Weekly AI Marketing & Web Update

Visual Agents, AI Search Visibility & the New Marketing Operations Stack

← All posts

· 11 min read · by Laura Ferruggia

This week’s headline: visual agents are getting good enough that “design → Elementor build → responsive QA” is becoming a realistic automation loop, not just a demo.

The most interesting near-term opportunity for us is the combination of GPT-6 Astra’s computer/visual capabilities with Elementor’s own agentic stack and its September MCP roadmap. Paid media is shifting too: ChatGPT Ads is becoming a more measurable performance platform, while Google is beginning to place conventional Search campaigns inside AI Mode.

1. Google Search Console now reports visibility in AI Overviews and AI Mode

As of August 31, Google says its Generative AI performance report has rolled out worldwide in Search Console. The report specifically measures how often a site’s links appear in AI Overviews and AI Mode, giving marketers first-party data on generative-search visibility rather than requiring third-party citation trackers or manual prompt testing.

The report currently focuses on impressions. We can see which pages receive the most AI visibility, how impressions change over time, and break the data down by page, country, device and date. The data can also be exported for reporting or analysis. Google does not currently provide query-level data or a dedicated AI click/conversion metric in this report, and Search Labs experiments are excluded.

That limitation matters, but this is still a major step forward for AEO measurement. Until now, a client could appear frequently in AI Overviews while traditional Search Console reporting made it difficult to isolate that visibility. We can now establish an actual baseline and track whether our content is gaining or losing exposure specifically within Google’s generative experiences.

There is also now a separate Search generative AI control in Search Console. By default, sites remain included, meaning their links can appear in AI Overviews, AI Mode and generative Discover experiences and their content can help ground generated answers. Sites can opt out, but doing so also means giving up impressions and traffic from those features.

Why it matters for us: AEO can increasingly become part of normal SEO measurement rather than a speculative side discipline. We should still care about rankings, clicks, leads and revenue, but we can now add a first-party metric for AI search visibility and identify which types of client content Google is actually using in its generative experiences.

What to try: pull the Generative AI report for our key SEO clients, identify the pages already earning the most AI impressions, and save a September baseline. Over time, we can compare those pages against traditional organic performance and conversions to see which content strategies are gaining visibility in AI Search and whether that visibility produces measurable business value.

Sources:

Generative AI in Discover report:

2. Astra + Elementor could collapse a large part of the web-production loop (Experimentation idea)

What changed. OpenAI released GPT-6 Astra on September 3 with major gains in computer use, browsing, visual judgment, software engineering, and long multi-step work. OpenAI explicitly shows Astra creating websites, troubleshooting software on-screen, and running frontend QA. It also says Astra can accept changing requirements mid-task without losing the original objective. OpenAI’s Astra launch details.

The Elementor side is just as important. Elementor’s Angie agent already operates inside WP Admin and the Elementor Editor, can read site context, create pages, build widgets, and execute site actions. Elementor’s public roadmap now lists an “Elementor MCP” for September 2026 that will connect tools such as Claude and Codex directly to Elementor sites to create and edit pages. The same roadmap lists AI control over headers, footers, archive templates, and display conditions. Elementor roadmap Angie documentation

Why it matters: The interesting workflow is not simply “AI writes HTML.” It is an agent that can both manipulate the structured site and visually judge the result. That gives us a plausible closed loop:

reference design / screenshot → inspect site styles → edit Elementor → preview frontend → compare visually → fix discrepancies → check desktop / tablet / mobile → repeat

The stronger version of this uses two modes together: structured editing through Elementor/Angie/MCP for reliable page changes, plus Astra’s browser/computer use for visual inspection and responsive QA. That is much less brittle than asking an agent to click every Elementor control by coordinates like a caffeinated ghost.

What to try: Run one internal staging-page pilot. Give Astra a Figma export, Claude Design screenshot, or approved visual reference and an existing Elementor page. Measure total time, manual interventions, visual defects, and breakpoint fixes versus the normal build process.

Suggested pilot prompt

Open the Elementor editor for this staging page. Reproduce the attached design using the site’s existing global styles, containers, components and widgets wherever possible. Do not alter global styles unless necessary. Preview the finished page on desktop, tablet and mobile, compare it visually with the reference, and correct visible discrepancies. Preserve existing functionality and do not publish changes without approval. When finished, summarize what you changed and any places where the reference could not be matched cleanly.

Important caveat: This is not yet a plug-and-play “Astra controls every Elementor site” product. Astra access is still rolling out, and Elementor MCP is listed on the roadmap with a September ETA. The opportunity is credible enough to pilot, but we should treat it as an emerging workflow rather than promise it to clients tomorrow.

Agency implication: If the loop works, the upside is not merely fewer developer hours. It can compress the full handoff from creative concept to production-ready responsive page. That changes landing-page economics, rapid campaign launches, design iteration, and potentially the margin structure of routine WordPress builds.

3. ChatGPT Ads is starting to look like a real performance-media platform

What changed. OpenAI has been expanding ChatGPT Ads beyond the U.S.; self-service Ads Manager access became available across 31 European markets on August 31. OpenAI says the platform now supports conversion optimization, geo-targeting, custom audiences, the OpenAI Pixel, Conversions API, and third-party measurement. OpenAI’s advertising update

This week’s more consequential layer is the ad-tech plumbing: new audience-management flexibility, expanded conversion matching, more granular product-level reporting, and a ChatGPT Ads Manager plugin that can create, manage, analyze and troubleshoot campaigns from ChatGPT or Codex with confirmation before changes are applied. Search Engine Land coverage of the September 4 update

Why it matters: This is the point where ChatGPT advertising starts moving from “interesting experimental inventory” toward something agencies can actually plan, measure and optimize. The Ads Manager plugin is especially notable because it puts campaign operations inside the same agent workflow marketers may already use for research, creative, analysis and reporting.

