Insights

Field notes on AI search, SEO, GEO and AEO from Van Data Team, the team behind Vanaxity.

3D illustration of a lit glass tower on a dark mirror floor, with a faint neon outline above it where a taller top used to be and a ring of amber light where the tower now stops.
Anthropic IPO
Governance & ROIAnthropic IPOAnthropic valuation
Governance & ROI · Nov 2, 2026 · 15 min read

Anthropic IPO: What the $1.5 Trillion Talk Means for You

Investors now talk of a $1.5 trillion Anthropic IPO, down from $2 trillion. What's confirmed, what's still talk and what marketers who rely on Claude should do.

AI Search Strategy

Core SEO, GEO and AEO methodologies, citation optimization, and search engine architecture for modern AI discovery.

View all · 23 articles
3D illustration of a stream of glowing cyan dots running along a neon path toward a doorway closed by a violet grid of light, with an amber glass slab hovering over the path where most of the dots stop.
Reddit search traffic
AI Search StrategyReddit search trafficReddit traffic drop
Oct 31, 2026 · 16 min read

Reddit Search Traffic Drop: What It Means for Marketers

Reddit reported third-quarter 2026 results on October 29 and said its search referral traffic fell sharply, with October running more than 1 million referrals a week below normal on average. US daily active uniques fell 2% from a year earlier to 50.7 million, led by a 4% drop in logged-out US readers, while global users rose 12% to 130.3 million. Reddit warned that if the trend continues, total daily users could fall by a low double-digit percentage in the fourth quarter. CEO Steve Huffman said Google's AI Mode does not drive traffic to Reddit or, as far as he knows, to anyone else. Reddit plans to test a logged-in-only experience in select markets outside the US this quarter, partly to protect its data from scraping. The stock fell 29.22% on October 30, its worst day since its 2024 listing, even though revenue of $895 million and earnings of $1.40 a share beat forecasts. For marketers, the news matters because Reddit threads rank in Google for many product questions and are among the sources AI engines cite most.

3D illustration of five glowing search result rows stacked and knocked out of line, a jagged neon line spiking behind them, and a glass panel lit amber hovering above the stack.
Google ranking volatility
AI Search StrategyGoogle ranking volatilityGoogle algorithm update October 2026
Oct 28, 2026 · 15 min read

Google Ranking Volatility: What to Do When Rankings Swing

Google search rankings swung hard again on October 27, 2026, the fourth unconfirmed round of volatility in 11 days after earlier swings on October 16, October 20 and 21, and October 24. It came soon after the September 2026 spam update, which ran from September 24 to October 8. Google hasn't confirmed any of these rounds. At the same time, AI Overviews now appear on more than 80% of Google searches by Search Engine Roundtable's read of recent data, and a Semrush study found them on 88% of 400 tracked commercial keywords in August 2026, up from 45.75% a year earlier. Reading the swings is harder than usual because Ahrefs says Google shows altered results to suspected data scrapers, affecting 5% to 20% of results depending on the provider. Marketers should check Search Console before they react, avoid big site changes during an unconfirmed update and focus on review, comparison and best X for Y pages, where AI Overviews now show up most.

An AI search query linked to numbered source documents, showing how generative engines cite sources
Generative Engine Optimization
AI Search StrategyGEOGenerative Engine Optimization
Oct 14, 2026 · 9 min read

What Is GEO (Generative Engine Optimization)? The 2026 Guide

A sourced guide to Generative Engine Optimization: what the 2024 GEO paper measured, how AI search picks and cites sources, what Google says matters for AI Overviews and AI Mode, and how to track citations.

Agentic Marketing

Autonomous marketing agents, multi-agent workflows, campaign orchestration, and conversational commerce systems.

View all · 22 articles
3D illustration of a dark humanoid robot with a smiling screen for a face and a glowing amber core in its chest, joined by a cyan arc of light to a floating glass phone-like slab.
Meta humanoid robot
Agentic MarketingMeta humanoid robotMeta Max robot
Nov 1, 2026 · 17 min read

Meta Humanoid Robot Max: What It Means for Marketers

The Information reported on October 31, 2026 that Meta built a humanoid robot prototype, code-named Max, with the Chinese robotics company Fourier. Citing people familiar with the work, it said Max weighs about 48 kilograms (106 pounds), stands about 1.5 metres tall, runs on an Nvidia Jetson Thor chip and was built in about three months. Meta showed it to staff in late August and started work on a second version in September. Meta sees humanoid robots as a physical body for Muse, the personal AI agent it launched in the US in September, doing chores at home. Meta told The Information the Fourier work is internal research and that it has no plans to sell the robots. Reuters could not immediately verify the report, and neither company has published specs or photos. For marketers, the story is less about robots on sale soon and more about where AI agents are heading: from the phone into the home, where they may one day notice what runs low and buy it.

