Google Sunsets Display Ads, Caps Data Retention

Estimated reading time: 26 minutes

A busy week to close out May. Google confirmed it’s retiring standalone Display campaigns in favour of Demand Gen, announced that granular Ads reporting data will be deleted after 37 months, and (via Sundar Pichai) conceded that AI Overviews are “more opinionated than they should be.” The May 2026 core update finished rolling over the bank holiday weekend, Preferred Sources passed 345,000 selections as Google expanded them across AI search, and OpenAI confirmed conversion-focused ads are coming to ChatGPT in June. Add fresh research on how AI search traffic diverges from organic, and there’s plenty of search marketing news to work through.

May 2026 Core Update Finished Rolling Over the Weekend

The May 2026 core update, which began rolling out on 21 May, completed over the bank holiday weekend. That’s faster than the up-to-two-weeks window Google gave at launch. Tracking tools showed the sharpest volatility on 23 and 24 May, settling by 25 May.

Early reads from the SEO community point to a fairly broad update that touched a wide spread of verticals, with some recovery visible for sites hit by the March update and fresh losses for thin or templated content. As always with core updates, the cleanest signal comes a week or two after the dust settles rather than during the rollout.

For anyone whose clients moved, the work now is comparison. Pull the before-and-after ranking and traffic data, segment the movement by content type and template, and look for whether the losers share signatures. If a client recovered ground lost in March, document what changed in the interim so the pattern is legible. We’re holding formal recovery recommendations until mid-June, since the update only just stabilised and the I/O-era AI changes from last week muddy the attribution.

Pichai Concedes AI Overviews Are "More Opinionated Than They Should Be"

“It’s probably more opinionated than it should be for the particular query you showed me.” That was Sundar Pichai on the Decoder podcast with Nilay Patel, responding to a live “best Chromebook” AI Overview Patel pulled up on his phone. It’s a notable admission from Google’s CEO that the AI Overviews product has a calibration problem, even if Pichai framed it as ordinary iteration in a fast-moving space.

The more pointed exchange was on publisher traffic. Pichai characterised the decline in clicks to publishers as “bounce clicks going down,” arguing that Google’s technology is filtering out low-quality clicks rather than starving publishers of valuable traffic. That’s a convenient framing, and one Google hasn’t backed with publisher-facing data. He also declined to push back on the Conde Nast “plan for zero search” stance covered two weeks ago, saying he wasn’t in a position to tell an iconic publisher what to think.

The honest read here is that the CEO of the company that runs AI Overviews just acknowledged the product is over-opinionated and reframed publisher traffic loss as a feature. Publishers and brands should take the “bounce clicks” framing with scepticism until Google shows the data, and plan for the click-suppression trend to continue regardless of how it’s labelled.

846,000 Searches Show How AI Overviews Change Scroll Behaviour

59% of users reverse their scrolling direction when an AI Overview is present, against 51% without one. That’s from ClickStream Solutions’ analysis of roughly 846,000 US Google Search sessions in February and March 2026, using anonymised Surfer SEO clickstream data.

The deeper finding is in the magnitude of the back-scroll. Among users who scrolled back up, those with AI Overviews spent 47.5% of their total scrolling moving upward, versus 27% without. For navigational searches specifically, back-scrolling jumped from 23% to 44% when an AI Overview appeared. People aren’t reflexively clicking the top result any more; they’re reading the AI answer, scrolling past it, then scrolling back up to reconsider.

The lesson for SERP strategy is that the result preview now does more work. Title tags and meta descriptions become competitive differentiators in a way they weren’t when the top organic result got the reflexive click. A vague or generic snippet fails the second-look test. Clarity and specificity in the preview are now a measurable advantage, which is a rare instance of an AI-era change that rewards exactly the discipline good SEOs already practise.

Google Preferred Sources Hit 345K and Expand Across AI Search

345,000 unique Preferred Sources have now been selected by users, and Google is expanding the feature across AI Mode and AI Overviews. Users are twice as likely to click through to a source they’ve marked preferred versus a non-preferred result. The expansion rolled out on 27 May, globally and in all languages.

