Table of contents
Introduction
English version of an article first published in Polish on 23 July 2026. Read the Polish original.
For well over a decade, working in Google Ads came down to one thing: the advertiser decides what to bid on, what to write and where the click leads. Automation kept taking away further pieces of that control, but it was always a choice. You could switch it on or leave it alone. In the first half of 2026 that arrangement stopped applying.
Three things came together at once: ads moved inside AI answers and now appear on almost one in three commercial queries, the new terms of service authorised Google to generate targets, ads and landing pages on the advertiser's behalf, and the old, predictable tools began to disappear in favour of AI Max. Each of these changes on its own would be an ordinary product update. Together they change the profession.
I break it down into parts: what exactly changed, how much of it is already live, what the first hard data show and which settings are worth reviewing before the automatic migration does it for you.
What you will learn:
- How often ads appear in AI Mode today, and on which queries most often
- Which ad formats run in AI Mode and what was announced at Google Marketing Live
- What exactly changed in the Google Ads terms of service in force since 1 July 2026
- What AI Max is, what it consists of and which controls stay on your side
- When DSA will disappear, what moves to automation in September and what only in February
- Where this model really fails: matching, brand protection and missing reporting
- A concrete list of settings to review before the automatic migration does it for you
Three changes that add up to one
It is worth seeing these events side by side, because only then does the pattern show. Within a few months Google moved ads from the model "the advertiser sets things up, the system executes" to the model "the system proposes and executes, the advertiser corrects".
Calendar of changes in Google Ads
-
15 April 2026
AI Max leaves beta
after eleven months of testing with hundreds of thousands of advertisers
-
20 May 2026
Google Marketing Live
the Conversational Discovery and Highlighted Answers formats are announced
-
11 June 2026
DSA migration postponed
from September 2026 to February 2027, after feedback from advertisers
-
1 July 2026
New terms of service take effect
all accounts covered automatically, with no acceptance window
-
September 2026
First wave of the automatic migration
automatically created assets and broad match; no more new DSA campaigns
-
February 2027
The end of Dynamic Search Ads
the remaining campaigns are moved to AI Max automatically
None of these changes is a scandal in itself. Only together do they describe a new division of roles: automation decides by default, a human corrects after the fact.
Scale: how many ads there are in AI Mode
Until July 2026 the discussion about ads in AI Mode rested mainly on individual screenshots. That changed with the SE Ranking study published on 15 July: it analysed 50,032 commercial queries from twenty industries, with the data collected on 30 June 2026.
The result: a text ad appeared on 29.45% of queries, that is 14,733 of the pool studied. Importantly, the researchers treat that figure as a floor, not a ceiling. AI Mode can be unstable, and the same query may show an ad in one session and none in the next.
Ad presence by cost per click
29.45%
of commercial queries with an ad
71.1%
of those showed two ads at once
50,032
queries studied across 20 industries
The link with the cost per click is the most interesting part. Neither search volume nor keyword difficulty translated into the presence of an ad. What counted was the price of the click. In other words, Google monetises AI Mode exactly where there is money to be made, and does so methodically.
The numbers Google announced at its I/O conference in May complete the picture of scale: AI Mode passed one billion monthly users within a year of launch, and the number of queries is said to be more than doubling every quarter. For comparison, AI Overviews reach around 2.5 billion users a month.
Ads in AI Mode stopped being an experiment the moment they started landing on one in three commercial queries. Now it is simply another placement, except that this one cannot be switched off.
Ad formats in AI Mode
In practice you will come across five layouts today. Some of them come close to organic content in style: the "Sponsored" label is there, but the block itself can look like a list of cited sources.
Five layouts you will see in practice
Text ads below the answer
at the end of the AI Mode answerThe most common format. Several ads stacked vertically in a grey block, styled much like organic citations.
Shopping carousel at the bottom
below the answerProduct cards with a photo, title, price, shop name and a review snippet, preceded by a heading.
Single products in the content
woven into the resultsSponsored products blended into their surroundings: labelled, but styled much like the neighbouring content.
Narrow product strip
nearer the top of the answerFavicon, shop name, price and attributes such as delivery information, often grouped on a grey background.
Sponsored product grid
above the free listingsUp to four sponsored results above the free list, often with labels of popular brands.
What was announced at Google Marketing Live
On 20 May 2026 Google showed two formats designed specifically for AI Mode, both powered by Gemini:
- Conversational Discovery: a creative built in real time for the specific wording of the question. Instead of matching a keyword to a ready-made headline, the model shapes the message around what the user actually asked.
- Highlighted Answers: when AI Mode generates a list of recommendations, relevant ads can enter that list directly, as one of its items.
Both keep the "Sponsored" label, and both get an independent commentary generated by Gemini next to the advertiser's creative. Google explains this with transparency and building trust. It is worth noting what that means in practice: next to your ad stands a text about your product that you have no influence over. The formats remain in testing, with no date given for a full rollout.
