies and consent in Germany and Austria.

For German visitors my default is basic consent mode, as it is for Irish ones. In basic mode, Google's tags are held back until the visitor interacts with the banner, and "when the user doesn't consent, no data is transferred to Google at all".¹⁰ The reasoning is in event tracking and consent mode, and I hold the same line on Meta in the Meta ads article for DACH. It has a price. A German order from a visitor who refused is invisible to Google Ads. So is that visitor's place on any remarketing list, because the tag that would have put them there never ran.

None of this is new for an Irish shop. At home, regulation 5 of the ePrivacy Regulations 2011 asks for the same yes before anything is stored on or read from a visitor's device, and the Data Protection Commission (DPC) says cookies related to advertising are not strictly necessary.¹¹¹² So if your Irish site already runs basic consent mode, your Irish campaign already learns only from visitors who accepted. The gap is familiar. What is not familiar is its size: the consent rate to plan the German campaign with is the one you measure on your German pages, not the one from your Irish banner, and German law governs your German visitors whatever your Irish setup does.

The conversions the platform observes are the true conversions times the consent rate.

$$C_{\text{obs}} = C \times r$$

Here $C_{\text{obs}}$ is the number of conversions Google Ads observes in a month, $C$ is the true number of orders from those ad clicks, which only your shop database knows, and $r$ is the consent rate, the share of visitors who accepted. It is a rate applied to a count, as in Foundations: ratios and rates.

Google's modelling can add an estimate on top, under conditions explained, with the thresholds, in the consent article. It does not put refusing visitors back onto your remarketing lists, and it does not turn an estimate into the real journeys the machine learns most from. For planning, I count on $C_{\text{obs}}$.

Worked example: how many German orders Google sees

Siobhán's shop database matches each order to its landing session. In May it shows 150 German orders from Performance Max clicks. I assume 45% of German visitors press accept, an assumption for the arithmetic, not a statistic. Measure your own.

$$C_{\text{obs}} = 150 \times 0.45 = 67.5$$

Here $C$ is 150 true orders, $r$ is 0.45, and $C_{\text{obs}}$ comes to 67.5, so about 68 conversions a month for the machine to learn from. The Irish campaign, behind the same kind of banner, feeds its machine several hundred. Same software, a fraction of the evidence. That, more than any adjective about German buyers, is why a copied campaign drifts.

Every asset group splits a thin signal further

Here I part company with the old version of this article, which rebuilt a struggling German campaign into six asset groups. With 68 observed conversions a month, that turns one thin signal into six very thin ones.

Google's guidance is to run a new Performance Max campaign for at least six weeks before judging it, and to avoid frequent changes to budget, bid strategy or status in that period, because each one can reset the learning.¹³ Every asset group needs its own conversions to tell good assets from bad. I use a plain ceiling:

$$G_{\max} = \left\lfloor \frac{C_{\text{obs}}}{n_{\min}} \right\rfloor$$

Here $G_{\max}$ is the largest number of asset groups I will run in the campaign, $C_{\text{obs}}$ is the observed conversions per month from the section above, $n_{\min}$ is the minimum observed conversions per asset group per month I want before I trust its reports, and the brackets mean round down to a whole number. It is one rate divided by another, as in Foundations: ratios and rates.

The $n_{\min}$ is mine, not Google's. In my experience 30 observed conversions a month per asset group is about the least that lets me read the reports without fooling myself. The reason it cannot be much lower is the standard error of a proportion, which you will meet properly two sections from now and in Foundations: sampling: with ten conversions, the noise is about a third of the signal.

Worked example: two asset groups, not six

With Siobhán's numbers:

$$G_{\max} = \left\lfloor \frac{68}{30} \right\rfloor = \lfloor 2.27 \rfloor = 2$$

Here $C_{\text{obs}}$ is 68, $n_{\min}$ is 30, and $G_{\max}$ rounds down to 2. Six asset groups would give each one about $68 / 6 \approx 11$ observed conversions a month. Two give each about 34. Siobhán's German campaign gets two asset groups. When the German campaign observes 120 conversions a month, it can have four.