Three connected changes shipped together:

Preferred sources now get distinct labels inside AI responses, so users can spot links from outlets they’ve already chosen.
A new perspectives carousel surfaces viewpoints from discussions, forums, and social media alongside timely articles.
The “highly cited” label expanded across more article links, flagging primary reporting that other stories reference frequently.

For publishers, this opens three distinct visibility pathways inside AI search: getting selected as a preferred source, earning the highly cited label through original reporting, and appearing in the perspectives carousel. The first is an audience-loyalty play (prompt your readers to mark you preferred), the second rewards primary reporting and original data, and the third favours genuine community discussion over polished marketing content. None of them are gameable through conventional on-page SEO, which is the point.

Google AI Mode Now Scales Faster Across Languages

Google said its newer multilingual models let AI Mode expand across countries and languages much faster than the previous generation. Where each new language used to require significant separate tuning, the current models generalise across languages with far less per-language work, which means AI Mode rollout to new markets accelerates from here.

For any agency with international clients, this matters sooner than it looks. AI Mode reaching a client’s non-English markets on a compressed timeline means the conversational-query optimisation work can’t stay an English-first project that gets translated later. Markets you assumed had a 12-month runway before AI Mode arrived may get it in a few months. We’re advising international clients to treat AI Mode readiness as a near-term requirement in their primary non-English markets rather than a deferred one.

Similarweb: ChatGPT Shows More Links, Referrals Jump 150%

Referral traffic from ChatGPT is up roughly 150% since 7 May, when OpenAI began surfacing more prominent links to brands inside its answers. The data comes from Similarweb, comparing the week before 7 May with the week after. Around 60% of that referral traffic now lands on brand homepages, with pageviews per visit up 24% and time on site up 11%.

This is the counter-narrative to the “AI kills referral traffic” story, at least for ChatGPT. When the AI surfaces prominent links, people click them, and the resulting visitors engage more than average. The earlier RESONEO finding that ChatGPT consolidated around fewer cited domains still holds; what changed on 7 May is link prominence, not citation breadth. The brands that were already in ChatGPT’s narrowed citation set are the ones capturing this 150% uplift.

The practical implication is to make sure your ChatGPT citations actually link somewhere useful. If your brand is cited but the link points at a thin page, you’re leaving the engagement uplift on the table. Audit which of your pages ChatGPT cites, and make sure those destinations are built to convert the more-engaged-than-average visitor who arrives.

The SEO-GEO Gap: Top Organic Pages Capture Just 29% of LLM Sessions

Your best organic pages and your best LLM-cited pages are probably not the same pages. Adam Gnuse’s analysis of 10 websites across 150,000 indexed pages found that top-10 organic pages captured 55% of organic sessions but only 29% of LLM referral sessions. The content that wins classical search and the content that wins AI citations are diverging measurably.

The content-type split is the useful part:

Trends and analysis posts: 78% citation rate in LLM traffic.
Year-in-review and original-data posts: 61% citation rate.
Educational how-to guides: 12% citation rate.
Generic comprehensive guides: consistently weak.

The pattern is consistent with everything else in the AI-citation research this month. LLMs reward original data, trends, and analysis (the things they can’t generate themselves) and penalise generic educational content (the things they can). Service and product pages also pulled more LLM sessions per organic session than articles did, which is a useful nudge for any client whose AEO strategy has been all-blog, no-product-page.

The action is a content audit that maps which pages earn organic traffic against which earn AI citations, then a deliberate investment in the original-data and analysis formats that the gap analysis rewards. The two content strategies need to run in parallel, not as one ported onto the other.

The Micro-Macro Shift: Measuring AI Visibility Without Precision

What do you measure when the same prompt gives a different answer every time you run it? That’s the problem this Search Engine Land piece tackles. AI responses are non-deterministic, personalised, and constantly changing, which means the micro-level precision SEOs are used to (this keyword, this rank, this day) doesn’t exist for AI visibility. Chasing it produces noise dressed up as data.