What to try: Identify one or two clients where ChatGPT ads would make strategic sense once inventory and targeting are available: high-consideration purchases, software, professional services, travel, or products where people naturally use ChatGPT to compare options. Build a test framework now: audience hypothesis, conversion event, creative angle, landing page, and acceptable CPA/ROAS.

What to ignore: Do not move meaningful budget just because “AI ads” sounds futuristic. We need enough inventory, targeting quality and conversion data to judge the channel like any other performance platform.

Agency implication: ChatGPT Ads should be on our paid-media roadmap as a new channel competency, not as a novelty. The interesting service opportunity is cross-channel intent capture: organic AI visibility + paid AI placement + landing-page optimization + measurement.

4. Google is testing conventional Search campaigns inside AI Mode

What changed. Google has begun a small experiment allowing text ads from traditional Search campaigns using exact and phrase match keywords to serve in AI Mode when Google detects explicit, direct intent. Advertisers do not necessarily need to move those campaigns into AI Max or Performance Max to participate in the test. Search Engine Land, September 7

Why it matters: AI Mode is not developing as a completely separate advertising universe. Google is testing ways to carry the existing keyword-based Search auction into conversational search where the user’s intent maps cleanly to a keyword. That preserves the relevance of traditional Search skills even as the interface changes.

What to try: For accounts that surface in the experiment, watch placement/query reporting closely and compare AI Mode traffic quality with conventional Search. Do not restructure campaigns just to chase the test.

What to ignore: There is no evidence yet that every exact/phrase campaign will suddenly pour into AI Mode. Google describes this as a small experiment tied to explicit intent.

Agency implication: Clients will increasingly need one paid-search strategy that spans conventional SERPs and AI interfaces. “AI advertising” and “search advertising” are starting to merge rather than remain separate packages.

5. AI Overviews may be making Shopping CTR look better while shrinking reach

What changed. A September 7 analysis of Shopping and Performance Max data found a recurring pattern: impressions have fallen more sharply than clicks, pushing CTR higher. The analyst’s hypothesis is that Google may be serving AI Overviews more often on lower-click-propensity queries and reserving Shopping placements for more commercially valuable searches. The data does not prove causation, but the pattern is worth watching. Search Engine Land analysis

Why it matters: A rising CTR can look like optimization success even when total visibility is contracting. If Google is changing which search experiences are eligible for Shopping ads, historical CTR benchmarks become less useful on their own.

What to try: In ecommerce reporting, look at impressions, absolute clicks, conversion volume, revenue and query mix alongside CTR. If CTR rises while impressions fall, treat that as a diagnostic question, not an automatic win.

What to ignore: Do not present AI Overviews as the confirmed cause. The current evidence is correlational and the analyst explicitly frames it as a hypothesis.

Agency implication: AI search is beginning to distort familiar KPIs. Reporting packages need to explain distribution changes, not just show prettier percentages.

6. Claude watermarking is a governance issue, not an SEO emergency

What changed. Anthropic updated its Claude watermarking documentation on September 1 with current detection-API information. Anthropic says the watermark does not add hidden characters, has negligible effect on output quality, carries no identifying information about the user or organization, and does not change ownership of the output. It is being implemented to comply with the EU AI Act. Anthropic’s watermarking explanation

Why it matters: Teams using Claude for client copy should understand what the watermark actually is before somebody starts inventing an “SEO penalty” narrative. The useful question is disclosure and content-governance policy, not how to evade detection.

What to try: Document when AI is used materially in content production, keep human review for factual accuracy and brand voice, and align disclosure practices with client or regulatory requirements. That is a process issue, not a reason to stop using Claude.

What to ignore: Claims that the watermark inserts visible artifacts, uniquely identifies the user, or inherently hurts Google rankings are not supported by Anthropic’s documentation.

What I would actually do this week

  1. Establish our Search Console AI visibility baseline. Pull the new Generative AI performance report for major SEO clients, record total AI impressions and top-performing pages, and save/export the data as a September 2026 benchmark. Then start incorporating AI Overview/AI Mode visibility into SEO reporting alongside traditional impressions, clicks, conversions and rankings. I’d also look at the highest-AI-impression pages across several clients and see whether any repeatable content characteristics emerge.
  2. Pilot Astra on one staging Elementor page using a real reference design and track build time, manual interventions, and responsive defects.
  3. Check whether our Elementor stack is ready for the upcoming MCP path and monitor the September roadmap release rather than installing random third-party automation just to say we did.
  4. Create a short list of clients that could plausibly test ChatGPT Ads once their market/account has access, with conversion measurement defined before launch.
  5. Watch AI Mode placement behavior and Shopping impression trends, not merely CTR.
  6. Treat Claude watermarking as an AI-governance/process topic, not an SEO fire drill.

The bigger pattern

The market is moving from AI that generates marketing assets toward AI that operates the systems where marketing work happens. That distinction matters. Astra can visually operate software and perform QA. Elementor is exposing page creation and editing to agents. ChatGPT Ads can increasingly be managed through an agent interface. Google is moving existing ad systems into AI Mode.

For an agency, the competitive advantage is therefore becoming less “our team knows how to prompt” and more “our workflows are structured so agents can safely build, inspect, measure and improve real client work.” That is a much more defensible capability — and a much more interesting service story.

Sources / further reading

AI Marketing SEO & AEO Web Technology