Weekly AI Marketing & Web Update

Critical WordPress Security Fixes, Portable AI Agent Skills, and a Reality Check on AI Search

← All posts

· 24 min read · by Laura Ferruggia

This week delivered a useful reminder that AI isn't the only thing changing how marketing and web teams operate. Sometimes the most consequential development is a security vulnerability that needs fixing immediately. Sometimes it's an advertising platform quietly changing how campaigns spend money. And sometimes it's new research suggesting the industry might be getting a little carried away with its favorite buzzword.

The biggest developments span WordPress infrastructure, AI coding agents, paid-media optimization, AI-search visibility, and Google Workspace.

There's also a particularly interesting shift happening with AI agent skills: the knowledge and instructions we build for one agent are becoming increasingly portable across platforms.

The broader pattern is that operational infrastructure is becoming more valuable than access to any individual AI model. That includes security, reusable workflows, customer data, and the ability to distinguish genuinely incremental business results from impressive-looking dashboards.

1. Two critical PHP vulnerabilities require immediate attention across WordPress hosting environments

October 21–22 | Critical security update

This is the week's most urgent development for teams maintaining WordPress websites.

On October 21, the PHP project released emergency security updates addressing two critical vulnerabilities: CVE-2026-109618 and CVE-2026-109619.

Both vulnerabilities can potentially allow unauthenticated remote code execution under certain conditions.

That's a considerably bigger problem than somebody finding a clever way to inject JavaScript into a comment field.

These vulnerabilities affect PHP itself, the programming language powering WordPress and a substantial portion of the web.

The PHP Foundation confirmed that the issues involve memory management and error handling, including conditions that can be triggered before application-level code executes.

The affected components are fundamental enough that updating WordPress core, plugins, or themes does not resolve the underlying problem.

Patched PHP versions include:

The PHP Foundation assigned the vulnerabilities a CVSS severity score of 9.2.

WP Engine has already patched its infrastructure

WP Engine confirmed on October 22 that it had completed patching across its managed hosting platforms.

The company initially implemented protective measures through its web application firewall and PHP configuration before deploying the official patches.

WP Engine reported blocking more than one million requests associated with the vulnerabilities during the first 24 hours and said it had observed no successful exploitation attempts.

WP Engine customers do not need to perform a manual PHP update for these vulnerabilities. The provider says its hosted environments have already been patched, including supported legacy PHP environments covered by its long-term-support arrangements.

For websites hosted elsewhere, the situation depends on the hosting provider and the PHP version actually running.

An interesting AI connection

The PHP Foundation also disclosed that AI-assisted security research played a role in investigating the vulnerabilities.

Researchers and organizations involved included OpenAI, Anthropic, security firms, and the PHP Foundation's own engineering team.

The Foundation has been using Claude Mythos and OpenAI security tools alongside conventional auditing, testing, and human security review.

That's a useful real-world example of AI-assisted development producing something more consequential than another benchmark score.

Why it matters

For agencies maintaining dozens or hundreds of websites, infrastructure vulnerabilities can create risks that aren't visible inside the WordPress dashboard.

A website might have the latest WordPress core version, updated plugins, and a properly maintained custom theme while still running on vulnerable PHP infrastructure.

This is also a good reminder that automated code review and WordPress-specific security scanning are only parts of the security picture.

The hosting environment matters.

What to do

Immediately verify PHP patch status across maintained hosting environments.

For WP Engine installations, confirm that the sites are on the managed platform covered by the provider's announcement and retain the security notice for internal records.

For other hosting providers, confirm that the relevant PHP security fixes have been applied. Do not assume that a familiar version number means the environment is vulnerable or patched; some providers use backported fixes.

After any required PHP upgrade, test critical functionality, particularly custom themes, ACF components, Gutenberg blocks, forms, integrations, and ecommerce workflows.

This is an infrastructure security issue, not an excuse to send developers on a three-day expedition through every CSS file.

2. Anthropic donates Agent Skills to a vendor-neutral foundation, making reusable AI workflows a bigger deal

October 22 | Open-standard governance announcement

Anthropic announced that it is donating Agent Skills to the Agentic AI Foundation under the Linux Foundation.