The proposed shift is from micro to macro. Instead of tracking exact citation positions on exact prompts, track trend direction across quarterly periods and across families of related queries. Map the funnel query pathways a buyer actually travels (explore, compare, decide) and measure whether your brand’s presence across those pathways is trending up or down over months, not days. The unit of analysis becomes the query family and the quarter, not the keyword and the day.

This lines up with the measurement reframes from Forrester and Indig in earlier roundups, and it’s the right correction. The agencies that try to report AI visibility with the false precision of a rank tracker will lose client trust the first time the numbers swing wildly for no reason. Macro trend reporting is less satisfying to look at and far more honest about what the data can actually support.

Reddit CEO: LLMs "Would Not Exist" Without Reddit Data

“Modern oil.” That’s how Reddit CEO Steve Huffman described user-generated content’s role in AI this week, arguing that large language models would not exist in their current form without Reddit’s data. He called Reddit “one of the single largest sources of training data” and “the most cited platform across all models.”

The business context matters. Reddit signed data licensing deals with Google and OpenAI around 2024, and Huffman made clear the company stays selective on terms: “commercial use of our data requires commercial terms.” This is the same Reddit whose AI-citation prominence we covered last week from the practitioner angle; the CEO is now saying the quiet part out loud about how structurally dependent the AI engines are on the platform.

For brands, the strategic read hasn’t changed but it’s hardened. Reddit visibility carries outsized weight in AI answers because the engines lean on Reddit so heavily. The catch, as covered last week, is that you can’t manufacture authentic Reddit presence without risking the moderation backlash that costs brands their accounts. The brands getting real AI-visibility returns from Reddit are participating genuinely, not spamming threads with repurposed blog content.

Machine-First Architecture: Building Sites Machines Can Cite

A thoughtful piece this week makes the case for “machine-first architecture”: designing sites for the most constrained consumer (the AI crawler that doesn’t execute JavaScript and parses content literally) on the principle that what works for the machine also works better for everyone else.

The argument is that AI crawlers are effectively the new accessibility constraint. A page that an LLM can cleanly identify, parse, cite, and reuse is also a page that loads fast, has clear semantic structure, and presents its key facts in plain HTML rather than behind interaction. Building for the machine forces the discipline (clean markup, factual clarity, retrievable key claims) that also improves the human experience.

For technical SEO teams, this is a useful framing to take to clients who resist the “rebuild for AI” pitch as speculative. Machine-first architecture isn’t a bet on a specific AI engine; it’s a return to semantic-web fundamentals that pay off across classical search, AI citation, and accessibility at the same time. We’re folding the principle into technical audit recommendations as a way to justify markup and rendering fixes that were always worth doing.

Agent-to-Agent Marketing: Selling to the AI, Not the Human

A new marketing discipline got named this week, at least according to Ahrefs. The premise of “agent-to-agent marketing” is that as AI assistants increasingly research products and make recommendations on behalf of people, the buyer you’re marketing to is sometimes the agent, not the human behind it. The Moltbook example (an experiment where AI agents recommend products to each other) is early and a little contrived, but the underlying shift is real enough to take seriously.

The logic follows directly from everything else in this month’s roundups. If an AI assistant is the intermediary deciding which three brands to surface, the optimisation target moves from persuading a person to being selected by a machine. That favours structured, machine-readable claims, verifiable data, and presence in the sources the agent trusts, over emotional brand storytelling aimed at a human who may never visit your site directly.

It’s too early to build real strategy around this, and we’re filing it under watch-don’t-act for now. But the direction of travel, from marketing to humans toward marketing to the agents that serve humans, is consistent enough across the research that it’s better understood before it becomes a live requirement rather than after.

Google Retires Standalone Display Campaigns for Demand Gen

Google is retiring standalone Display campaigns and folding them into Demand Gen. The phase-out runs gradually through 2027. From June 2026, eligible advertisers get a migration tool in Google Ads; eventually new Display campaigns can’t be created, existing ones stay editable until migrated, and anything left behind gets migrated automatically.