An ad written by a model, described by a commentary written by another model. The advertiser's role shifts from author to supplier of source material.
The new terms of service from 1 July
The most important change of this half-year was not announced on stage. It came into force quietly. The updated Google Ads terms say that the customer authorises Google and its affiliates to serve ads, including through automated features of the programme that format, select or generate targets, ads and landing pages on the customer's behalf.
The earlier provisions read differently: Google made tools available that may help generate these elements, and the advertiser switched them on or off. The shift is subtle in language and fundamental in effect: a feature you could tick has become a default authorisation.
What changed in practice
Before
until 30 June 2026
- Google makes automation tools available
- the advertiser switches them on or off
- the decision is a conscious choice
After
from 1 July 2026
- the authorisation to generate is the default
- it covers targets, ads and landing pages
- responsibility stays with the customer
The second part of the provision is just as important and easy to miss: responsibility for the generated material remains with the advertiser. It is the advertiser who has to review, approve, correct or remove automatically created campaigns and assets, regardless of who or what created them.
An asymmetry worth keeping in mind
The system may decide what an ad says, but the consequences (reputational, legal, budgetary) fall on the advertiser. For regulated industries such as medicine, finance and law, this is not a theoretical problem. A regular review of generated assets and landing pages stops being good practice and becomes part of risk management.
AI Max: what it is and what it does
AI Max is a layer of automation inside Search campaigns. It is not a separate campaign type like Performance Max. You stay on Search, but you agree to let the system choose queries more freely, rework ad text and swap landing pages. It left beta on 15 April 2026, after eleven months of testing.
What AI Max is made of
Search term matching
which queries
The system goes beyond the classic match types and judges for itself which queries fit your offer.
Text customisation
what the ad says
Headlines and descriptions are reworked for the specific query, based on your assets and the content of your site.
Final URL expansion
where the click lands
The system may send traffic to a different page from the one you set, if it judges it more relevant to the query.
What stays on your side: brand inclusion and exclusion lists, locations of interest and URL controls, which are carried over unchanged when you migrate from DSA. These are real tools, but you have to use them deliberately, because the default settings are liberal.
Google quotes its own result: on average 7% more conversions or conversion value at a similar CPA or ROAS when the full feature set is used, compared with search term matching alone. The figure comes from the vendor of the tool, so I treat it as a pointer to the direction, not as a promise. That is exactly what experiments in the Google Ads interface are for.
The DSA migration timetable
It is easy to get lost here, because the dates have changed and the old version still circulates in industry round-ups. As things stand today:
- Now until August 2026: the voluntary stage. You can move to AI Max yourself, at your own pace, setting everything up your way.
- September 2026: the automatic migration covers automatically created assets and campaign-level broad match settings. The option to set up new DSA campaigns ends.
- February 2027: the remaining DSA campaigns are moved automatically. The date was pushed back from September 2026 by a decision announced on 11 June.
Ginny Marvin, Ads Product Liaison at Google, explained the postponement plainly: advertisers had signalled that they needed more time, and Google did not want to go into the busy fourth quarter. It is a rare case of market feedback actually changing the timetable.
In the automatic move, AI Max is switched on with settings that mirror the existing configuration, and DSA campaigns become standard ad groups that keep their historical settings and URL controls. Former DSA campaigns get all three AI Max features switched on, while accounts using automatically created assets and broad match get search term matching and text customisation.
A migration you do yourself and a migration done by automation end up in the same place. The difference is who set the exclusions, and whether anyone set them at all.
Where it hurts
Automation in Google Ads can be effective and there is no point pretending otherwise. But three problems recur regularly enough to be worth knowing before you flip the switch.
Matches that should never have happened
In reviews of search terms reports in AI Max campaigns the same set keeps coming back: brand keywords matched to non-brand queries, non-brand keywords drifting onto competitors, and sometimes queries about your brand landing on competitor terms. Brand filters help, but typos and variants can slip through them, which is why solid negative keyword lists remain essential.
There is a small contradiction here that nobody settles out loud: Google recommends using negative keywords sparingly and adding them only once a given term consistently fails to meet your goals, because too many of them limit the benefits of intent-based matching. Practice is often the reverse. It is precisely negative keywords that protect the budget from broad matching.
Reporting that does not show what it should
There is no report segmentation for ads served in AI Overviews or in AI Mode. In standard reporting they simply appear as top-of-page results. So you cannot check what share of your clicks and costs comes from AI surfaces, or how they convert compared with classic results.
On top of that comes a criticism of AI Max itself: tests by Adalysis indicate that the reporting structure can inflate perceived performance by moving impressions over from exact and phrase match. The effect is that a campaign looks better than the real incremental gain justifies.