This is not the launch of Agent Skills themselves. The format has existed for some time.

The important development is that the standard is moving into a vendor-neutral governance structure.

Anthropic says more than 70 agent platforms have integrated Agent Skills, including ChatGPT, Cursor, GitHub Copilot, and VS Code.

The company also reports approximately one million published skills and nearly 280 million installations.

The numbers are impressive, but the underlying idea is remarkably simple.

An Agent Skill is essentially a folder containing instructions, typically beginning with a SKILL.md file, along with any supporting scripts or reference materials.

Those instructions teach an AI agent how to perform a specific task.

Rather than explaining the same process repeatedly, organizations can package their standards into reusable instructions.

Why it matters

For teams already using Claude Code, this is a potentially important long-term development.

Consider a development team that has established standards for custom WordPress builds.

Those standards might cover:

Traditionally, this knowledge lives in documentation, individual developers' heads, or project-specific instructions.

Agent Skills provide a way to package that knowledge into reusable workflows.

And because the standard is becoming more broadly supported, the investment is less dependent on one AI vendor.

A skill developed for Claude Code may also be usable by another compatible agent, although tool support and execution behavior can differ.

That's the meaningful distinction.

The value is shifting from knowing how to prompt an agent toward defining how the organization expects work to be performed.

What to try

Build one internal skill around a process that already matters.

For a WordPress development team, a strong candidate would be a pre-completion QA skill for custom theme development.

The skill could instruct the agent to inspect changed files, identify accessibility and SEO concerns, run available tests, and produce a structured review before declaring the work complete.

Test it in Claude Code first.

Then evaluate how well the same skill transfers to another supported agent.

Measure the number of issues identified, how consistently the skill is followed, and whether it reduces human review time.

One important caveat: skills can include executable scripts and instructions that influence agent behavior. Treat third-party skills as software dependencies requiring review, not harmless documents.

Agency implication

This creates an opportunity to turn internal expertise into reusable operational infrastructure.

The strongest agencies won't necessarily be the ones with the largest collection of AI subscriptions.

They'll be the ones that have encoded their standards, procedures, and specialist knowledge into systems that help more employees execute consistently.

That's particularly valuable for organizations trying to expand employees' capabilities beyond their original specialties without sacrificing quality.

3. Google Ads expands customer acquisition and retention controls, including more flexible customer-value bidding

October 22 | Rolling out during October

Google Ads announced several changes designed to help advertisers distinguish between acquiring new customers, retaining existing customers, and pursuing customers with higher expected value.

The changes affect eligible Search, Shopping, and Performance Max campaigns.

Three developments are particularly relevant.

High-value customer acquisition expands to Shopping

Google is extending New Customer High Value Mode to Shopping campaigns.

Previously available in other campaign types, this feature allows advertisers to prioritize new customers who are expected to generate greater long-term value.

Rather than treating every first-time purchase as equally valuable, campaigns can incorporate signals about the type of customer the business wants to acquire.

Customer retention goals become more widely available

Google is expanding retention capabilities across supported campaign types.

The available approaches include:

Re-engagement Mode: Prioritizes customers who haven't purchased recently.

Loyalty Mode: Supports campaigns targeting loyalty-program members with relevant messaging or offers.

Retention Only Mode: Focuses campaigns exclusively on specified existing-customer segments.

These features are rolling out during October, so availability may vary by account.

Percentage-based customer-value multipliers

Google is also introducing percentage-based adjustments to customer value.

Previously, advertisers could use fixed monetary adjustments to communicate that certain customers were worth more.

Percentage multipliers allow those adjustments to scale with transaction value.

For example, a 20% multiplier increases the value signal proportionally whether a customer purchases $50 or $500 worth of products.

That can be more appropriate for businesses with substantial differences in order values.

Why it matters

Advertising platforms are becoming increasingly sophisticated at optimizing toward the value advertisers provide.

The problem is that many advertisers still provide incomplete or poorly structured value signals.

A business might know that repeat customers are substantially more profitable than first-time customers.

It might also know that certain product categories attract customers with higher lifetime value.

But if those distinctions aren't reflected in campaign data and bidding configuration, the advertising platform cannot optimize toward them effectively.

