Google’s latest blog post claims AI is making Search more useful than ever. Google says people are asking new kinds of questions, clicking on more links, and spending more time on the content they visit.
But with no supporting data or clear definitions, the message reads more like reassurance than transparency.
Rather than take Google at its word or assume the worst, you can use your own analytics to understand how AI in Search is affecting your site.
Here’s how to do that.
Google Says: “Quality Clicks” Are Up
In the post, Google says total organic traffic is “relatively stable year over year,” but that quality has improved.
According to the company, “quality clicks” are those where users don’t bounce back immediately, indicating they’re finding value in the destination.
This sounds good in theory, but it raises a few questions:
Look at CTR changes before and after May 2024 (when AI Overviews began expanding).
Pay attention to queries that are longer, question-based, or likely to trigger summaries.
You may find impressions are holding steady or rising while CTR declines. That suggests your content is still being surfaced, but users may be getting their answers directly in Google’s AI-generated response.
2. Approximate “Quality Clicks” With Engagement Metrics
To test Google’s claim about higher quality clicks, you’ll need to look beyond Search Console.
In GA4, examine:
Engaged sessions (sessions lasting more than 10 seconds or including a conversion or multiple pageviews).
Average engagement time per session.
Scroll depth or video watch time, if applicable.
Compare these engagement metrics to the same period last year. If they’re improving, you may be getting more motivated visitors, supporting Google’s view.
But if they’re dropping, it could mean that AI Overviews are sending fewer, possibly less interested, visitors your way.
3. See Which Content Formats Are Gaining Visibility
Google says people are increasingly clicking on forums, videos, podcasts, and posts with “authentic voices.”
That aligns with its integration of Reddit and YouTube content into AI Overviews.
To see how this shift might be playing out for you:
Compare the performance of listicles, tutorials, and original reviews to more generic content.
If you create video or podcast content, track any uptick in referral traffic from Google.
Watch for changes in how your forum threads, product reviews, or community content perform compared to static pages.
You may find that narrative-style content, first-hand experiences, and multimedia formats are gaining traction, even if traditional evergreen pages are flat.
4. Watch For Redistribution, Not Just Declines
Google acknowledges that while overall traffic is stable, traffic is being redistributed.
That means some sites will lose while others gain, based on how well they align with evolving search behavior.
If your traffic has declined, it doesn’t necessarily mean your content isn’t ranking. It may be that the types of questions being asked and answered have changed.
Analyzing your top landing pages can help you spot patterns:
Are you seeing fewer entries on pages that used to rank for quick-answer queries?
Are in-depth or comparison-style pages gaining traffic?
The patterns you spot could help guide your content strategy.
Looking Ahead
When you rely on Search traffic, you deserve more than vague reassurances. Your analytics can help fill in the blanks.
By keeping an eye on your CTR, engagement, and how your content performs, you’ll get a better sense of whether AI in Search is helping you. This way, you can tweak your strategy to fit what works best for you.
A new study commissioned by Meta and conducted by Deloitte finds that advanced personalization strategies are associated with a 16 percentage point increase in conversions compared to more basic efforts.
The research also introduces a maturity framework to help organizations evaluate their personalization capabilities and identify areas for improvement.
What the Data Shows
According to the study, 80% of U.S. consumers say they’re more likely to make a purchase when brands personalize their experiences. Consumers also report spending 50% more with brands that tailor interactions to their needs.
The report connects these behaviors to broader business outcomes. In the EU, Meta’s personalized advertising technologies were linked to €213 billion in economic activity and 1.4 million jobs.
While the economic impact data is specific to Meta, the findings reflect a wider trend in digital marketing: personalized engagement influences purchase decisions and brand loyalty.
Derya Matras, VP for Global Business Group at Meta, commented:
“As people want content and services that are more relevant to them, they are increasingly drawn to brands that make them feel understood.”
Maturity Model for Personalization
The report outlines a four-level maturity model to help you assess where you stand with personalization. The study links higher maturity levels with measurable business outcomes.
Level 1: Low Maturity
Data remains siloed, and messaging tends to be generic. Personalization, if present, is rule-based and limited to a few channels.
Level 2: Medium Maturity
Some systems are integrated, enabling basic audience segmentation and limited customization across channels. These organizations may also use analytics tools and consent management.
Level 3: High Maturity
Unified customer profiles and identity resolution enable greater personalization across multiple touchpoints. Predictive modeling and dynamic content are more common.
Level 4: Champion Maturity
Real-time personalization, generative AI, and clean-room tech support tailored omnichannel experiences. Teams collaborate across departments, with AI governance integrated into decisions.
Three Personalization Strategies
The study outlines three personalization strategies:
Customer-based: Tailors experiences to individuals based on personal data and behavior.
Cohort-based: Segments audiences based on shared traits or behaviors.
Aggregated data-based: Uses anonymized, large-scale datasets to identify general trends.
The report doesn’t suggest a single best method. Instead, it offers examples to help you evaluate what fits your capabilities and goals.
Looking Ahead
For marketers assessing their next steps, the maturity framework offers a structured way to evaluate readiness across people, processes, and technology.
Rather than treating personalization as a software problem, the report frames it as a long-term shift in how organizations structure teams and manage data.
Global companies today face a paradox. Search is more important than ever, yet how it’s managed across markets is often inconsistent, inefficient, and misaligned with broader digital goals.
Too often, SEO is seen as a localized effort, tactically delegated to regional teams or outsourced agencies.
While local knowledge is critical, international SEO success demands structure, governance, and repeatable processes. Otherwise, companies waste resources, duplicate efforts, and fail to capitalize on scalable gains.
This article offers a blueprint for designing an effective SEO organizational structure for global companies, rooted in real-world service-level governance.
We’ll explore what to centralize, what to localize, and how to balance best practices with market nuance.
Drawing from the Service Level Agreement (SLA)-based SEO model used at leading enterprises, we’ll break down the building blocks of a successful international SEO operation, from key performance indicators (KPIs) and tooling to budget models and agency management.
What To Centralize Vs. Localize
An effective SEO structure isn’t just about resourcing; it’s about allocation logic. Knowing which tasks belong at corporate, brand, or market levels prevents duplication, preserves strategic clarity, and empowers those closest to the customer.
This may be one of the most challenging aspects of international SEO operations, particularly for decentralized organizations. You’ll need to evaluate what must be done at each level thoughtfully.