The right structure for Germany is neither the Irish one translated nor the most granular one you can draw on a whiteboard.

Split asset groups by what Performance Max honours

With two asset groups, the split has to be on something the asset group enforces. Google recommends separating asset groups by "content category, theme, language, or target audience".¹⁴ The audience is a suggestion. The other three are enforced, because they live in the assets, the products and the landing pages.

So the old article's splits by "stage of awareness" and "intent segment" fall away. They are audience ideas, and the campaign shows each group's ads to whoever it predicts will convert. A split the machine may ignore is not control.

What I split on, in order:

1. Products with different margins or different buyers. Each asset group gets its own listing group, so a frying pan headline never sits next to an enamel casserole image.
2. Landing pages in the right language. Each asset group gets a final URL on the German section of the site.
3. Language, if the campaign serves more than one. From September 2026, the Search part of Performance Max matches ads by the language of the ads themselves, and the campaign language setting only still applies to YouTube, Display, Discover and Gmail.¹⁵ A German asset group is now German because its words are German.

Then one setting that quietly undoes all of it. Final URL expansion is on by default in Performance Max and lets the campaign "replace your final URL with a more relevant landing page from the final URL domain".¹⁶ Where Irish and German pages share a domain, the most relevant page for a German search can be the Irish one, in English and written for Irish delivery. Google offers URL exclusions, which only work while expansion is on, and URL contains rules, with a warning that rules that are too broad limit reach.¹⁷ I keep expansion on and confine it to the German section with a rule.

Worked example: Siobhán's two asset groups

Siobhán's two asset groups follow her ranges: frying pans and skillets (high volume, lower price), and Dutch ovens and enamelware (lower volume, higher price, often a gift).

Asset group one, Pfannen. Listing group: the 34 pan and skillet products. Final URL: shandonironware.example/de/pfannen. Search themes: eight, in German, written from a German speaker's keyword research rather than translated, for example "gusseisenpfanne 28 cm" and "pfanne für induktion". Audience signal: a custom segment built from those themes, plus the German site visitors list, which only contains people who accepted.

Asset group two, Bräter und Emaille. Listing group: the 28 Dutch oven and enamel products. Final URL: shandonironware.example/de/braeter. Six search themes, including "bräter gusseisen emailliert". Same visitors list as a signal.

Final URL expansion stays on, with a URL contains rule that admits only addresses containing /de/. Location: Germany, presence only. Austria is a separate campaign, not a third asset group.

Write the German assets inside the character limits

German nouns are long. A headline is 30 characters including spaces, and at least one must fit in 15.⁴ A translated English headline usually overflows, or gets cut into something nobody would say.

Write the German assets from scratch with a native speaker, and fill the asset group: Google recommends using the Ad Strength indicator and adding up to 15 headlines and five descriptions, and waiting two to three weeks before replacing an asset that looks weak.¹⁴ Upload your own videos in all three orientations, horizontal, square and vertical, or Google will make some for you from what it has.⁴ For a German campaign, "what it has" may be your Irish lifestyle photos.

In my experience German shoppers respond better to an asset that states the product, size, price and delivery time than to one that performs excitement. The longer argument is in why your home brand tone can undermine trust in DACH, and fake urgency, which is a legal question rather than one of taste, is covered in the Meta ads article.

Prices are where the law reaches into the headline. Germany's standard VAT rate is 19%.¹⁸ The Price Indication Ordinance (PAngV) defines the total price as the price including VAT and other price components,¹⁹ and requires a business that offers or advertises goods to consumers with a price to state that total price.²⁰ So a price in a German consumer headline is the gross price, never the net one. If your shop system holds net prices, the gross price is a percentage increase:

$$P_{\text{gross}} = P_{\text{net}} \times (1 + t)$$

Here $P_{\text{gross}}$ is the total price the German shopper pays and must see, $P_{\text{net}}$ is the price before VAT, and $t$ is the VAT rate as a decimal, 0.19 for the German standard rate. A percentage increase applied to a number, as in Foundations: percentages.