The Display Network inventory itself isn’t going anywhere. It’s relocating into the Demand Gen workflow, where it sits alongside YouTube, Discover, Gmail, and Google Maps placements. The move unlocks features Display campaigns never had: carousel ads, expanded video formats, lookalike segments, generative AI image tools, channel-level reporting, and new bidding options.

The catch is control. Advertisers who have spent years building refined placement exclusions, app exclusions, and brand-safety controls on Display campaigns risk losing some of that granularity in the automatic migration. The strong recommendation across the PPC community is to migrate manually rather than wait for the automatic process, specifically so you can re-establish those controls deliberately in the Demand Gen structure.

For agency-side planning, this is a 2026-2027 project to schedule now. Audit which client accounts run material standalone Display spend, prioritise the ones with carefully tuned exclusions, and migrate those manually before the automatic process reaches them. Leaving it to Google’s auto-migration is the option most likely to quietly degrade brand-safety setups that took years to build.

Google Ads Will Delete Granular Reporting Data After 37 Months

Starting 1 June 2026, Google Ads will delete granular historical reporting data once it passes set retention windows. Hourly, daily, and weekly reporting data will only stay available for 37 months. Monthly, quarterly, and annual reports are kept for 11 years. Reach and frequency metrics get the strictest limit of all: three years.

Once a retention window expires, the data is gone from both the Google Ads interface and the API. There’s no archive to request it back from. For any agency or client that does long-term trend analysis, year-on-year seasonal benchmarking, or multi-year performance modelling at daily or weekly granularity, this is a real loss unless you act before the window closes.

The action is unambiguous and time-sensitive: set up automated export pipelines to warehouse granular Google Ads data before 1 June, and backfill as much historical daily and weekly data as the API still holds right now. BigQuery is the obvious destination for most setups. This is the kind of housekeeping that’s invisible until the day a client asks for a five-year daily trend and you discover the underlying data was deleted 25 months ago. Do it this week.

OpenAI Confirms Conversion-Focused ChatGPT Ads for June

OpenAI confirmed that conversion-focused ad campaigns are coming to ChatGPT in June, with tracking infrastructure that lets advertisers measure the actions ads drive after a user interacts with them. This is the next step on from the self-serve Ads Manager launch covered earlier this month, which started with CPC-based campaigns.

The shift from awareness-priced CPC to conversion-optimised campaigns with proper measurement is the moment a new ad channel becomes seriously plannable. Until you can attribute conversions, an ad surface is a brand-awareness experiment. Once conversion tracking exists, it competes for performance budget on its own merits. June is when ChatGPT ads stop being an experiment for most advertisers and start being a line item that needs a real ROAS target.

We’d still approach it cautiously. Conversion tracking is only as trustworthy as the attribution model behind it, and a brand-new ad platform’s first conversion-measurement implementation deserves scepticism until it’s validated against your own analytics. Run controlled tests, compare ChatGPT’s reported conversions against your GA4 and back-end data, and don’t shift meaningful budget until the numbers reconcile.

Google Ads Real-Time Policy Reviews Speed Up Approvals

Google launched Real-Time Policy Reviews in Google Ads, giving advertisers instant editorial feedback during ad creation rather than after submission. Compliant ads can serve almost immediately instead of sitting in a review queue, and policy issues surface while you’re still building the ad, when they’re cheap to fix.

For high-volume accounts and agencies running frequent creative refreshes, this removes one of the genuinely annoying friction points in Google Ads. The hours-to-days limbo between submitting an ad and learning it was disapproved for a fixable reason has been a standing tax on campaign velocity. Real-time feedback during creation should cut that materially.

The thing to verify is how the real-time checks handle borderline cases. Instant feedback is only useful if it’s accurate; a system that over-flags compliant ads in real time would be worse than the queue it replaces. Watch the false-positive rate on regulated-vertical accounts in the first few weeks before assuming the new flow is strictly better.

How to Get Google Ads Seen in AI Overviews

With ads now appearing inside AI Overviews and AI Mode, how do you actually get a client’s ads to surface there? A practical Search Engine Land guide this week laid out the levers. The short version: it’s less about new ad types and more about feeding Google’s AI the signals it needs.