Consider where content is created, how websites are maintained, how diverse market content truly is, and how mature your localization process is.
Unfortunately, there is no one-size-fits-all solution, not even a “one-size-fits-most” option. Each organization must assess its structure, workflows, and existing capabilities.
In many cases, it’s advisable to begin with a few uncontroversial initiatives, such as aligning on what is already established in brand or web standards, content themes, topical coverage, and entity research, and establishing consistent reporting.
Once those foundational elements are in place, you can move toward more sensitive and territorial elements such as Webmaster Tools account management, diagnostic methodology standardization, and global governance of webpage templates.
Centralized Functions (Corporate Center Of Excellence)
These activities are best housed within a corporate SEO function or Center of Excellence (CoE), where scale, tooling, and data access are leveraged across the enterprise:
Training and enablement of brand and market teams.
Tool governance and platform procurement.
Shared Responsibilities (BU And Editorial)
Some functions require cross-functional collaboration between the brand/business unit and central teams:
Editorial workflow integration.
Quarterly content planning tied to search trends.
Performance reviews of strategic campaigns.
Metadata refinement and topics alignment.
KPI alignment between SEO, PPC, and social media.
Localized Responsibilities (Market Or Regional Teams)
Localization is more than just translation. Market teams need autonomy in areas that require cultural fluency, deep customer knowledge, and search behavior insight:
Local-language topic and content research and mapping.
Not all localization adds value. Avoid local divergence when:
The infrastructure doesn’t support market-specific subdomains or folders.
The same product or offer is consistent across regions.
Central models can outperform local improvisation (e.g., PLPs).
There’s limited market-specific search volume or opportunity.
Standardization Of Best Practices
To succeed at scale, international SEO must rely on shared standards that create consistency and reduce avoidable errors.
Standardization accelerates execution and allows for cross-market insights.
Key Elements Of Standardization:
Enterprise SEO Playbook: Documented standards, processes, templates, and escalation paths.
SEO Training Curriculum: Modular training by role type, from content creators to developers.
Content Optimization Templates: Consistent formats for metadata, searcher intent, and markup.
Glossary and Taxonomy: Shared terminology dictionary and content tagging schema.
Governance Reviews: Scheduled audits of adherence to SEO standards by markets and BUs.
Standardization doesn’t mean rigidity. It means creating a foundation that enables innovation and agility at the local level while preserving enterprise-wide integrity.
KPIs That Matter At Each Level
Metrics must reflect both operational performance and business impact, and be meaningful at each layer of the organization.
In one real-world example, a company managing SEO through multiple agencies across markets experienced significant inefficiencies due to inconsistent reporting standards.
Regional and global teams were forced to spend time reconciling disparate metrics, definitions, and formats.
Enforcing consistent KPIs and using standardized reporting templates eliminated this wasted effort, freeing up time for analysis and action rather than reconciliation.
Corporate-Level KPIs
Organic market share growth.
Revenue or lead contributions.
Topical and answer shelf space across global regions.
If data collection and presentation are consistent, it is easy to roll up data across markets and business units to see the total impact on the business, opportunities, and problems.
Consistent and business-oriented metrics are critical to making the business case for continued funding and support of your initiatives.
Ensure KPIs are actionable, standardized across teams and markets, and demonstrate business value to stakeholders.
Process Design & SLA Governance
Clearly defined processes eliminate ambiguity and ensure that SEO deliverables happen on time and with quality.
SLAs are formal commitments defining expected service levels, responsibilities, and response times across collaborating teams.
As organizations mature in their SEO operations, introducing SLAs becomes essential, especially when coordinating between global, brand, and market-level stakeholders.
For example, suppose a global or brand team is responsible for actions that impact a lower level, such as a local market. In that case, those responsibilities must be documented and bound to SLA metrics.
This not only clarifies accountability but reinforces cross-functional support. Consider a global product launch: If the worldwide team owns the standardized topic taxonomy, it must be delivered to local markets in time for localization and adaptation.
Failure to meet these timeframes puts pressure on markets at launch and risks missed visibility. An SLA helps prevent this by enforcing alignment through timelines and accountability.
Core SLA Components:
Defined Turnaround Times: For topical or taxonomy research, page audits, and performance reporting.
Prioritization Levels: Normal, high-priority, emergency, with response timelines.
Escalation Paths: For unmet KPIs or technical blockers.
Quarterly Review Cadence: For content clusters, PLPs, and editorial integration.
Feedback Loops: Structured inputs from local teams into topic and content models and optimization cycles.
All SLAs should be clearly documented and agreed upon by both internal stakeholders and external agency partners. This alignment ensures that expectations are mutually understood and that accountability is shared.
In addition, a defined escalation process, covering both operational delays and performance disputes, must be in place and visible across all participants in the SEO workflow.
Process governance should be transparent, with clear ownership between corporate, brand, and local roles.
A robust tool utilization strategy ensures consistency, visibility, and collaboration across geographies.
The proper tool structure minimizes duplication, improves time-to-insight, and supports efficient SEO workflows, yet it does not impede any unique requirements at market levels.
Core Elements:
Centralized Tool Procurement: Licensing enterprise-grade platforms at scale and using automation or appropriate seat licenses for brands and markets.
Shared Access & Dashboards: Central teams provision access and enforce naming conventions and tagging protocols.
Integration With Tech Stack: SEO tools integrated into content management system (CMS), digital asset management (DAM), analytics, and campaign platforms.
Local Adaptation Guidelines: Empower markets to use supplementary tools while maintaining reporting standards.
Tools should be centrally funded to ensure consistency, leverage volume-based pricing, and simplify vendor relationships.
When centralized funding is not feasible, a “tin cup” model may be used, with markets contributing based on utilization and need. This hybrid approach helps ensure broad access to necessary tools while aligning budgets to value creation.
A real-world example underscores the importance of strategic tooling governance. In one organization, the enterprise licensed a powerful SEO diagnostic platform, but with a cap on the number of URLs that could be crawled.
Since U.S.-based teams initiated most crawls, smaller markets were often excluded due to exhausted crawl credits.
This led to a lack of visibility into localized issues, missed global diagnostic signals, and an inability to surface SEO problems across the full portfolio.
Organizations must ensure tooling limits don’t inadvertently prioritize one region over another and that diagnostic equity is built into global processes.
Budget & Resource Allocation Models
Budgets must reflect strategic intent, balancing centralized enablement with market agility.