Worked example: fifteen headlines that fit

Siobhán's best selling pan sells in Germany for €69 including VAT, a German shelf price in the same currency as her own books. The rate in it is the German 19%, not the Irish standard rate, because I assume her German orders carry German VAT. Which rate yours carry under the EU distance selling rules is a question for your accountant, and the headline rule is the same either way. With $P_{\text{gross}} = 69$ and $t = 0.19$, the net price is $P_{\text{net}} = 69 / 1.19 = 57.98$, so €57.98, and the VAT in it is $69 - 57.98 = 11.02$, so €11.02. The German headline writes it the German way, 69 €, with the symbol after the number.

Her translator's first draft for the Pfannen group included "Gusseisenpfannen aus Cork, Irland", which is 33 characters and does not fit. The German speaker she hired rewrote the set from scratch. Five of the fifteen, with their lengths:

• "Gusseisen" is 9 characters, the one that must fit in 15.
• "Gusseisenpfanne 28 cm" is 21 characters.
• "Bratpfanne 28 cm für 69 €" is 25 characters, and 69 € is the total price.
• "Gusseisen aus Irland" is 20 characters.
• "Lieferung in 3 bis 5 Tagen" is 26 characters.

The business name, Shandon Ironware, is 16 characters, inside the 25 allowed, and sits in the brand guidelines once for the whole campaign.⁴² Siobhán added three German videos shot on a phone in her kitchen, one in each orientation, so YouTube stops showing a slideshow of Irish photos.

Keep your own brand searches out of the German numbers

Before you judge the German campaign, take your own name out of it. People who already know your shop search for it by name, and those sales would mostly have happened anyway. When Performance Max takes credit for them, it looks better than it is.

There are two tools. Brand exclusions "keep your campaigns from serving for branded queries you want to avoid", and in Performance Max they apply to Search, Shopping and YouTube search inventory, with an option to still allow Shopping ads on searches that mention the excluded brand.²¹ Since 27 May 2025, brand exclusions for Search campaigns have been moving into AI Max, and Google states that this "does not impact Performance Max campaigns".²¹ Then negative keywords: Performance Max now takes campaign negative keywords, which apply to Search and Shopping inventory only, and account level negatives apply to it automatically.²² Google announced campaign negatives for all advertisers in January 2025.²³

My setup is a small Search campaign for the brand in each country, with the brand as an exact match keyword, which takes priority over Performance Max for exactly that query,¹ and a brand exclusion on the Performance Max campaign so it does not reach for the near misses either.

To see how much the brand flatters the campaign, split its return:

$$\text{ROAS}_{\text{nonbrand}} = \frac{V - V_{\text{brand}}}{S - S_{\text{brand}}}$$

Here $\text{ROAS}_{\text{nonbrand}}$ is the return on ad spend without your own brand searches, $V$ is the campaign's total conversion value and $S$ its total spend for the period, and $V_{\text{brand}}$ and $S_{\text{brand}}$ are the value and spend on search terms that contain your brand. A ratio of two differences, as in Foundations: ratios and rates. Report margin as the value, not revenue, for the reasons in auto bidding on your own data.

Worked example: what the German campaign earns without the brand

Siobhán's conversion value is the margin on each order, averaging €48 for German orders after cost of goods, delivery and payment fees. In May the German campaign observed 68 conversions, so $V = 68 \times 48 = 3{,}264$, on $S = 3{,}000$. The campaign ROAS on margin is $3{,}264 / 3{,}000 = 1.09$. It looks like it just about pays.

The search terms report shows 20 of the 68 conversions came from searches containing "shandon". Those cost €180 and brought $20 \times 48 = 960$ of margin.

$$\text{ROAS}_{\text{nonbrand}} = \frac{3{,}264 - 960}{3{,}000 - 180} = \frac{2{,}304}{2{,}820} = 0.82$$

Here $V$ is €3,264, $V_{\text{brand}}$ €960, $S$ €3,000 and $S_{\text{brand}}$ €180. Without the brand, every euro of German spend brought back 82 cent of observed margin. That is the honest starting line, and it is lower than it looked.

Two caveats. These are observed conversions only, and reconciling them with the shop database per market is in cross language attribution. And a brand search is not worth zero, because some of those people found the brand through the campaign. The split gives a range, not a point.