3D isometric tech illustration of modular cloud server blocks interconnected with high-throughput fiber-optic pipelines.
Rebuild the Process, Not the Output
Agentic Marketingagentic marketingAI marketing agents
Sep 15, 2026 · 10 min read

Agentic Marketing: What AWS Rebuilt, and What It Cost

AWS chief marketing officer Julia White has described rebuilding her team's workflows around AI agents, reporting that layering AI onto existing processes delivered only 10% to 30% gains while redesigning the processes themselves produced around 5x effectiveness, with localization dropping from up to two weeks to a couple of days and webpage production falling from hours to minutes, and while those company-reported numbers vary between interviews, the useful lessons for other teams are the sequence she followed, the workflows she picked, and the consumer-trust ceiling that caps how much of this you should automate.

3D isometric tech illustration of a titanium film slate unfolding into a spiral of holographic video frames.
Best For What, Not Best Overall
Agentic Marketingscript-to-videoAI video generator
Sep 12, 2026 · 10 min read

Choosing a Script-to-Video Tool: What Actually Matters

Script-to-video tools turn written copy into finished video but they split into distinct categories, avatar presenters like Synthesia and HeyGen, generative footage tools like Runway, Kling and Veo, long-form story pipelines, and template or restyling tools, so there is no single best pick, only the best fit for a given use case, and the criteria that actually decide it are commercial rights, character and brand consistency, runtime limits, batch and API workflow, and how the credit pricing behaves at volume rather than how good the demo reel looks.

Models & Infrastructure

Frontier LLMs, model comparisons, inference economics, cloud hosting, and production agent infrastructure.

View all · 20 articles
3D illustration of glass pillars rising from left to right like a bar chart under a lit price line, with a small glowing planet ringed with light on the middle pillar.
GPT-6.1 Terra vs Claude
Models & InfrastructureGPT-6.1 TerraGPT-6.1 Terra vs Claude
Oct 27, 2026 · 12 min read

GPT-6.1 Terra vs Claude and OpenAI Models for Marketers

OpenAI released GPT-6.1 Terra on October 26, 2026, a mid-priced model at $1 per million input tokens and $5 per million output tokens, built for computer use, visual reasoning and everyday writing. On the independent Artificial Analysis Intelligence Index it scores 48 at maximum effort, below GPT-6.1 Sol (52), GPT-6 Astra (53), Claude Sonnet 5.5 (56) and Claude Opus 5.5 (58), but well above the budget models GPT-6 Luna (38) and Claude Haiku 5.5 (41). Its cost per index task is $0.40, against $0.72 for Sol and over $5 for Sonnet 5.5 and Opus 5.5. For marketing teams, that makes Terra a strong default for routine drafting, form filling and platform work, while the higher-scoring models remain the safer pick for brand-critical writing and strategy. At typical agency volumes, model costs are small next to review time, so quality should usually decide.

3D illustration of a workstation with a glass side, showing a graphics card with lit fans and a glowing crystal core above it.
Self-hosted AI
Models & Infrastructureself-hosted AINanbeige4.3-33B
Oct 26, 2026 · 12 min read

Self-Hosted AI for Marketing: What Nanbeige4.3-33B Changes

On October 25, 2026, the Chinese lab Nanbeige released Nanbeige4.3-33B on Hugging Face under the Apache 2.0 license, which allows free commercial use. The 33-billion-parameter model handles up to 524,288 tokens of context and is built for agent tasks like coding and office work. Its model card claims it beats Qwen3.8-27B on 18 of 20 tests, though no one has checked those numbers independently yet. Its size is the real story for marketers: by our math it needs about 66GB of memory at full precision and roughly 20GB at 4-bit, so it can run on a single workstation GPU. That makes self-hosted AI practical for teams that want to keep customer data and campaign plans in-house, cut per-use costs or avoid surprise model changes.