A key benefit of adopting a three-level management structure aligned to global and local goals is the ability to accurately identify actual resource needs.
This structure helps link local execution to global outcomes, providing the data and justification needed to support budget requests.
When budget allocation aligns with tactical needs and enterprise goals, securing executive sponsorship and scaling successful models becomes easier.
Pay-for-Play Services: Market-funded services like local content research, link building, or page audits.
Joint-Funded Pilots: CoE co-invests with business units to explore new opportunities.
Agency Rate Cards: Pre-negotiated pricing and scope packages to streamline engagement.
ROI Justification Models: Frameworks to link SEO investment to lead gen, conversion uplift, and efficiency gains.
Allocating resources based on market opportunity modeling helps prioritize high-impact work and avoid waste.
Managing Local Agencies And Execution Partners
International SEO execution often depends on external support, but market inconsistency can erode gains.
A lack of coordination in one multinational SEO initiative involving multiple agencies led to numerous tickets being submitted for nearly identical issues.
Some tickets addressed the same problem using different approaches, while others attempted to undo recently completed work based on alternate recommendations from a local agency.
This fragmentation caused unnecessary backlog, confusion, and frustration, highlighting the need for strong alignment on how SEO issues and changes are approached.
Key guidelines may be integrated directly into contracts with external partners. One proven approach references the corporate SEO Center of Excellence playbooks, brand-specific standards, and Google’s SEO fundamentals as foundational compliance requirements.
These guidelines should be codified in contractual language, with a clause stating that any unapproved deviations will be corrected at the partner’s expense.
This ensures that new websites, SEO experiments, or localization practices do not introduce non-compliant structures or technical risks without visibility and alignment.
Best Practices:
Approved Vendor Lists: Curated list of pre-vetted agencies aligned with corporate standards.
Onboarding Templates: Playbooks for briefing agencies on brand voice, workflows, and KPIs.
Monthly Performance Reviews: Standard cadence of reporting and performance analysis.
SEO Task Scoping Tools: Templates for briefs, content, and searcher interest research requests, and content updates.
Audit Trail Protocols: Visibility into agency deliverables, implementation logs, and turnaround times.
With effective agency governance, local teams can move fast, without compromising quality or consistency.
Transitioning To A Mature SEO Operating Model
A successful shift to an international SEO structure requires staged planning and executive alignment.
The saying goes that Rome wasn’t built in a day, and neither will your global search program be. However, the framework outlined here provides a structured starting point.
With the accelerating change in AI-driven search, having a uniform and consistent process that is well integrated across marketing, development, content creation, and all teams responsible for visibility and engagement is critical for future success.
Roadmap Elements:
Stakeholder Interviews: Capture local challenges, needs, and barriers to change.
Current-State Coverage Map: Understand what is done, where, by whom.
SEO Maturity Assessment: Evaluate readiness across people, process, tools, and performance.
Pilot Programs: Test governance, SLA models, and tooling structures with one region or BU.
Training & Change Management: Ongoing enablement to embed new practices and workflows.
Phased rollouts ensure learning loops and scalability.
Building An SEO Organization Built For Scale
As search becomes more multimodal and AI-driven, companies can no longer afford disjointed SEO practices.
A strong SEO organizational structure balances strategy and execution, global alignment and local nuance, standardization and innovation.
By embracing a service-level model, aligning KPIs to business outcomes, and establishing clear governance, global enterprises can:
Improve search visibility.
Reduce operational waste.
Enable consistent, scalable content performance.
Ultimately, SEO becomes not just a marketing function but a critical enabler of digital growth and global value creation.
I remember seeing those “God is my co-pilot” bumper stickers since I was old enough to read them.
I was a precocious little agnostic, so they always struck me as weird. God can’t be your co-pilot because God isn’t a physical manifestation of someone who can help you drive a car.
I eventually figured out that “God is my co-pilot” was less a literal statement and more a declaration of faith that there is an omniscient presence available to help you navigate life’s construction zones (if you believe, anyway).
So, fast forward to 2025, and marketers have a new omniscient presence that they can put their faith in. Something that seems equally all-knowing but perhaps a little more … unpredictable.
AI.
Large language models (LLMs) – like ChatGPT, Claude, Gemini – feel delightfully divine when you first try them. They answer instantly, confidently, and often with an authority that makes you wonder if they do know everything.
But, spend enough time with these tools, and you discover something unsettling: AI isn’t just your god-like guide. It can also act like the devil, gleefully granting your wishes exactly as asked – and letting you suffer the consequences.
This is why the healthiest way to think of AI in your SEO and content workflows is as a co-pilot. Not God. Not Lucifer. But, a powerful partner that can elevate your work, if you exercise your free will (and make good choices).
The God-Like Qualities Of AI
There’s a reason AI feels god-like in a marketing context:
It seems omnipresent, embedded in your search results, your content management system (CMS), your analytics.
It delivers answers instantly, with confidence and authority.
It processes far more data than any human ever could, instantly finding patterns we mere mortals miss on the first (or third) pass.
Ask it to draft a content brief, summarize competitive search engine results pages (SERPs), generate topic clusters, or even shape a brand narrative – and it performs in seconds what would have taken you hours.
That kind of power can feel miraculous.
But, just as theologians remind us that God’s will is mysterious and not always aligned with ours, LLMs work on their own unknowable internal logic.
The outputs may not match your intent. The answer may not come in the form you wanted. And you may not even fully grasp why it chose the answer it did.
The Devilish Side Of AI
On the flip side, AI can also be a trickster: seductive, transactional, and literal. It will grant you exactly what you wish for – and sometimes that’s the worst thing possible.
When you prompt an LLM poorly, you’re effectively making a deal with the devil. The model will fulfill your request to the letter, even if what you asked was misguided, incomplete, or poorly articulated.
The result? Content that’s technically correct but off-brand, off-tone, or even factually wrong – yet delivered with such confidence it lulls you into publishing it.
The moral: Be careful what you ask for. The clarity of your prompt determines the quality of your output.
What AI Is Good At
When treated as a co-pilot, not as a god, AI can supercharge your workflow:
Research & Insights
Competitive landscape analyses.
SERP gap identification.
Tracking how competitors frame their unique value propositions.
Summarizing multiple opinion pieces or reviews into one clear insight.
Identifying overlooked audience segments based on forums and social media discussions.