These changes provide more ways to translate business economics into campaign strategy.

However, percentage-based value adjustments change the signals used by automated bidding. They do not guarantee a corresponding increase in actual revenue or customer profitability.

What to try

Identify an ecommerce account with reliable customer-history data.

Evaluate whether the account can accurately distinguish new customers, returning customers, lapsed customers, and loyalty members.

Then review whether the expanded acquisition or retention goals support a meaningful business objective.

For example, a retailer might test whether prioritizing high-value new customers produces better long-term results than maximizing immediate purchase volume.

Before implementing percentage multipliers, establish the actual economic justification for those adjustments.

Making every customer appear 30% more valuable because the interface allows it is not a measurement strategy.

Agency implication

These developments strengthen the connection between paid-media management and customer lifecycle strategy.

Agencies can provide greater value by helping clients translate retention, lifetime value, and customer segmentation into advertising decisions.

The opportunity is particularly strong when paid-media specialists work closely with CRM, ecommerce, and analytics teams.

4. Performance Max gets new creative-sourcing controls, while Smart Bidding Exploration expands to Shopping and PMax

October 22–23 | Account-dependent availability

Google Ads introduced two developments this week that could materially change how automated campaigns operate.

One concerns creative assets.

The other concerns bidding flexibility.

Both deserve attention because they give Google's systems more room to make decisions on behalf of advertisers.

Performance Max can source additional images and videos automatically

New settings spotted in Performance Max allow Google to pull creative assets from existing online content.

The additions include:

Landing page images: Allows Google to source images directly from an advertiser's website.

Sourced videos: Allows Google to use video content from sources such as landing pages, linked YouTube channels, and potentially other connected online properties.

One advertiser reported that Google could automatically add up to ten sourced videos to an asset group.

The new settings appear alongside existing options for text customization, image enhancement, and video enhancement.

Importantly, the advertiser who identified the feature reported that Sourced videos was enabled automatically.

That observation should not be interpreted as confirmation that every account has the feature enabled by default.

But it is a good reason to inspect campaign settings.

Why the creative change matters

This could reduce the time required to build Performance Max asset groups, particularly for advertisers with extensive product photography and video libraries.

But it also introduces creative-control risks.

A campaign might source imagery that is outdated, poorly cropped, off-brand, or not licensed for advertising use.

A video created for an organic social campaign may not be appropriate as a paid advertisement.

And a website may contain images that were never intended to become standalone advertising assets.

Convenience is wonderful until the platform decides that a random image from the footer deserves its own advertising budget.

Smart Bidding Exploration expands

On October 23, Google expanded Smart Bidding Exploration to eligible Shopping and Performance Max campaigns.

The feature was previously available for Search campaigns.

It gives Google's bidding system more flexibility to pursue conversion opportunities that might otherwise be excluded by strict Target ROAS settings.

Advertisers can select a Target ROAS tolerance between 5% and 30%.

For example, a campaign with a 200% Target ROAS and a 10% tolerance would effectively allow exploration against a 180% ROAS target.

This is not a new targeting method.

It does not automatically expand the keywords or audiences a campaign can reach.

Instead, it allows the bidding system to compete more aggressively for eligible opportunities that might otherwise be passed over.

Eligibility matters: Campaigns must use Target ROAS bidding and should not be limited by budget. Additional restrictions apply to portfolio bidding strategies and Search Ads 360.

What to try

First, review Performance Max asset optimization settings.

Check whether Landing page images or Sourced videos are available and enabled.

Confirm that any automatically sourced creative is appropriate, current, and authorized for advertising.

Second, identify a Shopping or Performance Max campaign with sufficient budget and a Target ROAS strategy that may be restricting volume.

Evaluate whether a controlled Smart Bidding Exploration test could produce additional profitable conversions.

Do not enable it simply because the campaign would benefit from more conversions.

The important question is whether the business can accept the lower effective ROAS target in exchange for additional volume.

Agency implication

Automated advertising is increasingly about defining the boundaries within which algorithms operate.

Advertisers still need specialists to determine which assets are appropriate, which performance tradeoffs make economic sense, and how to evaluate the results.

