Insights

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

3D illustration of a neon path of cyan light motes running across a dark mirror floor and through an arch of cyan light, while a dim amber branch leads three amber motes wearing dark visors to a glowing amber barrier where they stop.
Fake AI crawlers
AI Search Strategyfake AI crawlersverify AI crawlers
AI Search Strategy · Nov 16, 2026 · 12 min read

Fake AI Crawlers: How We Verified a Week of Bot Traffic

We matched 23,962 requests from 15 crawler names to the IP lists OpenAI, Google, Anthropic and others publish. 11.1% came from elsewhere, and the fakes went for .env files.

AI Search Strategy

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

View all · 27 articles
3D illustration of a neon path of cyan light motes running across a dark mirror floor and through an arch of cyan light, while a dim amber branch leads three amber motes wearing dark visors to a glowing amber barrier where they stop.
Fake AI crawlers
AI Search Strategyfake AI crawlersverify AI crawlers
Nov 16, 2026 · 12 min read

Fake AI Crawlers: How We Verified a Week of Bot Traffic

We took one week of request logs from vandatateam.com, the site of Van Data Team, the company behind Vanaxity, and checked every request that named a crawler against the address list its operator publishes. From November 9 to 15, 2026, 23,962 requests named one of 15 crawlers with a published list, from Googlebot and Bingbot to OpenAI's, Anthropic's and Perplexity's bots. 21,303 of them (88.9%) came from the operator's addresses and 2,659 (11.1%) did not. Seven addresses sent 627 of the outsiders while switching between as many as 15 crawler names, and 321 requests went to paths only a scanner asks for, such as .env files and .git folders. Verified crawlers asked for none of those files. They read mostly our AI model benchmark pages, Meta's crawlers read 1,086 different pages in the week, and no major AI crawler asked for our llms.txt. A user agent is only a claim, and the address list is how you check it.

3D illustration of three lit glass blocks shaped like a podium on a dark mirror floor, knocked out of line with the tallest one tipped over, under a wide neon arc, while a small amber cube glows on the floor to the right.
Core update
AI Search StrategyNovember 2026 core updateGoogle core update
Nov 11, 2026 · 12 min read

November 2026 Core Update: What to Do Before Black Friday

Google started rolling out the November 2026 core update on November 10, 2026, at 10:06 a.m. Pacific time. It applies globally and to all languages, and Google says it may take up to two weeks. It's the third core update of 2026, after March and May, and the eighth ranking update on Google's Search Status Dashboard this year. A full two-week rollout would end around November 24, three days before Black Friday on November 27, so many stores and publishers will see rankings move in their busiest stretch. The March 2026 core update gives a sense of scale: SE Ranking data showed 79.5% of top-three URLs changed position, up from 66.8% after the December 2025 update, and about 24.1% of top-ten pages fell out of the top 100. Google advises waiting until the rollout ends, then at least a week more, before judging the impact in Search Console. For marketers, the best plan is to freeze big site changes, watch revenue rather than rank trackers, line up paid and email backups for key products and save the deep analysis for early December.

3D illustration of a glass hourglass between two dark discs ringed in neon on a mirror floor; its top bulb is full of small cyan sparks, while only a few amber drops fall through the narrow neck into a small warm heap in the bottom bulb.
AI impressions
AI Search StrategyAI impressionsAI impressions vs clicks
Nov 9, 2026 · 13 min read

AI Impressions vs Clicks: What Ziff Davis's 37% Drop Shows

On November 6, 2026, Ziff Davis said its content appeared in Google's AI Overviews and AI Mode about 2.7 billion times in the third quarter, nearly ten times the clicks its sites got from Google Search. In the same quarter, its organic Google referrals fell about 37% from a year earlier, after a 23% drop the year before, so it has lost more than half in two years. Revenue fell 5.6% to $289.3 million, and advertising and performance marketing fell 7.7% to $186 million. The 2.7 billion figure comes from Google Search Console's new generative AI report, which counts impressions only, with no clicks or queries. OpenAI appears to be building a similar dashboard for ChatGPT: a leaked screen shows daily impressions and clicks for verified businesses, but no prompts, user data or competitor views. Chartbeat says search fell to 12% of North American publisher pageviews, from 21% a year earlier. The lesson for marketers is that AI impressions measure reach, not demand. Track them next to clicks, conversions and branded search, and judge AI visibility by what it does to revenue.