Content Ideation & Briefing
Generating alternative angles on stale topics: e.g., turning “best practices” into “common mistakes” or “myths to avoid.”
Rewriting existing briefs to prioritize experience, expertise, authoritativeness, and trustworthiness (E-E-A-T) signals
Drafting Q&A content by scanning customer service transcripts or Reddit threads.
Suggesting specific examples or metaphors to make dry topics more engaging.
Narrative Shaping & Messaging
Reworking messaging for different formats: a LinkedIn post, an email subject line, and a webinar title – all aligned.
Auditing your current messaging to highlight jargon and suggest plain-language alternatives.
Helping articulate your brand’s point of view in ways that differentiate it from competitors.
Stress-testing your messaging by generating “devil’s advocate” objections you can preemptively address.
Workflow Enhancements
Drafting a competitive heat map: strengths, weaknesses, opportunities, threats – with citations.
Organizing customer testimonials into themed categories and crafting pull quotes.
Generating follow-up email sequences based on webinar transcripts or meeting notes.
Converting white papers into tweet threads, infographic outlines, and video scripts.
It’s like an intern with infinite energy and decent taste – incredibly helpful, but still in need of supervision.
What AI Is Not Good At
Don’t confuse the fluency of AI with wisdom. Here’s where it stumbles:
Judgment & Nuance
It doesn’t understand your brand’s unique sensibility, your audience’s emotional context, or when not to say something. You have to give it that context and direction. You cannot assume it will figure it out.
Accuracy & Truth
It is still prone to “hallucinations” – confidently wrong statements presented as fact.
We have limited understanding of why this happens, but it is so frequent that you almost have to assume there are at least a few hallucinations in the output somewhere.
Accountability
It cannot make decisions, nor does it bear the consequences of your choices. That’s on you.
In short, AI lacks your free will. And free will is what allows you to question, interpret, and choose what to do with its suggestions.
The Co-Pilot Mindset: Free Will Wins
To work effectively with your AI co-pilot, you need to strike the right balance between trust and control.
Here’s how:
Stay In The Pilot’s Seat
Never hand over full control. You’re still ultimately responsible for the vehicle.
Treat AI as a partner – or maybe not even a full partner, more like an exceptionally bright and quick research assistant – but never a replacement for you in any equation.
Be Precise In Your Prompts
Don’t assume it “knows what you mean.” Giving the AI instructions is like giving instructions to a particularly clever child who enjoys maliciously complying with your orders, except the AI doesn’t actually experience the joy.
You need to articulate your expectations clearly: format, tone, audience, and purpose. Add as much context and as many constraints as you can. The more data points and context you can provide, the better the outputs will be.
Use It To Accelerate, Not Replace
AI can speed up research, help shape narratives, and generate ideas, but it can’t replace your expertise or final judgment.
Review & Revise
Never, never, never, never publish output unedited. Always apply your brand’s perspective, always fact-check, and always ensure alignment with your goals.
Read everything you’re about to publish carefully. It’s okay to trust, but always verify.
Here’s an example of how that looks in practice:
I recently took a client’s complete keyword ranking report – not just the terms they were tracking, but every single ranking URL and query – and filtered out any URL already on page 1.
Then, I narrowed the data to just rankings in positions 11-20 (to keep it manageable) and fed that into an LLM.
I asked it to estimate the potential lift in organic traffic if each term improved to position 1 and to rank the list by estimated lift, highest to lowest.
But, I also gave the LLM context about the client’s business, explaining what kinds of customers and services were most valuable to them.
Then, I asked the model to highlight the keywords that made the most business sense for this client, because not every keyword you rank for is one you actually want to rank for.
With that context, the LLM was able to match keyword intent to the client’s goals and call out the terms that aligned with their business priorities.
In just minutes, I had a prioritized roadmap of high-impact, high-fit opportunities – something that would have taken hours to produce manually.
Practical Ways To Work With AI
Here are some more actionable ways you can incorporate AI into your workflow effectively:
Research Smarter And Faster
Create a competitive matrix with links and pros/cons.
Summarize customer sentiment across reviews, highlighting recurring pain points.
Surface conflicting expert opinions to inform balanced thought leadership pieces.
Forecast upcoming trends based on chatter in niche forums and early adopters.
Build Better Briefs
Include competitive positioning suggestions in briefs, not just keywords.
Add tone-of-voice examples aligned to audience segments.
Incorporate real data sources and reference points to help writers anchor their copy.
Generate sample social captions to support a campaign.
Strengthen Your Messaging
Stress-test a headline by generating objections and counterpoints.
Rewrite complex product descriptions into benefit-driven language for different audiences.
Propose alternate positioning statements for product launches or rebrands.
Audit your FAQ section to make it more conversational and AI-friendly.
Repurpose And Expand Content
Turn webinar transcripts into ebooks, blog series, and email drips.
Extract key insights from research reports to create shareable social graphics.
Draft SEO-friendly meta descriptions and titles for old content.
Identify missed opportunities in evergreen content for updates or expansion.
AI can do so much more than just “help you ideate.” It can help you uncover blind spots, repurpose assets, and deepen your strategic thinking, but only when you stay in the driver’s seat to guide and refine the outputs.
Final Thought: You, And Only You, Are The Pilot
I think we tend to treat our collective relationship with AI the same way we look at religion – you’re either a believer or an atheist.
Some have complete faith and trust it without question, while others reject it entirely and are convinced there is nothing there to believe in. The truth is somewhere in the middle (as it often is).
AI can be a powerful, tireless, but imperfect partner. It can help carry and manage heavy mental loads, work with you to map out routes and decide on destinations, but it can not take responsibility for driving the car. That’s got to be on you.
Your free will – your ability to keep your hands on the wheel – is what ensures the journey ends where you intended. If you actually let go, you’re certainly going to crash. You’re asking for assistance, not a magical autopilot.
So, go ahead: Let AI ride shotgun and keep your hands at 10 and two, where they belong.
OpenAI has released two new open-weight language models under the permissive Apache 2.0 license. These models are designed to deliver strong real-world performance while running on consumer hardware, including a model that can run on a high-end laptop with only 16 GB of GPU.
Real-World Performance at Lower Hardware Cost
The two models are:
gpt-oss-120b (117 billion parameters)
gpt-oss-20b (21 billion parameters)
The larger gpt-oss-120b model matches OpenAI’s o4-mini on reasoning benchmarks while requiring only a single 80GB GPU. The smaller gpt-oss-20b model performs similarly to o3-mini and runs efficiently on devices with just 16GB of GPU. This enables developers to run the models on consumer machines, making it easier to deploy without expensive infrastructure.
