The biggest thing this week is that Google is making publisher preference and in-SERP generated experiences materially more important, which has direct consequences for SEO strategy and what agencies should be selling. There’s also a useful signal from local AI citation research, plus one coding-platform move worth watching.
1. Google just gave publishers a real “Preferred Sources” growth lever
On August 20, Google updated Search Central with a new embeddable Preferred Sources button. A user who clicks it can designate your domain as a preferred source and then gets returned to your site. Preferred sites are more likely to appear in Top Stories, and can get a preferred badge inside AI Overviews and AI Mode. Google says readers who mark a site as preferred are about twice as likely to click through to it. (Google for Developers)
This is consequential because it’s one of the first genuinely actionable audience → AI visibility mechanisms. Instead of trying to reverse-engineer some mystical “GEO optimization” bullshit, you can directly encourage loyal readers to increase your prominence in Google’s AI surfaces.
What to try: add the button to publisher/content-heavy client sites, especially blogs, local publications, thought-leadership properties and recurring-resource sites. I’d treat this almost like an SEO-adjacent newsletter CTA: article footer, sidebar, author pages, perhaps after a second pageview rather than obnoxiously above the fold. Google’s recommended implementation is literally two HTML lines, so this is a cheap experiment. (Google for Developers)
Agency angle: this belongs in an AI Search Readiness / AEO implementation package. Not as a standalone upsell—that would be hilariously flimsy—but as one concrete deliverable alongside crawlability, entity clarity, citation monitoring, structured data and content refresh work.
2. Google is starting to build interactive tools inside AI Overviews
Google’s generative UI is expanding beyond AI Mode into AI Overviews. Instead of merely summarizing pages, Google can dynamically create interactive layouts, diagrams, calculators, simulations and other mini-tools directly in the SERP. (Search Engine Journal)
This is potentially a nastier zero-click development than ordinary AI summaries. A page ranking because it provides a mortgage calculator, comparison widget, estimator, converter, quiz or interactive explainer may increasingly find Google recreating the function of the page rather than just summarizing its prose.
That changes how I’d think about “tool content.” Interactive assets are still valuable for links, conversion and brand authority—but “we built a calculator, therefore Google must send us traffic” is becoming a weaker assumption.
What to try: inventory client pages whose traffic depends heavily on a simple utility. For those, add value Google can’t trivially synthesize: proprietary data, saved results, downloadable outputs, personalized recommendations, integrations, historical comparisons, expert interpretation, lead capture, or account-level functionality.
Agency angle: there’s a good consulting/service opportunity around SERP substitution risk audits: identify content Google can replace versus content that retains reasons to click.
3. Google’s August spam update finished — now is the time to inspect volatility
Google’s third spam update of 2026 ran from August 18 through August 21, taking roughly 2 days and 16 hours. Google announced no new spam policies with it. (Google Search Status)
The important part isn’t “OMG ALGORITHM UPDATE.” It’s that you can now cleanly compare client performance around the rollout window without waiting for further turbulence.
What to try: flag any material Search Console changes beginning Aug. 18, but don’t attribute every sneeze to the update. Prioritize sites with scaled AI content, aggressive programmatic SEO, parasite-style pages, thin affiliate content, or recent mass publishing. If ordinary strong editorial pages moved slightly, I would not start chainsawing them because Google sneezed.
For AI-content workflows specifically, the takeaway remains boring but useful: AI generation is not itself the risk; low-value scaled publishing is. Google’s current generative-search guidance still explicitly centers conventional SEO fundamentals and non-commodity content rather than some separate magical AEO markup layer. (Google for Developers)
4. Local AI search appears to favor the business’s own website more than you might expect
A new study covering 14,472 Gemini citations across 1,487 local queries found that business websites accounted for nearly 60% of citations. The catch: repeated identical queries frequently returned different sources, showing significant volatility. (Search Engine Land)
That matters for agencies because local AEO isn’t just “get mentioned on Reddit and Yelp.” The client’s actual site remains a major source candidate.
The practical strategy here is very SEO-ish: strong service/location pages, clear business attributes, accurate hours/services/locations, corroborating third-party profiles, reviews, and locally specific content. In other words: local SEO did not die; somebody just stapled an LLM onto it and invented seven new acronyms.
Agency angle: local SEO packages should increasingly include AI recommendation visibility tracking and prompt-based competitive audits, but I would position this as an extension of local SEO rather than pretending it is an entirely alien discipline.
5. AI Overview tracking is still crappy, but there’s a clever GA4 workaround worth testing
A nine-month dataset published this week tracked 51,200 AI Overview referral events and found that Google sometimes appends #:~:text= fragments when users click cited passages. The authors used that pattern as a GA4 custom dimension to partially isolate AI Overview traffic. They also found citation lifecycles were volatile: some snippets rose, faded or emerged months after publication. (Search Engine Land)
This is not a perfect attribution method, so I would absolutely not turn it into a bullshit dashboard claiming “100% accurate AIO traffic.” But it’s good enough for experimentation.
What to try: implement the fragment-based GA4 dimension on one or two sites and compare it against citation-tracking data over a month. If it behaves reasonably, this could become part of your SEO reporting stack.
Agency angle: there’s a legitimate emerging deliverable here: AI-search visibility + referral attribution reporting, provided you communicate uncertainty instead of painting a very expensive graph and pretending epistemology has been solved.
6. llms.txt v2 exists; still do not worship the file
The llms.txt proposal received a v2 update this week, adding more formal linking to Markdown versions of pages and an HTTP-header mechanism for agents. But Google’s own guidance has previously clarified that llms.txt is not required for Google Search’s AI features. (Search Engine Journal)
My recommendation remains: if implementation takes fifteen minutes and fits your stack, fine. Add it. But do not sell “we installed llms.txt” as meaningful AEO transformation. That is the 2026 equivalent of charging someone $800 to submit their website to Yahoo.
7. Cursor launched its own GitHub competitor — watch the direction, not necessarily the product yet
Cursor launched Origin, an early-beta source-code hosting platform with repos, PRs, code browsing/editing and GitHub synchronization. The obvious strategic play is to make the entire development environment agent-native, instead of bolting AI coding onto a Git-centric workflow designed for humans. (IT Pro)
For agency web development, this is more important as a direction-of-travel signal than something I’d migrate production repos to Monday morning.
The coding-agent competition is shifting from “which model writes better PHP?” to who owns the repository, agent context, review loop, deployment pipeline and observability layer. That could radically simplify small WordPress/site work over the next year.
What to try: nothing production-critical yet. Keep watching Origin, Claude Code, Codex and competing agent-native workflows. The interesting metric isn’t benchmark score; it’s whether you can reliably assign something like “update these 20 pages, verify staging, run tests and open the PR” with minimal babysitting.
What I’d actually do this week
If I were prioritizing this for an agency rather than collecting shiny AI Pokémon, I’d do three things:
- Implement/test Google Preferred Sources on one content-heavy property and build it into your AEO checklist.
- Create an AI-search measurement experiment combining GA4 #:~:text= tracking, Search Console and manual/third-party citation checks.
- Add “SERP substitution risk” to content audits, especially for calculators, simple comparison tools and generic informational utilities now that Google can generate interfaces inside AI Overviews.
The larger pattern is getting clearer: AEO is converging back toward strong SEO, audience loyalty and brand authority—not replacing them. Google is simultaneously making AI answers more capable of eliminating clicks and giving known/liked publishers mechanisms to retain visibility. That makes differentiated content and recognizable brands more valuable, while commodity information gets progressively more screwed.