← All insightsAI Weekly Roundup

The AI answer is now for sale

Google started testing ads inside personalized AI Mode, ChatGPT Ads added buyer tools, and a tagging bug made ChatGPT traffic look like it fell off a cliff.

By Matt Lawler, President and CEO6 min read

This week in one line

AI answers are becoming paid media, so give them their own test budget and check your tracking before you trust the numbers.

012 items

Major LLM & Model Developments

OpenAI's models are coming to Google Cloud.

Why it matters

Your IT team's choice of cloud no longer decides which model your marketing tools can use.

OpenAI and Google Cloud said OpenAI's frontier models and Codex will be sold through Google Cloud's Model Garden and Marketplace, with general availability planned in the coming weeks. Companies can apply existing Google Cloud spend commitments to the purchase. Microsoft stays OpenAI's primary cloud partner.

SourceOpenAI, Oct 21

Anthropic handed Agent Skills to an open foundation.

Why it matters

A playbook you write once, like your brand voice or your reporting steps, gets more portable across tools.

Anthropic is donating Agent Skills, its open standard for packaging the instructions an AI agent follows, to the Agentic AI Foundation, a Linux Foundation project. Anthropic says more than 70 agent platforms already use the standard, and engineers from Anthropic, Google and OpenAI will maintain it.

SourceAnthropic, Oct 22

022 items

User-Centric Use Cases & Features

Claude Duet works in the background and asks before it acts.

Why it matters

The assistant that watches your buyer's inbox is now a consumer product, so your first touch may be filtered by one.

Anthropic launched Claude Duet in research preview, a standalone agent app for professionals and small business owners. It sends a morning brief, checks connected apps every few hours, and handles tasks like scheduling, dashboards and forecasts. It asks for approval before it sends messages, makes payments or deletes anything.

SourceClaude, Oct 21

Google is putting more source links into AI answers.

Why it matters

Being chosen as a source is becoming a setting your buyer controls, not just a ranking.

Google says it's showing inline links higher in AI responses more often and is testing a section of official sites at the top of some answers. News carousels built on a user's Preferred Sources are expanding, and Google says Preferred Sources use has grown more than 7x since the start of the year.

SourceGoogle, Oct 15

034 items

Industry Trends & Policy Signals

Google is testing ads in personalized AI Mode.

Why it matters

If you buy search, the AI answer is your next paid channel.

Google confirmed it's testing ads in AI Mode for users who connect apps like Gmail through Personal Intelligence, reversing its March position that those answers would stay ad-free. Its help page says ads won't be based directly on Gmail or Photos content, but Search may use summaries and inferences from connected apps. Sensitive categories are excluded.

SourceSearch Engine Roundtable, Oct 19

ChatGPT Ads keeps adding tools for buyers.

Why it matters

It's being built as a real media channel, measured by partners your team may already use.

OpenAI told advertisers it's testing larger visual ad formats, redesigned the Conversions page, added weekly performance emails, and added measurement partners including Northbeam, DV Rockerbox and INCRMNTAL. The news came by advertiser email, not a public post.

SourceSearch Engine Roundtable, Oct 19

Your ChatGPT traffic drop may be a tracking bug.

Why it matters

If AI referrals fell on your dashboard between Oct 9 and 15, check the tagging before you change the plan.

After GPT-6 rolled out, many ChatGPT links lost their tracking tag, so visits showed up as Direct in GA4. Profound measured answers with at least one tagged link falling from about 98% on October 8 to 66% by October 14, and Graphite reported ChatGPT-attributed sessions down 22% to 57% on sites it tracks. OpenAI hasn't confirmed a bug publicly.

SourceSearch Engine Journal, Oct 15

GEO is feeding a content-farm boom.

Why it matters

Your programmatic budget is likely touching some of it.

DoubleVerify data reported by Axios shows low-quality AI-generated sites rose more than fivefold in Q2 2026 versus a year earlier. Monthly ad requests to those pages grew from 184 million in January to 834 million in September.

SourceAxios, Oct 17

04 · Decisions

Strategic Takeaways for Marketing Leaders

Three calls we'd make this week if we ran your marketing.

Decision 01

Give AI answers their own test budget.

Google and OpenAI are both selling space inside the answer. Run it as a test with its own incrementality read, not as a line inside paid search.

Decision 02

Check the tagging before you trust AI traffic.

One missing tag erased up to 57% of reported ChatGPT sessions for some sites in a week. Verify before a falling number changes the plan.

Decision 03

Measure presence in the answer, not just clicks.

Clicks undercount AI influence, and mentions and citations tell different stories. Track whether you're named, on a fixed question set, on a schedule.

05 · Actions

What CMOs Should Do This Week

Each one fits before Friday. Check them off as you go.

LVL3 data

127 of 620

Reachable top domains with a real llms.txt file

From our data

The content-farm story is an industry spending money to look good to AI. We tested one of the most-sold tactics. Of the 1,000 most popular domains, about one in five reachable sites publish a real llms.txt file. Across 397 home services companies, having one showed no measurable lift in Google AI Overview appearances. Don't pay for one. Spend that money on crawler access and real search strength.

Read the llms.txt Reality Check

Find out how you actually show up in AI answers.

A systems audit runs our method on your business: a citation baseline on your own buying questions, plus the entity and citation gaps behind it.

Book a systems auditYou get the findings either way.
Let's talk.