Agentic Marketing

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

View all · 23 articles
3D illustration of a glass shopping bag glowing amber inside, with two thin neon handles, on a dark mirror floor; two neon tracks, one cyan and one violet, curve in from each side with a bright orb on each, racing toward the bag.
AI shopping agents
Agentic MarketingAI shopping agentsAlexa for Shopping
Nov 8, 2026 · 14 min read

AI Shopping Agents: What Amazon vs Meta Means for Marketers

Amazon and Meta are racing to own the AI agent that shops for people. On November 6, 2026, Amazon let invited US customers text Alexa for Shopping to search, compare, build a cart and check out inside the message thread, with millions more due in the coming weeks. The same day, Meta added spaces to its Muse agent, small tools people build by describing them, such as a holiday gift tracker. A day earlier, Muse gained real-time voice and expressive avatars in the US. The race follows Amazon's block on Muse, in place since September 20, which Amazon says is about permission, bots that don't identify themselves and stored passwords. Both sides report strong use: more than 400 million customers used Alexa for Shopping in the past year, and over 25% of Muse users shop with it, at more than $200 an order. Adobe forecasts $275.1 billion in US online holiday spending and a 130% rise in traffic from AI services. For marketers, the shift means agents, not people, will read more product pages, so product data, feeds and measurement need to work for machines as well as shoppers.

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.

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 · 33 articles
3D illustration of a tall dark column covered in a grid of identical violet squares beside a low block holding three small amber spheres of different sizes, with a cyan neon magnifying glass hovering in front of the tall column.
GA4 bot traffic
Governance & ROIGA4 bot trafficDirect traffic spike GA4
Nov 16, 2026 · 8 min read

GA4 Bot Traffic: How One Visitor Made 88% of Our Sessions

In October 2026, vanaxity.com's Google Analytics 4 property showed 15.8 times as many sessions as in September. Without one automated visitor, the rise was 1.7 times. From October 1 to November 15, 2026, that visitor made 1,312 sessions, 88.1% of all sessions and 91.7% of those GA4 filed under Direct. Every one of them came from Singapore, on a 1280x1200 screen, in Chrome on Windows, as a new user who viewed one page, stayed zero seconds and fired only GA4's automatic events. GA4's built-in bot exclusion didn't catch it, and GA4's data filters can't use country or screen resolution. We built an exploration segment that removes it from our reports, and we changed our tag so it no longer loads in browsers that announce automation through navigator.webdriver. Other sites have reported the same pattern since October 2025, so check your own Direct traffic before you report any growth.

3D illustration of a lit glass block on a dark round plinth at the center of a mirror floor, circled by six rings of small glowing cyan and violet beads with a few warm amber beads among them.
Anthropic S-1
Governance & ROIAnthropic S-1Anthropic IPO
Nov 13, 2026 · 12 min read

Anthropic S-1: What the IPO Filing Says About AI Spending

Anthropic's IPO is about six times oversubscribed, Bloomberg reported on November 12, 2026, citing people familiar with the deal. Its amended S-1 filing of November 10 offers 825 million Class A shares at $70 to $78 each, to list on Nasdaq as ANT. Seven companies, including Microsoft, Nvidia, Google and AMD, agreed to buy $26.0 billion more in a private placement at the IPO price, and up to 20% of the IPO shares are expected to go to retail investors. At the top of the range, the IPO would raise about $64.4 billion, or about $90.4 billion with the placement, which reports say would be the largest listing ever. The filing also shows how fast companies are spending on AI: revenue was $33.2 billion in the first nine months of 2026, up from $2.5 billion a year earlier, and more than 11,900 customers spent $100,000 or more over the past year, up from more than 950. Yet Anthropic says its Fortune 500 customers spend on average less than 0.1% of their total expenses with it. For marketing teams, that means AI vendors will push hard to grow those budgets, so benchmark your spend, negotiate on volume and keep a second model ready.

3D illustration of a small four-rotor drone with glowing cyan rotor guards hovering above a round dock inside the neon outline of a house on a dark mirror floor, with one small amber light on its nose.
Ring Drone Cam
Governance & ROIRing Drone CamRing Rover Cam
Nov 12, 2026 · 12 min read

Ring Drone Cam: How to Market AI Products People Fear

On November 10, 2026, Amazon's Ring launched Ring Robotics with two products. The Ring Drone Cam is a $249.99 indoor flying camera that records 2K video, flies for up to eight minutes per charge by Ring's own test and ships in the US on January 27, 2027, if the FCC approves it first. The Ring Rover Cam is a $999.99 wheeled security robot built on Amazon's Astro, records up to 4K video, maps up to 5,000 square feet and ships December 16, 2026. Both map a space with AI and need a Ring Protect plan for scheduled patrols. The launch comes nine months after Ring's Super Bowl ad for its AI Search Party feature drew a privacy backlash, and Ring canceled a planned Flock Safety integration four days later. So Ring leads with privacy: the drone records only while flying, has no microphone and goes dark in its dock. Coverage of the specs also conflicts, from flight time to weight, which shows why brands need one clear spec page that AI answers can quote. For marketers, the lesson is to answer the fear before the feature, publish facts in one place and capture demand honestly when a product can't be sold yet.

All articles

103 articles

October 2026

September 2026

August 2026

July 2026