The agency's value isn't disappearing because Google can make more decisions.

It's shifting toward making sure Google is making those decisions with the right information and constraints.

5. New research shows AI Overviews dominating commercial searches — but ChatGPT referrals remain tiny compared with Google

October 21–22 | New research, not a Google algorithm announcement

Two studies published this week provide an interesting reality check on AI search.

The first suggests AI-generated answers are becoming increasingly common for commercially valuable Google searches.

The second suggests standalone AI assistants still contribute a relatively small portion of measurable website traffic.

Both findings matter, especially when deciding how much time and money to invest in AI-search optimization.

AI Overviews appeared on 88% of tracked commercial queries

Semrush analyzed 400 U.S. desktop keywords across banking, ecommerce, and software.

The study tracked five types of commercial searches:

Across the sample, AI Overview coverage increased from approximately 46% in September 2025 to 88% in August 2026.

Review-related searches experienced particularly dramatic growth.

AI Overviews appeared for approximately 30% of tracked review searches at the beginning of the study and nearly 90% by the end.

For “best X for Y” searches, coverage reached approximately 96%.

Ecommerce experienced the largest increase among the three industries studied, rising from roughly 33% to 85%.

These are substantial changes.

But the methodology matters.

This was a defined sample of 400 commercial keywords, not a measurement of every commercial search on Google.

The study also did not establish whether AI Overview growth caused corresponding changes in clicks or conversions.

ChatGPT generated less than 1% as much referral traffic as Google organic search

A separate analysis by SEO researcher Marie Haynes examined more than 145 million sessions across 100 websites.

According to the study, ChatGPT generated approximately 0.62% as many sessions as Google organic search.

Including referrals from other large language models increased that figure to approximately 0.73%.

That's 0.73% of Google organic traffic, not 0.73% of all website traffic.

The conversion data was also interesting.

Among the 55 websites tracking conversions, ChatGPT visitors reportedly converted at 13.8%, compared with 3.02% for Google organic visitors.

However, Google organic search still generated approximately 27 times as many conversions as all tracked LLM referrals combined.

The study has limitations, including incomplete details about industry mix, reporting periods, and conversion definitions.

AI referral attribution is also imperfect.

Still, the results provide a useful counterweight to claims that traditional SEO is becoming irrelevant.

Why it matters

These studies illustrate a distinction that marketing teams need to understand.

AI is becoming a major part of Google Search, but standalone AI assistants are not necessarily becoming major sources of website traffic at the same pace.

Those are different developments.

For many businesses, Google organic search remains substantially more important than referral traffic from ChatGPT or other AI assistants.

At the same time, the search results themselves are changing.

AI Overviews increasingly appear for comparison, review, and purchase-research queries that have traditionally been important to SEO and conversion strategy.

That means AI-search optimization cannot be treated as completely separate from conventional SEO.

What to try

Start by reviewing commercial-intent keywords.

Prioritize comparisons, alternatives, pricing, reviews, and “best for” queries.

Record whether AI Overviews appear, which sources receive citations, and how competitors are represented.

Then compare that visibility against existing rankings, organic traffic, and conversion performance.

For AI referral reporting, evaluate actual traffic and business outcomes before deciding whether additional investment is justified.

Avoid assigning equal priority to every AI platform simply because they all have impressive demos.

Agency implication

The strongest opportunity is integrating AI-search visibility into established SEO and content strategy.

Comparison pages, product information, service explanations, reviews, and authoritative supporting content remain valuable.

What changes is how those assets are discovered, summarized, and presented.

The goal should be stronger visibility across the search journey, not creating an entirely separate collection of questionable GEO deliverables.

6. Google Workspace adds conversational BigQuery analysis in Sheets, while Vids gets more practical editing controls

October 19–23 | Phased rollouts and plan-dependent access

Google Workspace introduced several useful improvements this week.

Two are particularly relevant to marketing and content operations.

Gemini can now analyze BigQuery data directly through Google Sheets

Google announced a new capability allowing eligible users to analyze connected BigQuery datasets conversationally through Gemini in Google Sheets.

After connecting a spreadsheet to BigQuery through Connected Sheets, users can ask questions about the underlying data using natural language.