An ad is not a citation
The most interesting finding of the SE Ranking study concerns something many clients ask me directly: by paying for presence in AI Mode, do I increase the chance that AI will point to my site as a source.
Overlap between advertisers and cited sources
11.53%
of advertiser domains among the sources cited for the same queries
1.95%
overlap at the level of individual URLs
The answer is no. Money buys ad space, not a citation. The latter is won in an entirely different way, through content structure, credibility and data, which I wrote about in my guide to citations in AI Overviews.
What to do now
A concrete list for the coming weeks, ordered so that the most urgent things get done before the September wave of the migration:
- 01 Take an inventory of what will be moved anyway. Check where you have DSA campaigns, automatically created assets and campaign-level broad match. These three things migrate regardless of your decision. The only question is whether you manage to set them up before automation does.
- 02 Tidy up your negative keyword lists. Competitor brands, job-seeking queries, searches for free solutions, variants and typos of your own brand. With broad matching this is the main line of defence for the budget.
- 03 Set up brand lists, in both directions. Inclusions for your own brand, exclusions for competitors. Then check the search terms report for variants that slipped through anyway.
- 04 Make a deliberate decision about final URL expansion. If you have pages that paid traffic should not reach (terms and conditions, careers, blog, test versions), exclude them in advance. This is the setting that surprises people most often.
- 05 Test with an experiment, not on the whole account. The interface lets you start a comparison with one click. The declared 7% comes from Google. Your own data are the only version that counts.
- 06 Introduce a regular review of generated assets. Since responsibility has stayed on your side, you need a rhythm: once every two weeks, go through the automatically created headlines, descriptions and landing pages. More often in regulated industries.
- 07 Improve the quality of your conversion data. The more decisions automation makes, the more everything depends on what you feed it. Measuring the right events and sensible conversion values matter more today than manual bid tuning.
- 08 Create a baseline before the migration. Save cost, conversions, CPA and the structure of queries from the last 90 days. Without it you will not be able to prove to yourself or to your client whether the change helped or hurt.
The specialist's role shifts from operator to controller: less setting of bids, more watching over what automation should not be doing.
What it means for the Polish market
AI Mode has been available in Poland since 8 October 2025, when Google extended it to 36 languages and 50 countries. Poland is therefore one of the first EU markets with access, which means Polish users have already had time to get used to this interface.
On the other hand, for now Google offers ads inside AI Overviews in English, in a dozen or so countries. Ads above and below the AI answer, however, run wherever the feature itself is available. The practical conclusion for a Polish account: you will see some of the formats described here only with a delay, but the changes to the terms of service, the DSA migration and the behaviour of AI Max apply to you immediately and in full.
There is one more thread worth remembering when planning for 2027: the EU proceedings concerning Google's preferential treatment of its own services also cover AI surfaces. It is not known what will come of this for ads, but assuming that the layout of the SERP in Europe is fixed for years would be careless today.
For the smaller budgets I most often work with, the practical consequence is simple: with broad matching and automatically generated messaging, the quality of exclusions becomes more important than the number of keywords. A small account will feel a few hundred PLN spent on irrelevant queries sooner than a large advertiser who averages it out over the month.
I set out what a similar trade-off looks like in Performance Max campaigns, and how to keep them under control, in a separate guide to Performance Max. The mechanics are different, but the principle is the same: automation works as well as the data and the limits you give it.
FAQ: frequently asked questions
Can I switch off ad serving in AI Overviews and AI Mode?
No. Google states plainly that you cannot target placements in AI Overviews directly or opt out of them. Ads get there from Search, Shopping and Performance Max campaigns, and which one appears is decided by an auction that takes into account both the user's query and the content of the AI answer itself. The only things you can influence are the quality and relevance of your material and the settings of the campaign as a whole.
What exactly did the Google Ads terms of service change on 1 July 2026?
The new wording authorises Google and its affiliates to serve ads, including through automated features of the programme that format, select or generate targets, ads and landing pages on the customer's behalf. The earlier provisions spoke rather of making tools available that the advertiser could use by switching them on or off. Now it is a default authorisation, not an option to tick. All accounts were covered by the change on 1 July 2026, with no acceptance window and no re-approval of the terms.
Who is responsible for an ad generated by automation?
The advertiser. The updated terms stress that it is the customer who is obliged to review, approve, edit or remove automatically created campaigns and ad assets, regardless of who or what generated them. In practice this means that a regular review of assets and landing pages stops being good practice and becomes part of risk management.
What is AI Max and do I have to switch it on?
AI Max is a layer of automation for Search campaigns that left beta on 15 April 2026. It consists of three elements: search term matching (broader than the classic match types), text customisation of ads and final URL expansion. Switching it on is not mandatory straight away, but the direction is unambiguous: campaigns using DSA, automatically created assets and campaign-level broad match will be moved automatically anyway, and new DSA campaigns can no longer be created after September 2026.