Advanced Reasoning, Tool Use, and Chain-of-Thought
OpenAI explains that the models outperform other open source models of similar sizes on reasoning tasks and tool use.
According to OpenAI:
“These models are compatible with our Responses API(opens in a new window) and are designed to be used within agentic workflows with exceptional instruction following, tool use like web search or Python code execution, and reasoning capabilities—including the ability to adjust the reasoning effort for tasks that don’t require complex reasoning and/or target very low latency final outputs. They are entirely customizable, provide full chain-of-thought (CoT), and support Structured Outputs(opens in a new window).”
Designed for Developer Flexibility and Integration
OpenAI has released developer guides to support integration with platforms like Hugging Face, GitHub, vLLM, Ollama, and llama.cpp. The models are compatible with OpenAI’s Responses API and support advanced instruction-following and reasoning behaviors. Developers can fine-tune the models and implement safety guardrails for custom applications.
Safety In Open-Weight AI Models
OpenAI approached their open-weight models with the goal of ensuring safety throughout both training and release. Testing confirmed that even under purposely malicious fine-tuning, gpt-oss-120b did not reach a dangerous level of capability in areas of biological, chemical, or cyber risk.
Chain of Thought Unfiltered
OpenAI is intentionally leaving Chain of Thought (CoTs) unfiltered during training to preserve their usefulness for monitoring, based on the concern that optimization could cause models to hide their real reasoning. This, however, could result in hallucinations.
“In our recent research, we found that monitoring a reasoning model’s chain of thought can be helpful for detecting misbehavior. We further found that models could learn to hide their thinking while still misbehaving if their CoTs were directly pressured against having ‘bad thoughts.’
More recently, we joined a position paper with a number of other labs arguing that frontier developers should ‘consider the impact of development decisions on CoT monitorability.’
In accord with these concerns, we decided not to put any direct optimization pressure on the CoT for either of our two open-weight models. We hope that this gives developers the opportunity to implement CoT monitoring systems in their projects and enables the research community to further study CoT monitorability.”
Impact On Hallucinations
The OpenAI documentation states that the decision to not restrict the Chain Of Thought results in higher hallucination scores.
The PDF version of the model card explains why this happens:
Because these chains of thought are not restricted, they can contain hallucinated content, including language that does not reflect OpenAI’s standard safety policies. Developers should not directly show chains of thought to users of their applications, without further filtering, moderation, or summarization of this type of content.”
Benchmarking showed that the two open-source models performed less well on hallucination benchmarks in comparison to OpenAI o4-mini. The model card PDF documentation explained that this was to be expected because the new models are smaller and implies that the models will hallucinate less in agentic settings or when looking up information on the web (like RAG) or extracting it from a database.
OpenAI OSS Hallucination Benchmarking Scores
Takeaways
Open-Weight Release OpenAI released two open-weight models under the permissive Apache 2.0 license.
Performance VS. Hardware Cost Models deliver strong reasoning performance while running on real-world affordable hardware, making them widely accessible.
Model Specs And Capabilities gpt-oss-120b matches o4-mini on reasoning and runs on 80GB GPU; gpt-oss-20b performs similarly to o3-mini on reasoning benchmarks and runs efficiently on 16GB GPU.
Agentic Workflow Both models support structured outputs, tool use (like Python and web search), and can scale their reasoning effort based on task complexity.
Customization and Integration The models are built to fit into agentic workflows and can be fully tailored to specific use cases. Their support for structured outputs makes them adaptable to complex software systems.
Tool Use and Function Calling The models can perform function calls and tool use with few-shot prompting, making them effective for automation tasks that require reasoning and adaptability.
Collaboration with Real-World Users OpenAI collaborated with partners such as AI Sweden, Orange, and Snowflake to explore practical uses of the models, including secure on-site deployment and custom fine-tuning on specialized datasets.
Inference Optimization The models use Mixture-of-Experts (MoE) to reduce compute load and grouped multi-query attention for inference and memory efficiency, making them easier to run at lower cost.
Safety OpenAI’s open source models maintain safety even under malicious fine-tuning; Chain of Thoughts (CoTs) are left unfiltered for transparency and monitorability.
CoT transparency Tradeoff No optimization pressure applied to CoTs to prevent masking harmful reasoning; may result in hallucinations.
Hallucinations Benchmarks and Real-World Performance The models underperform o4-mini on hallucination benchmarks, which OpenAI attributes to their smaller size. However, in real-world applications where the models can look up information from the web or query external datasets, hallucinations are expected to be less frequent.
Featured Image by Shutterstock/Good dreams – Studio
Over the past decade, digital marketers have witnessed a dramatic shift in how search budgets are allocated.
In the past decade, companies were funding SEO teams alongside PPC teams. However, a shift towards PPC-first has dominated the inbound marketing space.
Where Have SEO Budgets Gone?
Today, more than $150 billion is spent annually on paid search in the United States alone, while only $50 billion is invested in SEO.
With Google Ads, every dollar has a direct, reportable outcome:
Impressions.
Clicks.
Conversions.
SEO, by contrast, has long been:
A black box.
As a result, agencies and the clients that hire them followed the money, even when SEO’s results were higher.
PPC’s Direct Attribution Makes PPC Look More Important, But SEO Still Dominates
Hard facts:
SEO drives 5x more traffic than PPC.
Companies pay 3x more on PPC than SEO.
Image created by MarketBrew, August 2025
You Can Now Trace ROI Back To SEO
As a result, many SEO professionals and agencies want a way back to organic. Now, there is one, and it’s powered by attribution.
Attribution Is the Key to Measurable SEO Performance
Instead of sitting on the edge of the search engine’s black box, guessing what might happen, we can now go inside the SEO black box, to simulate how the algorithms behave, factor by factor, and observe exactly how rankings react to each change.
With this model in place, you are no longer stuck saying “trust us.”
You can say, “Here’s what we changed. Here’s how rankings moved. Here’s the value of that movement.” Whether the change was a new internal link structure or a content improvement, it’s now visible, measurable, and attributable.
For the first time, SEO teams have a way to communicate performance in terms executives understand: cause, effect, and value.
This transparency is changing the way agencies operate. It turns SEO into a predictable system, not a gamble. And it arms client-facing teams with the evidence they need to justify the budget, or win it back.