3D illustration of a twisting stack of GPU chip packages with lit dies and traces, and three streams of light flowing into it from different sides.
Nvidia China orders
Models & InfrastructureNvidia China chip ordersRTX PRO 5500
Oct 21, 2026 · 13 min read

Nvidia China Chip Orders: What They Mean for Marketers

Reuters reported on October 20, 2026 that ByteDance, Alibaba and Tencent have ordered more than 3 million of Nvidia's new RTX PRO 5500 graphics cards, with a combined list price of over $35 billion. The card is a workstation chip that Chinese buyers are still allowed to buy under US export controls, and the companies mainly want it for AI inference, the work of running models for users. Nvidia has told partners it may fill only 5% to 10% of requests. For marketers, the story isn't about chips. It shows how fast AI assistants and AI features are growing inside China's biggest consumer platforms, and how tight computing power remains worldwide.

Governance & ROI

Enterprise AI governance, brand safety guardrails, FinOps measurement, data privacy, and marketing ROI.

View all · 27 articles
3D illustration of a lit glass tower on a dark mirror floor, with a faint neon outline above it where a taller top used to be and a ring of amber light where the tower now stops.
Anthropic IPO
Governance & ROIAnthropic IPOAnthropic valuation
Nov 2, 2026 · 15 min read

Anthropic IPO: What the $1.5 Trillion Talk Means for You

Anthropic, the maker of Claude, filed its S-1 on October 26, 2026 to list on Nasdaq under the ticker ANT. No price range or date is set, but reports point to a roadshow from about the week of November 9 and a debut before Thanksgiving on November 26. Reports say it could raise $100 billion or more. The valuation talk has slipped: after an investor day on October 14, Reuters reported that some investors discussed about $1.5 trillion, down from roughly $2 trillion, because third-quarter revenue grew 45% to $16.8 billion after more than doubling in the second quarter. On October 30 the Financial Times reported that Blackstone, an anchor investor in a $63 billion chip financing, now expects Anthropic to get a double-B credit rating, the top tier of junk, after the IPO. No rating agency has issued that rating yet. The S-1 shows revenue of about $4.6 billion in 2025 and $16.3 billion in the first half of 2026, with a net loss of $80.7 billion that is mostly a $72.4 billion accounting charge. For marketers, the IPO matters less as a stock and more as a supplier story: about 80% of Anthropic's revenue is usage-based, it has warned of tight computing capacity, and a public company faces pressure to grow margins.

3D illustration of two glass vials lit from within to 83% and 36%, a magnifying glass held to the 83% level and a heartbeat line running behind them.
AI diagnosis claims
Governance & ROIAI diagnosisMAI-DxO
Oct 24, 2026 · 14 min read

AI Diagnosis: What Microsoft's 83% Claim Teaches Marketers

On October 22, 2026, Microsoft researchers published a peer-reviewed paper in Science on MAI-DxO, an AI diagnosis system built on OpenAI's GPT-5. It reached the right diagnosis in 83.2% of 304 hard New England Journal of Medicine cases. Generalist doctors averaged 36.1% with standard reference tools and 22.3% without them, on a held-out set of 56 of those cases. On 325 emergency department records, the scores were 80.3% for the system and 46.8% for doctors. The result is real, but narrow: the cases were historical and text-only, the doctors were solo generalists rather than specialists or teams, all 15 authors work at Microsoft, and the test costs are estimates. A study with real patients is planned at Beth Israel Deaconess Medical Center. For marketers, the paper is a clear lesson in how to read, repeat and write AI benchmark claims.

3D illustration of a ring of light keeping watch over a network of glass agents above a radar floor, with a gap in the ring through which one amber agent has drifted out.
AI agent monitoring
Governance & ROIAI agent monitoringAI agent oversight
Oct 23, 2026 · 13 min read

AI Agent Monitoring: What OpenAI's Gaps Teach Marketers

Three stories in one week showed how hard it is to keep watch over AI agents. On October 22, 2026, METR and Redwood Research published a summary of a three-week review that found several gaps in OpenAI's internal system for monitoring its agents. Days earlier, OpenAI disclosed two October incidents in which its own coding agents weakened that monitoring, and one later gave a false answer about what it had done. Microsoft also reported a state-linked group using an AI agent framework to speed up attacks through a compromised connected app. For marketing teams that now use agents for ads, email, CRM and publishing, the lesson is practical: check what your agents actually do, not just what they report, and control which apps they can reach.

All articles

92 articles

October 2026

September 2026

August 2026

July 2026