Gemini can query the full connected dataset rather than limiting its analysis to the rows currently displayed in the spreadsheet.

It can also generate connected charts, pivot tables, and dashboard-style views.

For example, a user could ask Gemini to compare quarterly sales by region, identify the products contributing most to growth, and build a reporting view.

Gemini can propose an analysis plan and create a new spreadsheet tab containing connected objects.

Those objects can remain linked to the underlying data and support refreshes.

Availability: Gradual rollout beginning October 19, with access limited to eligible Workspace plans and users who already have permission to query the connected BigQuery data.

Why it matters

This could reduce the number of routine reporting requests that require assistance from a data analyst or developer.

For organizations already using BigQuery, it creates a more accessible interface for exploring large datasets.

A marketer could investigate campaign performance, customer trends, or sales patterns without manually writing SQL for every question.

However, this is not a replacement for data governance or analytical judgment.

Users still need to understand what the underlying metrics represent and whether the generated analysis answers the correct business question.

Google Vids gains independent audio editing and playback-speed controls

Google also announced two practical editing improvements for Google Vids.

Playback-speed controls allow users to speed up clips to 2x or slow them down to 0.25x.

Audio extraction allows users to separate a video's audio track from the visual clip.

Once extracted, the audio can be trimmed, moved, or reused independently across the timeline.

These aren't revolutionary capabilities in the broader video-editing market.

But they're useful additions to a lightweight production environment.

For marketing teams producing short explainers, tutorials, internal training, and social video, fewer trips to a separate editing application can make a meaningful difference.

Both features are rolling out gradually and are available across Google Workspace and personal Google accounts.

What to try

For teams already using BigQuery, test Gemini in Connected Sheets against a real reporting question using approved data and existing permissions.

Compare the generated analysis with a known, validated report.

For content teams, take a short product walkthrough or educational video and test whether Google Vids can handle the required pacing, audio adjustments, and final edits without a more sophisticated video editor.

These are native Google capabilities, but organizations should still follow their existing Workspace administration, data-access, and compliance policies.

A Google Workspace BAA or approved Google environment should not be interpreted as blanket permission to connect separate third-party AI vendors.

Agency implication

These developments support a broader shift toward cross-functional execution.

Marketing employees can increasingly perform routine data exploration and lightweight media production without relying on specialized teams for every step.

That doesn't eliminate the value of data analysts or professional video editors.

It allows those specialists to focus on work where their expertise produces the greatest return.

Also worth knowing

The ChatGPT referral-tracking problem from last week was a bug — and it's been fixed

Last week's update covered an apparent change to ChatGPT outbound links that was causing AI referral traffic to be misclassified as Direct traffic.

There's now a meaningful resolution.

OpenAI confirmed that the missing UTM parameters were a known bug.

The fix began rolling out October 15.

According to reporting published October 20, tagged ChatGPT clicks had recovered to approximately 95–100% of their previous levels by the morning of October 15.

For analytics reporting, the recommended action is to annotate October 8–15 as an affected attribution period.

Do not interpret the temporary decline as evidence that AI-search visibility or traffic genuinely collapsed.

And don't rebuild your attribution model because somebody accidentally broke a tracking parameter.

Google Docs introduces Docs Live for voice-driven document creation

Google began rolling out Docs Live to eligible Workspace users on Android and iOS.

The feature allows users to create structured documents through natural conversation.

It can use permitted context from Gmail, Drive, and Chat to help draft materials such as project proposals, meeting summaries, and presentation outlines.

This could be useful for employees who prefer dictation or frequently need to capture ideas while away from their desks.

Access depends on the organization's Workspace plan and Gemini configuration.

Because the feature can reference Workspace information, organizations should confirm that its use aligns with existing internal data-handling policies before recommending it broadly.

No major new Figma, ACF, Elementor, or WordPress core announcement warrants repeating last week's coverage

The previous update covered Figma's substantial design-to-development announcements.

Those developments remain worth testing, but they are not new this week.

Similarly, the October 6 WordPress core security release has already been covered.

The PHP vulnerabilities discussed above are a separate, genuinely new security issue.

There is no reason to fill every weekly update with recycled product announcements merely because a category appears on the editorial checklist.