How Agencies Are Replacing PPC With Measurable Organic SEO
For agencies, attribution opens the door to something much bigger than better reporting; it enables a completely new kind of offering: performance-based SEO.
Traditionally, SEO services have been sold as retainers or hourly engagements. Clients pay for effort, not outcomes. With attribution, agencies can now flip that model and say: You only pay when results happen.
Enter Market Brew’s AdShifted feature to model this value and success as shown here:
Screenshot from a video by MarketBrew, August 2025
The AdShift tool starts by entering a keyword to discover up to 4* competitive URLs for the Keyword’s Top Clustered Similarities. (*including your own website plus 4 top-ranking competitors)
Screenshot of PPC vs. MarketBrew comparison dashboard by Marketbrew, August 2025
AdShift averages CPC and search volume across all keywords and URLs, giving you a reliable market-wide estimate and details for your brand towards a monthly PPC investment to rank #1.
Screenshot of a dashboard by Marketbrew, August 2025
AdShift then calculates YOUR percentage of replacement for PPC to fund SEO.
This allows you to model your own Performance Plan with variable discounts available to the Market Brew license fees with an always less than 50% of PPC Fee for clicks replaced by new SEO traffic.
Screenshot of a dashboard by Marketbrew, August 2025
AdShift simulates a PPC replacement plan option selected based on its keywords footprint to instantly see savings from the associated Performance Plans.
That’s the heart of the PPC replacement plan: a strategy you can use to gradually shift a clients’ paid search budgets into measurable performance-based SEO.
What Is A PPC Replacement Plan? Trackable SEO.
A PPC replacement plan is a strategy in which agencies gradually shift their clients’ paid search budgets into organic investments, with measurable outcomes and shared performance incentives.
Here’s how it works:
Benchmark Paid Spend: Identify the current Google Ads budget, i.e., $10,000 per month or $120,000 per year.
Forecast Organic Value: Use search engine modeling to predict the lift in organic traffic from specific SEO tasks.
Execute & Attribute: Complete tasks and monitor real-time changes in rankings and traffic.
Charge on Impact: Instead of billing for time, bill for results, often at a fraction of the client’s former ad spend.
This is not about replacing all paid spend.
Branded queries and some high-value targets may remain in PPC. But for the large, expensive middle of the keyword funnel, agencies can now offer a smarter path: predictable, attributable organic results, at a lower cost-per-click, with better margins.
And most importantly, instead of lining Google’s pockets with PPC revenue, your investments begin to fuel both organic and LLM searches!
Real-World Proof That SEO Attribution Works
Agencies exploring this new attribution-powered model aren’t just intrigued … they’re energized. For many, it’s the first time in years that SEO feels like a strategic growth engine, not just a checklist of deliverables.
“We’ve pitched performance SEO to three clients this month alone,” said one digital strategy lead. “The ability to tie ranking improvements to specific tasks changed the entire conversation.”
“Instead of walking into meetings looking to justify an SEO retainer, we enter with a blueprint representing a SEO/GEO/AEO Search Engine’s ‘digital twin’ with the AI-driven tasks that show exactly what needs to be changed and the rankings it produces. Clients don’t question the value … they ask what’s next.”
Several agencies report that new business wins are increasing simply because they offer something different. While competitors stick to vague SEO promises or expensive PPC management, partners leveraging attribution offer clarity, accountability, and control.
And when the client sees that they’re paying less and getting more, it’s not a hard sell, it’s a long-term relationship.
A Smarter, More Profitable Model for Agencies and SEOs
The traditional agency model in search has become a maze of expectations.
Managing paid search may deliver short-term wins, but it comes to a bidding war with only those with the biggest budgets winning. SEO, meanwhile, has often felt like a thankless task … necessary but underappreciated, valuable but difficult to prove.
Attribution changes that.
For agencies, this is a path back to profitability and positioning. With attribution, you’re not just selling effort … you’re selling outcomes. And because the work is modeled and measured in advance, you can confidently offer performance plans that are both client-friendly and agency-profitable.
For SEOs, this is about getting the credit they deserve. Attribution allows practitioners to demonstrate their impact in concrete terms. Rankings don’t just move, … they move because of you. Traffic increases aren’t vague, … they’re connected to your specific strategies.
Now, you can show this.
Most importantly, this approach rebuilds trust.
Clients no longer have to guess what’s working. They see it. In dashboards, in forecasts, in side-by-side comparisons of where they were and where they are now. It restores SEO to a place of clarity and control where value is obvious, and investment is earned.
The industry has been waiting for this. And now, it’s here.
From PPC Dependence to Organic Dominance — Now Backed by Data
Search budgets have long been upside down, pouring billions into paid clicks that capture a mere fraction of user attention, while underfunding the organic channel that delivers lasting value.
Why? Because SEO lacked attribution.
That’s no longer the case.
Today, agencies and SEO professionals have the tools to prove what works, forecast what’s next, and get paid for the real value they deliver. It’s a shift that empowers agencies to move beyond bidding-war PPC management and into a lower cost & higher ROAS, performance-based SEO.
This isn’t just a new service mode it’s a rebalancing of power in search.
Organic is back. It’s measurable. It’s profitable. And it’s ready to take center stage again.
The only question is: will you be the agency or brand that leads the shift or watch as others do it first?
OpenAI has given itself a dual mandate. On the one hand, it’s a tech giant rooted in products, including of course ChatGPT, which people around the world reportedly send 2.5 billion requests to each day. But its original mission is to serve as a research lab that will not only create “artificial general intelligence” but ensure that it benefits all of humanity.
My colleague Will Douglas Heaven recently sat down for an exclusive conversation with the two figures at OpenAI most responsible for pursuing the latter ambitions: chief research officer Mark Chen and chief scientist Jakub Pachocki. If you haven’t already, you must read his piece.
It provides a rare glimpse into how the company thinks beyond marginal improvements to chatbots and contemplates the biggest unknowns in AI: whether it could someday reason like a human, whether it should, and how tech companies conceptualize the societal implications.
The whole story is worth reading for all it reveals—about how OpenAI thinks about the safety of its products, what AGI actually means, and more—but here’s one thing that stood out to me.
As Will points out, there were two recent wins for OpenAI in its efforts to build AI that outcompetes humans. Its models took second place at a top-level coding competition and—alongside those from Google DeepMind—achieved gold-medal-level results in the 2025 International Math Olympiad.
