Mobile had its paperwork month in August. Digital marketing had a different one. Search, paid social, and retail media all circled the same question from different angles: what counts as AI now, who has to disclose it, and what can you actually measure once you do.
None of it landed as one big headline. A search update here, a law taking effect there, a setting quietly disappearing from an ads dashboard most people never touched until it was gone. Here’s what actually happened across search, paid social, retail media, and AI regulation in August and September, and what to do about each one.
1. Google’s August Spam Update Puts Teeth Behind Its Ban on AI-Search Manipulation
Google released its third spam update of 2026 on August 18, and the rollout wrapped on August 21, roughly two days and sixteen hours later, making it the longest spam update of the year. That’s a separate event from the ranking volatility that rattled trackers in late July and early August. Publishers and third-party tools were convinced something big had hit, but Google’s own Search Status Dashboard showed nothing logged for those dates, and the company never confirmed an update during that window at all.
The part that actually matters for AEO work sits underneath the date. A few months earlier, Google quietly extended its spam policy to cover attempting to manipulate generative AI responses, putting content engineered purely to bait an AI Overview citation in the same enforcement bucket as classic keyword stuffing. This August 2026 spam update doesn’t target anything new, Google says, and it doesn’t specifically go after link spam or site reputation abuse either. But the policy covering AI manipulation was already on the books when the update rolled out, which means it’s now a live enforcement risk rather than a hypothetical one.
Next steps:
- Treat any content built purely to bait an AI citation, rather than answer a real question, as carrying the same risk as any other manipulation tactic. Content that does both is still fine.
- Before blaming a traffic dip on this update, check your dates. A drop starting August 18 through 21 is worth investigating. A drop from the noisy week before it is probably something else entirely.
2. AI Overviews Now Expand Into Full AI Mode Answers by Default
On August 27 and 28, Google confirmed that AI Overviews can automatically grow into a longer, more detailed answer that resembles the AI Mode experience, for queries where its systems decide a fuller response helps the searcher. The manual “Show more” button disappears for those queries. SEO watcher Chris Long spotted the behavior first, and Google’s Robby Stein, who leads the Search product, confirmed the logic behind it: if someone’s already scrolling past the AI Overview, the dynamic expansion cancels so it doesn’t disrupt their reading position.
It’s not a ranking change. It’s an interface change, and it moves in one direction: more of the answer shown before anyone ever reaches a link to a website. A related, earlier update had already brought Google’s generative UI, the feature that builds custom visual layouts on the fly, from AI Mode into AI Overviews as part of a back-to-school rollout, so this expansion is part of a pattern rather than a one-off test.
Action items:
- Weight content strategy toward comparison and evaluation queries where possible. Single-fact informational queries are the most exposed to this kind of expansion.
- Flag this alongside the spam update in client-facing search reporting this month. Neither one is a penalty, and both quietly shift organic traffic for reasons that have nothing to do with a page’s own quality.
3. Search Console’s New Generative AI Reports Are Live Worldwide
As of August 31, Google’s dedicated generative AI performance reports in Search Console are available for every property globally, following a phased rollout that started in June. The reports break out impressions from AI Overviews, AI Mode, and Discover’s AI features by page, country, device, and date.
The limitation is the one that matters most: no click data, and no visibility into the actual prompts that triggered the impression. You can see a page showed up inside an AI answer. You can’t see what someone asked to get there, or how many people clicked through afterward. Google says the rollout is complete, but a site still needs enough AI-feature impressions before the report populates at all, so an empty tab right now doesn’t necessarily mean anything is wrong.
What to do next:
- Add the Generative AI tab to every recurring SEO report starting this cycle. It’s first-party Google data, and for accounts with real AI Overview visibility, it’s the closest thing to ground truth available.
- Pair it with a third-party AEO or AI-visibility tool for what Search Console still won’t show: prompts, competitor citations, and any read on actual traffic.
4. The EU’s AI Act Transparency Rules Take Effect
Article 50 of the EU AI Act became enforceable on August 2. It requires disclosure whenever someone interacts with an AI system, or encounters synthetic audio, image, video, or text that could pass for the real thing. National authorities can enforce it immediately, and the European Commission’s own Article 50 transparency obligations guidance sets fines at up to 15 million euros or 3% of global annual turnover, whichever is higher. Only the machine-readable marking requirement for systems already on the market gets a grace period, running to December 2.