Read the 2025 reports before you change anything

"Performance Max hides its reporting" was true when the old version of this article was written. In 2025 Google added most of what was missing.

In April 2025 Google announced full search terms reporting for Performance Max, cost and clicks in asset reporting, and a channel performance page.²⁴ That report shows impressions, clicks, conversions, value and cost per channel, from Search to Gmail, for any date after 6 June 2025. It does not let you move budget between channels: Google states that "you can't directly control budget allocation per channel".²⁵ The asset group report shows results per asset group, and Google warns not to remove an asset group on that data alone.²⁶

So the order of work is: read first, change one thing, then wait. Google's own advice is one to two weeks, or a full conversion cycle, after a significant change to an existing campaign.¹³

The flow runs from consent at the top to the monthly review at the bottom. The first boxes are decided before any asset is written, because most of the control sits in the structure, not the creative. The decision allows one change at a time, because two changes at once cannot be told apart.

Worked example: where the German money went

Siobhán opens the channel performance report for May, with the old single asset group still running.

The horizontal axis shows the six channels the campaign served on. The vertical axis shows what each one cost in May, in euro, adding up to €3,000. Search took 55% of the money and brought 52 of the 68 conversions. YouTube took 18% and brought five, served from videos Google had generated out of Irish photos.

She cannot cut YouTube's share directly. She can change what YouTube works with: the German videos from the asset section. The search terms report adds a cluster of searches on how to season cast iron, clicks without orders, which get German negative keywords such as "einbrennen". Then six weeks of patience, because the restructure counts as a new start.

Judge asset groups with intervals, not with a league table

With two asset groups, the temptation is to rank them every Monday and cut the loser. At German volumes the ranking is often noise, so put a range around each conversion rate first:

$$p \pm 1.96 \times \sqrt{\frac{p(1-p)}{n}}$$

Here $p$ is the asset group's conversion rate, observed conversions divided by clicks, $n$ is the number of clicks, the square root is the standard error of a proportion, and 1.96 stretches it to a 95% confidence interval. If two asset groups' intervals overlap, the data does not yet say which is better. The interval is explained step by step in Foundations: confidence intervals, and the standard error in Foundations: sampling. Google's advice to keep judging at campaign level fits the arithmetic.²⁶

Worked example: comparing Siobhán's two asset groups

Eight weeks after the restructure, Siobhán compares a month of the two asset groups.

Pfannen: 1,300 clicks, 39 observed conversions. $p = 39 / 1{,}300 = 0.030$. The standard error is $\sqrt{0.030 \times 0.970 / 1{,}300} = \sqrt{0.0000224} = 0.00473$. Times 1.96 is 0.00927. The interval is 2.07% to 3.93%.

Bräter und Emaille: 1,000 clicks, 24 observed conversions. $p = 24 / 1{,}000 = 0.024$. The standard error is $\sqrt{0.024 \times 0.976 / 1{,}000} = \sqrt{0.0000234} = 0.00484$. Times 1.96 is 0.00949. The interval is 1.45% to 3.35%.

Here $p$ is each group's conversion rate and $n$ its clicks. Pfannen looks better, 3.0% against 2.4%, but the intervals overlap between 2.07% and 3.35%. Siobhán cannot yet say which converts better, and leaves both alone. For a real answer she needs more data, and how much data you actually need has the sums.

What I would set up for an Irish shop advertising in Germany

The old version of this article ended with "automation needs boundaries". That stands. What changed is where I put them: in the few places the machine has to obey.

In order: basic consent mode and a shop database that knows every order. One campaign per country. A brand Search campaign on exact match and a brand exclusion. Let observed conversions set the number of asset groups, usually two or three. Split by product range and German landing page, with Final URL expansion kept inside the German section. German assets written from scratch, with gross prices and your own videos. Then six weeks, the three reports, and one change at a time.

None of that needs a theory about German buyers. It needs a campaign built for the evidence the German market actually gives you.