People who believe that AI doesn’t pose genuine competition to human-level intelligence might actually take some comfort in that. AI is good at the mathematical and analytical, which are on full display in olympiads and coding competitions. That doesn’t mean it’s any good at grappling with the messiness of human emotions, making hard decisions, or creating art that resonates with anyone.
But that distinction—between machine-like reasoning and the ability to think creatively—is not one OpenAI’s heads of research are inclined to make.
“We’re talking about programming and math here,” said Pachocki. “But it’s really about creativity, coming up with novel ideas, connecting ideas from different places.”
That’s why, the researchers say, these testing grounds for AI will produce models that have an increasing ability to reason like a person, one of the most important goals OpenAI is working toward. Reasoning models break problems down into more discrete steps, but even the best have limited ability to chain pieces of information together and approach problems logically.
OpenAI is throwing a massive amount of money and talent at that problem not because its researchers think it will result in higher scores at math contests, but because they believe it will allow their AI models to come closer to human intelligence.
As Will recalls in the piece, he said he thought maybe it’s fine for AI to excel at math and coding, but the idea of having an AI acquire people skills and replace politicians is perhaps not. Chen pulled a face and looked up at the ceiling: “Why not?”
This is today’s edition of The Download, our weekday newsletter that provides a daily dose of what’s going on in the world of technology.
These protocols will help AI agents navigate our messy lives
A growing number of companies are launching AI agents that can do things on your behalf—actions like sending an email, making a document, or editing a database. Initial reviews for these agents have been mixed at best, though, because they struggle to interact with all the different components of our digital lives.
Anthropic and Google are among the companies and groups working to fix that. Over the past year, they have both introduced protocols that try to define how AI agents should interact with each other and the world around them. If they work as planned, they could give us a crucial part of the infrastructure we need for agents to be useful. Read our story to learn more.
—Peter Hall
A glimpse into OpenAI’s largest ambitions
—James O’Donnell
OpenAI has given itself a dual mandate: on the one hand, it’s a tech giant rooted in products, including of course ChatGPT, which people around the world reportedly send 2.5 billion messages to each day. But its original mission is as a research lab that will not only create “artificial general intelligence” but ensure that it benefits all of humanity.
My colleague Will Douglas Heaven recently sat down for an exclusive conversation with the two figures at OpenAI most responsible for the latter ambitions. The whole story is worth reading for all it reveals—about how OpenAI thinks about the safety of its products, what AGI actually means, and more—but here’s one thing that stood out to me.
This story is from The Algorithm, our weekly newsletter all about the latest goings-on in AI. Sign up to receive it in your inbox every Monday.
The must-reads
I’ve combed the internet to find you today’s most fun/important/scary/fascinating stories about technology.
1 OpenAI is adding mental health guardrails to ChatGPT It’s set to give less direct advice, and encourage users to take breaks from lengthy chats. (NBC) + What happens when doctors fail to spot AI’s mistakes? (The Verge) + OpenAI has released its first research into how using ChatGPT affects people’s emotional well-being. (MIT Technology Review)
2 The US wants to build a nuclear reactor on the moon And it hopes to do that before Russia and China, who are planning to do exactly the same. (Politico) + NASA’s latest mission to the moon just failed. (Engadget) + Nokia is putting the first cellular network on the moon. (MIT Technology Review)
3 How to live forever (or at least get rich trying) Love them or hate them, the people behind the explosion in longevity research are a fascinating bunch. (New Yorker $) + Longevity clinics around the world are selling unproven treatments. (MIT Technology Review)
4 Welcome to Silicon Valley’s ‘hard tech’ era Goodbye, consumer software. Hello, massive military contracts. (NYT $) + Phase two of military AI has arrived. (MIT Technology Review)
5 There’s a big problem with the Gulf’s trillion-dollar AI dream Building data centers in a region that already has water scarcity issues seems…unwise. (Rest of Water) + There’s a data center boom in the US desert too. (MIT Technology Review) + Google has promised to scale back its energy usage during certain times to reduce stress on the grid. (Quartz $)
6 Tesla’s board awarded about $30 billion of shares to Elon Musk “Retaining Elon is more important than ever before,” they wrote in a letter to shareholders yesterday. (FT $) + Tech CEOs pay packets are reaching stratospheric new records. (WSJ $)
7 What happens if you respond to those scam job texts? You get exploited, obviously—but you’d be surprised just how weird it can get along the way. (Slate)
8 Why there’s so much uproar over Vogue’s AI-generated ad It’s the latest flashpoint in the war over when AI should (and shouldn’t) be used. (TechCrunch)
9 Earth’s core seems to be up and leaking out of Earth’s surface It’s a finding that’s forcing geoscientists to rethink some long-held assumptions. (Quanta $) + How a volcanic eruption turned a human brain into glass. (MIT Technology Review)
10 Could lasers help us see inside people’s heads? It seems possible, but big hurdles remain to this new method being adopted in clinical settings. (IEEE Spectrum)
Quote of the day
“Hate it! Don’t want anything to do with it.”
—Weezy Simes, a 27-year-old florist, sums up her feelings about AI to Business Insider.
One more thing
ANDREA D’AQUINO
What happened to the microfinance organization Kiva?
Since it was founded in 2005, the San Francisco-based nonprofit Kiva has helped everyday people make microloans to borrowers around the world. It connects lenders in richer communities to fund all sorts of entrepreneurs, from bakers in Mexico to farmers in Albania. Its overarching aim is helping poor people help themselves.
But back in August 2021, Kiva lenders started to notice that information that felt essential in deciding who to lend to was suddenly harder to find. Now, lenders are worried that the organization now seems more focused on how to make money than how to create change. Read the full story.
—Mara Kardas-Nelson
We can still have nice things
A place for comfort, fun and distraction to brighten up your day. (Got any ideas? Drop me a line or skeet ’em at me.)
+ I want this guy to draw my portrait. + Highly recommend making these lemongrass chicken lettuce wraps. So tasty and easy! + This encyclopedia teaches you about ancient gods and forgotten deities from around the world. + Some of the architecture in Iran looks breathtakingly beautiful.
OpenAI has finally released its first open-weight large language models since 2019’s GPT-2. These new “gpt-oss” models are available in two different sizes and score similarly to the company’s o3-mini and o4-mini models on several benchmarks. Unlike the models available through OpenAI’s web interface, these new open models can be freely downloaded, run, and even modified on laptops and other local devices.
