AI Is Reading Your Searches. And Taking Notes.

If you've checked Google Search Console recently, you've probably spotted some odd queries: "yes," "what about Resend?," "compare with the previous version." They're not spam and they're not scraper bots. They're fragments from conversations users are having with AI Mode in Google Search, and they're now showing up as separate queries in your reports.

A recent piece on Search Engine Journal documented the phenomenon in detail. Follow-up responses from AI Mode (things like "yes, but for a smaller budget" or "what alternatives exist?") are recorded by Search Console as distinct queries. In practice, you can now see, for the first time, what people ask after the AI gives them an initial answer. The fragments cluster into seven types: reply artifacts, pivot follow-ups, conversational questions, tracker probes, agent commands, pasted strings, and long uncategorized queries.

The truly interesting part? 57.7% of impressions carry no query string at all. Anonymization hides the majority of the data. What you see in Search Console is just the tip of the iceberg. And the new Generative AI report in Search Console shows impressions and pages but intentionally excludes queries and clicks. Google gives you a window, but with the blinds half drawn.

What you can do right now: search the standard query report for conversational fragments. If you find queries like "compare X with Y" and don't have a dedicated page making that comparison, you've identified a content gap that AI is exposing without knowing it. Exactly the kind of insight that traditional SEO metrics no longer provide. The free classifier tool mentioned in the article automates this analysis, but even a manual review of 50 queries can reveal patterns no standard report contains.

73% of Marketers Use AI Daily. 70% Don't Measure the Impact.

Social Media Examiner published data from their 2026 AI marketing report, and the numbers tell two completely different stories. First: adoption has exploded. 73% of marketers use AI daily, up from 37% two years ago. 90% use it at least weekly. Tools like ChatGPT, Claude, or Jasper have become as routine as Google Docs or Canva. Second story: almost nobody measures what it actually does. Only 30% can quantify concrete results, even though 86% say AI saves them time and boosts productivity.

The numbers get even more interesting around learning. 85% of marketers teach themselves AI through daily experimentation. 53% pay for tools out of their own pocket, without waiting for the company budget. This means adoption isn't trickling down from executive strategy decks or formal training programs. It's happening organically, one employee at a time, and management is only now starting to notice.

Where is the biggest gap? Surprisingly, not in text or image generation. It's in video. 74% want to learn AI video creation, but only 37% actually do it. That 37-percentage-point spread is the largest gap in the entire report between aspiration and action. And the next wave building? Autonomous AI agents, which 79% of marketers plan to adopt. The real differentiator between companies that win and those that don't won't be whether they use AI, but whether they measure what truly matters. When 73% adopt but 70% don't measure, you have an entire market flying on autopilot with no instruments.

llms.txt V2: The Spec Google Ignores, But AI Agents Don't

The llms.txt specification received a major V2 update that adds formal Markdown linking for AI agents. Specifically, the update introduces two new URL patterns (you can serve Markdown at /page.html.md or /page.md) and two relationship attributes: rel="alternate" and rel="describedby". Through these, AI agents can automatically discover Markdown versions of your pages via HTML link tags or HTTP headers.

Google has said clearly: it doesn't use these files and their presence won't affect rankings. But that's an answer to the wrong question. llms.txt isn't for Google Search. It's for Claude, for GPT, for Copilot, for any AI agent trying to understand your site's content when generating a response. Thousands of sites already publish it. Platforms like Mintlify auto-generate it. And Anthropic, OpenAI, and even Google themselves publish llms.txt for their developer documentation.

If visibility in AI-generated answers is on your priority list, llms.txt V2 is a small step with big potential. Not because it solves everything, but because it makes your site easier to process for the agents already reading your content. In an ecosystem where more and more purchase decisions pass through AI, being machine-readable is no longer optional.

YouTube No Longer Counts Views the Way You Think

Starting August 24, YouTube is fundamentally changing how it counts views on long-form videos and live streams. The new rule: a view is registered the moment a video starts playing, with no minimum watch time threshold. Until now, you needed to watch for several seconds for a view to count. Starting next week, the first fraction of a second is enough. The predictable result: public view counts will increase significantly for all creators, regardless of content quality.

YouTube is simultaneously introducing a separate metric, Engaged Views, which measures viewers who stay past the initial seconds (for Shorts) or who actually watch long-form content. Monetization stays tied to Engaged Views, not the new inflated view count. In practice, YouTube is officially separating exposure from engagement, something marketing agencies should have been doing for a long time.

For brands and agencies, the change raises an immediate practical question: if you evaluate creator performance based on public views, starting next week those numbers will be artificially higher. Comparisons with previous campaigns become invalid. Engaged Views becomes the only relevant metric for real budget decisions. And for those producing branded video content, the gap between an empty view and an Engaged View will separate genuine reporting from cosmetic numbers.

A Common Thread

All four changes this week share something: the data you rely on has either changed or doesn't show what you think it shows. YouTube views no longer mean watching. Search Console is surfacing fragments of AI conversations, not traditional searches. 73% of marketers use AI daily, but 70% have no idea whether it's working. And a simple text file might be the key to your visibility in the AI engines you've been ignoring.

Your dashboard isn't lying. But it's not telling you everything either.