The factors that matter, per the guide: well-optimised product and service feeds, contextual landing page content that matches conversational intent, conversational ad creative rather than keyword-stuffed copy, and strong audience signals. The campaign types positioned to surface in AI Overviews are Shopping, Performance Max, and AI Max, which is consistent with Google steering advertisers toward its AI-mediated campaign products.

For PPC teams, the practical move is to treat feed quality and landing-page contextual depth as AI Overview ranking factors, not just conversion-rate factors. The same feed hygiene and landing-page relevance work pays off twice now: once for conversion rate and once for AI Overview ad eligibility. That’s a rare two-for-one in a year of mostly new, separate workstreams.

YouTube Auto-Detects and Labels AI-Generated Video

YouTube is now automatically detecting undisclosed photorealistic AI-generated video and labelling it for viewers, alongside repositioning its disclosure labels more prominently. This extends the SynthID-in-Search verification push covered two weeks ago into the video layer, and it shifts the burden: creators who don’t disclose AI content may find YouTube discloses it for them.

For brands and creators using AI-generated or AI-assisted video, the practical question is whether the auto-detection affects reach or trust. YouTube hasn’t said labelled content gets deprioritised, but a prominent “AI-generated” label changes how viewers receive a video. The safe move is proactive disclosure: label your own AI content clearly and control the framing, rather than letting YouTube’s auto-detection apply a label you didn’t choose.

The wider trend across this month is consistent. AI content verification is moving from opt-in creator honesty toward platform-enforced detection, on both Google Search and YouTube. Plan for AI-content labelling to become standard across every major platform within the next year.

Google Appeals Search Monopoly Ruling as Brussels Weighs a Fresh Fine

Google formally appealed the US search monopoly ruling on 25 May, the latest move in the long-running antitrust saga. In parallel, the EU is reportedly weighing a fresh fine over Google favouring its own services in search results. Two regulatory fronts, both live, both with potential to reshape how Search operates in different markets.

For search marketers, the near-term impact is minimal; appeals and EU deliberations move slowly, and nothing changes in the SERP this week. The longer-term watch item is whether any remedy forces structural changes to how Google presents its own services versus third parties, which could open or close visibility opportunities depending on the outcome. It’s a slow-burn story rather than an actionable one, but it’s the regulatory backdrop against which every AI search change this year is happening.

The thing to keep in mind is that Google is making its biggest AI-search product changes in a decade at the exact moment its core search business is under the heaviest regulatory pressure in its history. Those two facts are not unrelated, and the tension between them will shape what Google ships through 2026 and 2027.

Strategic Direction: Where Search Marketing Is Heading

Three threads run through this week.

First, Google is consolidating advertisers onto its AI-mediated products and tightening the data it gives back. Standalone Display folds into Demand Gen. Granular reporting data gets a 37-month shelf life. Ads surface inside AI Overviews through Shopping, Performance Max, and AI Max. The pattern is more automation, more AI mediation, and less granular control or historical data for the advertiser. Agencies need to get comfortable adding value through strategy and measurement discipline rather than manual lever-pulling, because the levers are disappearing.

Second, the evidence keeps stacking up that AI search is a genuinely different discipline from classical SEO, not an extension of it. The SEO-GEO gap analysis found top organic pages capture barely a quarter of LLM sessions. The micro-macro measurement piece argues precision tracking is dead for AI visibility. Preferred Sources and the highly-cited label reward original reporting over optimised content. Every serious piece of research this month points the same way: original data, machine-readable structure, and genuine authority win AI citations; generic optimised content doesn’t.

Third, the transparency and trust layer is hardening. Pichai conceding AI Overviews are over-opinionated, YouTube auto-labelling AI video, SynthID in Search, the regulatory pressure on both sides of the Atlantic. The era of AI features shipping without accountability is ending, and the next phase will be defined as much by verification, labelling, and regulation as by capability.

The agencies that thrive through the back half of 2026 will be the ones that rebuild their measurement and content strategies around these realities rather than retrofitting old playbooks. The direction is set; the work is in the execution.