Footnotes

1. Google, About Performance Max campaigns, Google Ads Help, 2026, https://support.google.com/google-ads/answer/10724817

2. Google, Build an asset group, Google Ads Help, 2026, https://support.google.com/google-ads/answer/10724492

3. Google, Performance Max asset groups, Google Ads API documentation, Google for Developers, 2026, https://developers.google.com/google-ads/api/performance-max/asset-groups

4. Google, Performance Max campaigns specs and format requirements, Google Ads Help, 2026, https://support.google.com/google-ads/answer/17091269

5. Google, Organize your account with ad groups, Google Ads Help, 2026, https://support.google.com/google-ads/answer/6372655

6. Google, About audience signals for Performance Max campaigns, Google Ads Help, 2026, https://support.google.com/google-ads/answer/14530785

7. Google, Add audience signals, Google Ads Help, 2026, https://support.google.com/google-ads/answer/14753570

8. Google, Use search themes with your Performance Max campaign, Google Ads Help, 2026, https://support.google.com/google-ads/answer/14767319

9. Telekommunikation Digitale Dienste Datenschutz Gesetz (TDDDG), § 25 Schutz der Privatsphäre bei Endeinrichtungen, Bundesministerium der Justiz, gesetze im internet, https://www.gesetze-im-internet.de/ttdsg/__25.html

10. Google, Consent mode overview, Google for Developers, Tag Platform, 2026, https://developers.google.com/tag-platform/security/concepts/consent-mode

11. European Communities (Electronic Communications Networks and Services) (Privacy and Electronic Communications) Regulations 2011, S.I. No. 336 of 2011, regulation 5, Irish Statute Book, 2011, https://www.irishstatutebook.ie/eli/2011/si/336/made/en/print

12. Data Protection Commission, Guidance note on cookies and other tracking technologies, DPC, April 2020, https://www.dataprotection.ie/sites/default/files/uploads/2020-04/Guidance%20note%20on%20cookies%20and%20other%20tracking%20technologies.pdf

13. Google, Optimisation tips for Performance Max campaign for all business types, Google Ads Help, 2026, https://support.google.com/google-ads/answer/11385582?hl=en-GB

14. Google, Best practices for asset groups in Performance Max campaigns, Google Ads Help, 2026, https://support.google.com/google-ads/answer/14528220

15. Google, About language targeting (the September 2026 change for Search and Performance Max), Google Ads Help, 2026, https://support.google.com/google-ads/answer/1722078

16. Google, About Final URL expansion in Performance Max, Google Ads Help, 2026, https://support.google.com/google-ads/answer/14337539?hl=en-GB

17. Google, About Final URL expansion (asset optimisation), Google Ads Help, 2026, https://support.google.com/google-ads/answer/16672777

18. Umsatzsteuergesetz (UStG), § 12 Steuersätze, Bundesministerium der Justiz, gesetze im internet, https://www.gesetze-im-internet.de/ustg_1980/__12.html

19. Preisangabenverordnung (PAngV), § 2 Begriffsbestimmungen, Bundesministerium der Justiz, gesetze im internet, https://www.gesetze-im-internet.de/pangv_2022/__2.html

20. Preisangabenverordnung (PAngV), § 3 Pflicht zur Angabe des Gesamtpreises, Bundesministerium der Justiz, gesetze im internet, https://www.gesetze-im-internet.de/pangv_2022/__3.html

21. Google, Apply brand exclusions to Performance Max or Search campaigns, Google Ads Help, 2026, https://support.google.com/google-ads/answer/14505308

22. Google, Negative keywords in Performance Max campaigns, Google Ads Help, 2026, https://support.google.com/google-ads/answer/15726455

23. Tal Akabas, Kick off 2025 with new Performance Max features, Google, The Keyword, 23 January 2025, https://blog.google/products/ads-commerce/new-performance-max-features-2025/

24. Tal Akabas, Channel performance and more reporting coming to Performance Max, Google, The Keyword, 30 April 2025, https://blog.google/products/ads-commerce/channel-performance-reporting-coming-to-performance-max/

25. Google, About the channel performance report, Google Ads Help, 2026, https://support.google.com/google-ads/answer/16260130

26. Google, About asset group reporting for Performance Max, Google Ads Help, 2026, https://support.google.com/google-ads/answer/13872527