When exactly will Dynamic Search Ads disappear?
The timetable has changed. Originally the automatic migration of DSA to AI Max was due in September 2026, but on 11 June 2026 Google pushed it back to February 2027. Ginny Marvin explained it by the need to give advertisers more time and to stay out of the busy fourth quarter. The September date remains in force, however, for automatically created assets and campaign-level broad match settings. After September 2026 the option to set up new DSA campaigns also ends.
Will my reports show how much performance comes from AI Mode?
As of today, no. Google Ads does not offer report segmentation for ads served in AI Overviews or in AI Mode. In standard reporting they are simply recorded as top-of-page results. Google says dedicated controls and reporting segments will appear as these surfaces develop, but has given no specific date. Until then, any conclusion about how well these placements perform rests on indirect signals.
Does an ad in AI Mode increase the chance that AI will cite my site?
The data suggest these are two separate mechanisms. In the SE Ranking study only 11.53% of advertiser domains appeared among the sources cited for the same queries, and at the level of individual URLs the overlap fell to 1.95%. Paying for presence in AI Mode therefore does not buy organic visibility in the answers, and working on citations is an entirely different front.
Does AI Max replace Performance Max?
No, they are two different things. Performance Max is a campaign type covering all Google channels, in which you hand the choice of placements to automation. AI Max is a layer of automation inside Search campaigns: you stay in a Search campaign, but you let the system match queries more broadly, rework ad text and swap landing pages. You can run them side by side, although then you need to watch your exclusions all the more carefully, so that the campaigns do not start competing for the same queries.
Worth remembering
- Ads appear on 29.45% of commercial queries in AI Mode, and the researchers treat that figure as a floor, not a ceiling.
- The more expensive the click, the greater the chance of an ad: 24.33% at a CPC below 2 dollars against 53.56% above 10 dollars.
- The terms in force since 1 July 2026 authorise Google to format, select and generate targets, ads and landing pages on your behalf, by default and without separate consent.
- Responsibility for the generated material stays with the advertiser, regardless of what created it.
- AI Max left beta on 15 April 2026 and consists of three elements: search term matching, text customisation and final URL expansion.
- In September 2026 automation will take over automatically created assets and campaign-level broad match; the DSA migration has been pushed back to February 2027.
- There is no report segmentation for AI Mode or AI Overviews, so you will not see how much of your performance comes from those surfaces.
- Only 11.53% of advertiser domains appeared among the sources cited for the same queries. Advertising and citation are two separate worlds.
Summary
Google is not hiding the direction and there is no point pretending it can be reversed. Ads have moved inside AI answers and you cannot opt out of them. The terms of service have shifted content generation from an option to a default authorisation. Predictable tools are disappearing, replaced by a layer that decides for itself about matching, messaging and the landing page.
What is left for the advertiser? More than the complaints suggest, and less than there used to be. What remains are exclusions, brand lists, URL controls, the quality of conversion data, experiments and the discipline of reviews. It is not the same job as five years ago, but it is still a job in which you can make a difference, just in a different place.
Automation will reconfigure your campaigns anyway. The only real question is whether it does so on your settings or on its own defaults.
September is a few weeks away. Reviewing exclusions, brand lists and URL controls is a few hours of work that later saves a quarter of explaining where the job-seeking queries in a sales campaign report came from.
Sources and materials
The Google documentation, announcements and studies on which I base this analysis:
- Google Ads Help – ads and AI Overviews (eligible campaigns, no opt-out)
- Google Blog – the next generation of ads for the AI era of Search (Google Marketing Live 2026)
- Google Blog – Dynamic Search Ads are moving to AI Max
- Search Engine Land – Google Ads updates terms of service ahead of July 2026 rollout
- Search Engine Land – Google AI Mode ads reach nearly 30% of queries (SE Ranking study, 15 July 2026)
- SE Ranking – AI Mode now shows ads on nearly 1 in 3 commercial queries
- Google Ads Help – reporting in AI Max campaigns
Want to go through the migration on your own terms?
I will review the account for exclusions, brand lists and URL controls, set up an experiment and show in numbers what AI Max really does with your budget.
Book a consultationHow this text was written
I will not pretend it was written entirely by hand, by candlelight. For the research, the proofreading and the code of this page I used a language model. It is a good tool for digging through documentation and catching the typos you no longer see on the third read of your own text. The English version was also prepared with a model, from my Polish original.
What the model did not do: it did not choose the topic, form the arguments or decide what this data means for your site. That stays with me, as does the last word on what goes into the text and what comes out of it.
Where the model helped
- digging through sources and organising data
- proofreading and the English version of the text
- code, page layout and infographics
What stays with me
- the choice of topic and what is worth saying about it
- the arguments, the reading of the data and the conclusions
- fact-checking and the decision on what goes into the text