Key Takeaways

Set up automated exports of granular Google Ads data before 1 June. Hourly, daily, and weekly reporting data gets deleted after 37 months with no recovery path, so warehouse it in BigQuery now and backfill what the API still holds.
Schedule manual migration of standalone Display campaigns to Demand Gen, prioritising accounts with carefully tuned placement and brand-safety exclusions. Don’t let Google’s automatic migration degrade controls you spent years building.
Run a content audit that maps organic-winning pages against AI-cited pages. They diverge: top organic pages capture only 29% of LLM sessions, and trends/analysis/original-data content earns far more AI citations than how-to guides.
Report AI visibility as a macro quarterly trend across query families, not micro per-prompt precision. False precision in AI visibility reporting will cost client trust the first time numbers swing for no reason.
Prompt loyal readers to mark client sites as Preferred Sources. With 345K selected and a 2x click-through advantage, it’s a concrete, audience-led AI visibility lever.
Treat Pichai’s “bounce clicks” framing of publisher traffic loss with scepticism. Plan for AI Overview click suppression to continue regardless of how Google labels it.
Proactively label AI-generated video on YouTube before the auto-detection labels it for you. Control the framing rather than inheriting a label you didn’t choose.

Frequently Asked Questions

When exactly do I need to export my Google Ads data?

Before 1 June 2026, when the new retention policy takes effect. Hourly, daily, and weekly reporting data will only be available for 37 months going forward, and once a window expires the data is gone from both the interface and the API. Set up automated exports to a warehouse like BigQuery now, and backfill as much granular history as the API still holds.

Do I need to do anything about Display campaigns right now?

Not this week, but plan for it. The migration tool arrives in June 2026 and the phase-out runs through 2027. The recommendation is to migrate manually rather than wait for automatic migration, especially on accounts with refined placement exclusions, app exclusions, or brand-safety controls, since the automatic process risks losing some of that granularity.

What’s the takeaway from the SEO-GEO gap research?

Your best organic content and your best AI-cited content are usually different pages. Top organic pages captured 55% of organic sessions but only 29% of LLM referral sessions in the study. AI engines favour original data, trends, and analysis, and penalise generic how-to content they could generate themselves. Run both content strategies in parallel rather than assuming one transfers to the other.

How should I measure AI visibility if precision tracking doesn’t work?

Shift from micro to macro. Track trend direction over quarterly periods across families of related queries, mapped to the funnel pathways buyers actually travel, rather than exact citation positions on exact prompts. AI responses are non-deterministic and personalised, so per-prompt precision is noise. Quarterly trend reporting across query families is the honest unit of measurement.

Are ChatGPT ads worth testing now that conversion tracking is coming?

They merit a controlled test from June, when conversion-focused campaigns and tracking arrive. The Similarweb data showing a 150% referral lift since ChatGPT started surfacing more links suggests genuine traffic value. But validate ChatGPT’s reported conversions against your own GA4 and back-end data before shifting meaningful budget; a new platform’s first conversion-measurement implementation deserves scepticism.

Conclusion

This week tied a clear bow on May’s bigger story. Google is moving advertisers onto AI-mediated products and giving back less granular data, the research consensus that AI search is its own discipline keeps hardening, and the trust-and-verification layer is tightening across products and regulators. Pichai’s admission that AI Overviews are over-opinionated is the kind of thing that gets quoted for months, but the more consequential items for day-to-day work are the Google Ads data retention deadline and the Display campaign retirement, both of which need action on real timelines.

For agencies, the practical priorities are concrete: export and warehouse granular Ads data before 1 June, plan manual Display-to-Demand-Gen migrations, rebuild AI visibility measurement around macro quarterly trends, and audit the gap between organic-winning and AI-cited content. Do those four things and you’re ahead of most of the market heading into June.

Need help adapting your search strategy for the AI era? Contact the Anicca team for expert SEO and PPC guidance.

Table of Contents

SEO & Algorithm Updates

Google’s AI Search Surface

AI Search Research & GEO

PPC & Paid Media

Trust, Verification & Regulation

Wrap-Up