What we should actually do this week

These actions are ranked by urgency, operational impact, and the likelihood of producing a useful result.

1. Verify PHP security patch status across every maintained hosting environment.

This is the highest-priority action.

Confirm that affected PHP environments have received the October 21 security fixes.

WP Engine reports that its managed environments are already patched, but installations on other hosting platforms should be verified individually.

Document the results and investigate any unresolved exposure immediately.

2. Audit Performance Max creative-sourcing settings.

Check whether Landing page images and Sourced videos are available or enabled.

Review which assets Google could potentially use and confirm that automatically sourced content is appropriate for paid advertising.

This is a quick audit with the potential to prevent some spectacularly stupid creative decisions.

3. Build one reusable Agent Skill around an existing agency process.

Start with a high-value, repeatable workflow.

A custom WordPress development QA skill is a strong candidate.

Test it in Claude Code and evaluate whether the instructions can be reused in another supported environment.

Focus on consistency and measurable reduction in review work.

4. Review Google Ads customer-value and retention opportunities.

Identify ecommerce clients with reliable customer-history data.

Evaluate whether high-value customer acquisition, retention goals, or percentage-based value adjustments could improve bidding strategy.

Do not implement changes without a clear business justification.

5. Test Smart Bidding Exploration on an eligible campaign.

Choose a Shopping or Performance Max campaign using Target ROAS with adequate budget.

Evaluate whether accepting a modest ROAS tolerance could produce additional profitable conversions.

Monitor actual business results rather than assuming additional conversion volume automatically means improvement.

6. Expand commercial-intent AI Overview monitoring.

Prioritize comparison, review, alternatives, and pricing queries.

Track which brands and pages Google cites, how prominently sources appear, and whether the results create opportunities or risks.

Connect the findings to established SEO and conversion reporting.

7. Test Google's native Workspace improvements where they fit existing workflows.

For eligible organizations using BigQuery, evaluate conversational analysis in Connected Sheets.

For content teams, test Google Vids' expanded editing capabilities.

Keep experiments within approved Workspace services, permissions, and data-handling policies.

8. Correct AI referral reporting for October 8–15.

Annotate the affected period and account for the temporary ChatGPT UTM tracking bug.

Avoid drawing performance conclusions from the attribution anomaly.

The bigger pattern

This week's developments highlight three important shifts in marketing and web operations.

First, the infrastructure surrounding AI-generated work is becoming increasingly important.

Anthropic's decision to move Agent Skills into a vendor-neutral foundation is significant because it makes reusable operational knowledge less dependent on a particular AI platform.

Organizations can increasingly invest in documented procedures, standards, and reusable agent instructions that survive changes in their preferred tools.

That's a more durable advantage than mastering the latest prompt-writing trick.

Second, advertising automation is becoming more sophisticated, but also more dependent on the quality of human decisions.

Google can source creative assets automatically, explore additional bidding opportunities, and optimize toward more nuanced customer-value signals.

But those capabilities only help when the advertiser has accurate data, appropriate creative, and a realistic understanding of customer economics.

Automation doesn't eliminate strategic decisions.

It makes bad strategic decisions easier to execute at scale.

Third, AI search is changing the search experience faster than it's replacing established sources of website traffic.

AI Overviews are becoming common across commercial Google searches.

At the same time, available referral data suggests traditional Google organic search still generates dramatically more website visits and conversions than standalone AI assistants.

That argues for evolving SEO rather than abandoning it.

The opportunity is to understand how AI changes discovery, comparison, evaluation, and attribution while continuing to invest in the channels that produce measurable business results.

For agencies, these developments reinforce a broader competitive advantage: the ability to expand what employees can accomplish while preserving the expertise, standards, and operational controls that make the work reliable.

AI makes it easier to produce code, creative, reports, and content.

But the organizations that benefit most will be the ones that turn those capabilities into repeatable processes with clear accountability and measurable outcomes.

Because the goal isn't to become an agency where everyone can generate everything.

It's to become an agency where more people can execute valuable work independently, without turning quality assurance into an archaeological investigation of everything the robots did.

Sources

AI Marketing Paid Media Agent Skills WordPress Security