Meta and Google were already building toward this. Meta’s AI info label, tucked inside a Facebook or Instagram ad’s About this ad menu, now applies automatically when it detects C2PA metadata embedded by third-party tools like Adobe Firefly or Canva’s AI features, on top of Meta’s own generative tools. Google embeds its own SynthID watermarks and C2PA metadata into AI-generated ad creative and began rolling those controls out across its advertising products in July. Both systems depend entirely on that metadata, and it doesn’t always behave the way you’d expect: it can survive an edit that removed every trace of AI involvement and trigger a false-positive label, or get stripped by a tool that doesn’t preserve it and let genuinely AI-made content through unlabeled.
Next steps:
- Build a disclosure workflow for any campaign running AI-generated creative into an EU market, not a one-time legal check before launch.
- Confirm which production tools in a client’s stack embed C2PA credentials and which strip them on export, since detection depends entirely on that metadata making it through.
- Don’t assume platform compliance and legal compliance are the same checklist. Meeting Meta’s or Google’s disclosure rules doesn’t automatically satisfy Article 50, and the reverse is also true.
5. Meta Strips Manual Placement Controls From Ad Sets
Starting around August 19, Meta advertisers began seeing a notice inside Ads Manager that excluding individual placements, platforms, devices, or operating systems would no longer be available at the ad set level. Social Media Today first reported the option to exclude placements, based on a screenshot shared by Meta ads specialist Jon Loomer. It’s a separate move from Messenger Stories getting cut as an ad placement entirely on August 27, part of the same v26.0 update rolling out across the month, but worth tracking alongside it since both shrink the same set of manual controls. For now, the placement exclusion change is a test limited to Sales and Leads objective campaigns, with no official rollout date announced.
The direction is familiar. Detailed targeting exclusions went first. Automated placements became the default years ago. This is simply the next manual lever headed the same way, replaced by value rules and account-level controls that apply across every campaign at once rather than one ad set at a time.
Actionable steps:
- If a specific placement has genuinely hurt performance on an account, pull that data now, while the ad set exclusion option is still there to reference against.
- Rebuild any defensible exclusion as a value rule, which lets you bid up or down by placement, device, or mobile OS, or as an account-level placement control. Both survive this change. The ad set setting won’t.
- Plan Q4 budgets assuming less manual placement control and more reliance on creative variety to steer delivery indirectly. An ad set running one creative gives the algorithm nowhere to redistribute spend once exclusions disappear.
6. Amazon Ads Extends Retail Media Into ChatGPT
On September 10, Amazon Ads confirmed a partnership with OpenAI that places a small group of US advertisers inside a ChatGPT ad surface, with Delta Vacations named as one of the first testers. The timing tracks: Sensor Tower reported ChatGPT ads crossing a billion-dollar annualized run rate on September 1, up from roughly $100 million as recently as March. It’s a small pilot rather than a general release, but Amazon, the largest retail media business by a wide margin, chose to sit inside that surface rather than compete against it, and that decision on its own is worth paying attention to.
For a brand already running Amazon DSP or Sponsored Products, the pitch is continuity: the same audience data and measurement infrastructure, extended into a conversational buying moment instead of a separate build from scratch.
What to do next:
- Nothing to activate yet if an account isn’t part of the pilot, but it’s worth a line in next year’s channel planning conversation now, rather than waiting until every competitor is asking the same question.
- Watch for other retail media networks to test similar arrangements once this pilot produces results worth citing. Retail media data is starting to travel across AI surfaces instead of staying locked inside one platform’s results page.
- Measure any future placement here as its own category. A recommendation inside a chat thread doesn’t behave like a search ad or a product feed ad, and folding it into existing ROAS models will misread whatever the pilot actually produces.
The Takeaway
Search, paid social, and retail media all leaned the same direction this cycle: less energy spent proving a system uses AI, more spent measuring what happens once people see it. Google’s new Search Console report shows where a page appeared in an AI answer, not who clicked. The EU’s disclosure law requires a label on AI content, but says nothing about whether that label helps or hurts performance once it’s there.
Two of these already carry deadlines. The EU’s rules are live now, with only the marking requirement’s grace period stretching to December. Meta’s placement change is spreading through test accounts with no published timeline, which is its own kind of urgency, since there’s no way to know when it lands on a given account until it does. The rest belongs in next month’s reporting, built in now rather than explained after a number moves and nobody can say why.
Stay tuned for our next roundup. For more marketing insights, visit the Moburst blog or explore our full range of digital marketing services.
Related Articles
AI Website Builders vs Agencies, Build Buy or Partner
Moburst’s Mobile Marketing Roundup #34