In the company’s many years without an open LLM release, some users have taken to referring to it with the pejorative “ClosedAI.” That sense of frustration had escalated in the past few months as these long-awaited models were delayed twice—first in June and then in July. With their release, however, OpenAI is reestablishing itself as a presence for users of open models.
That’s particularly notable at a time when Meta, which had previously dominated the American open-model landscape with its Llama models, may be reorienting toward closed releases—and when Chinese open models, such as DeepSeek’s offerings, Kimi K2, and Alibaba’s Qwen series, are becoming more popular than their American competitors.
“The vast majority of our [enterprise and startup] customers are already using a lot of open models,” said Casey Dvorak, a research program manager at OpenAI, in a media briefing about the model release. “Because there is no [competitive] open model from OpenAI, we wanted to plug that gap and actually allow them to use our technology across the board.”
The new models come in two different sizes, the smaller of which can theoretically run on 16 GB of RAM—the minimum amount that Apple currently offers on its computers. The larger model requires a high-end laptop or specialized hardware.
Open models have a few key use cases. Some organizations may want to customize models for their own purposes or save money by running models on their own equipment, though that equipment comes at a substantial upfront cost. Others—such hospitals, law firms, and governments—might need models that they can run locally for data security reasons.
OpenAI has facilitated such activity by releasing its open models under a permissive Apache 2.0 license, which allows the models to be used for commercial purposes. Nathan Lambert, post-training lead at the Allen Institute for AI, says that this choice is commendable: Such licenses are typical for Chinese open-model releases, but Meta released its Llama models under a bespoke, more restrictive license. “It’s a very good thing for the open community,” he says.
Researchers who study how LLMs work also need open models, so that they can examine and manipulate those models in detail. “In part, this is about reasserting OpenAI’s dominance in the research ecosystem,” says Peter Henderson, an assistant professor at Princeton University who has worked extensively with open models. If researchers do adopt gpt-oss as new workhorses, OpenAI could see some concrete benefits, Henderson says—it might adopt innovations discovered by other researchers into its own model ecosystem.
More broadly, Lambert says, releasing an open model now could help OpenAI reestablish its status in an increasingly crowded AI environment. “It kind of goes back to years ago, where they were seen as the AI company,” he says. Users who want to use open models will now have the option to meet all their needs with OpenAI products, rather than turning to Meta’s Llama or Alibaba’s Qwen when they need to run something locally.
The rise of Chinese open models like Qwen over the past year may have been a particularly salient factor in OpenAI’s calculus. An employee from OpenAI emphasized at the media briefing that the company doesn’t see these open models as a response to actions taken by any other AI company, but OpenAI is clearly attuned to the geopolitical implications of China’s open-model dominance. “Broad access to these capable open-weights models created in the US helps expand democratic AI rails,” the company wrote in a blog post announcing the models’ release.
Since DeepSeek exploded onto the AI scene at the start of 2025, observers have noted that Chinese models often refuse to speak about topics that the Chinese Communist Party has deemed verboten, such as Tiananmen Square. Such observations—as well as longer-term risks, like the possibility that agentic models could purposefully write vulnerable code—have made some AI experts concerned about the growing adoption of Chinese models. “Open models are a form of soft power,” Henderson says.
Lambert released a report on Monday documenting how Chinese models are overtaking American offerings like Llama and advocating for a renewed commitment to domestic open models. Several prominent AI researchers and entrepreneurs, such as HuggingFace CEO Clement Delangue, Stanford’s Percy Liang, and former OpenAI researcher Miles Brundage, have signed on.
The Trump administration, too, has emphasized development of open models in its AI Action Plan. With both this model release and previous statements, OpenAI is aligning itself with that stance. “In their filings about the action plan, [OpenAI] pretty clearly indicated that they see US–China as a key issue and want to position themselves as very important to the US system,” says Rishi Bommasani, a senior research scholar at the Stanford Institute for Human-Centered Artificial Intelligence.
And OpenAI may see concrete political advantages from aligning with the administration’s AI priorities, Lambert says. As the company continues to build out its extensive computational infrastructure, it will need political support and approvals, and sympathetic leadership could go a long way.
Generative AI platforms such as ChatGPT, Perplexity, and Claude now execute live web searches with all prompts. Ensuring a site is crawlable by AI bots is therefore essential for mentions and citations on those platforms.
Here’s how to optimize a website for AI crawlers.
Disable JavaScript
Make sure your pages are readable with JavaScript disabled.
Unlike Google’s crawler, AI bots are immature. Many tests from industry practitioners confirm AI crawlers cannot always render JavaScript.
Most publishers and businesses no longer worry about JavaScript crawlability since Google has rendered those pages for years. Hence there’s a huge number of JavaScript-heavy sites.
The Chrome browser can render a site without JavaScript. To activate:
Go to your site using Chrome.
Open Web Developer tools at View > Developer > Developer Tools.
Click Settings (behind the gear icon) on the right side of the panel.
Scroll down and check the option “Disable JavaScript” under “Debugger.”
Disable JavaScript in Chrome’s Developer Tools panel.
Now browse your site, making sure:
All essential content is visible, especially behind tabs and drop-down menus.
The navigation menu and other links are clickable.
For video embeds, there’s an option to click to the original video, access a transcript, or both.
You can use Aiso, an AI optimization platform, to ensure AI bots can access and crawl your site. With a free trial, the platform allows a few free checks. Go to the “Website crawlability” section and enter your URL.
The tool will conduct a thorough review with suggestions on improving access for AI crawlers and even show the appearance of pages with JavaScript disabled.
Aiso can review a site’s use of JavaScript and suggest improvements for AI bot access.
Ensure AI Access
Make sure your site allows access for AI bots. Some content management platforms and plugins disallow AI access by default — site owners are often unaware.
To check, review your robots.txt file at [yoursite.com]/robots.txt.
The AI platforms themselves can interpret the file to ensure it allows access. Paste your robots.text URL into a ChatGPT prompt, for example, and request an analysis.
Schema markup makes it easier for AI bots to extract essential information from a page (or bypass a block) without crawling it in full.
For example, many website FAQ sections have collapsible elements that prevent access to AI bots. Schema’s FAQPage Type replicates all questions and answers, enabling bot visibility.
Similarly, Schema’s Article Type can communicate context and authorship of content.
