The Visibility Report #10

Week of March 24-30, 2026

Six weeks. That's how long it took ChatGPT's ad pilot to cross $100 million in annualized revenue, and it's still only showing ads to about one in five eligible users. Self-serve access opens in April. This week's issue focuses on what that means for marketers on both sides: the ones buying ads and the ones whose brands are being surfaced (or poached) inside AI conversations.

🔥 ChatGPT Ads Hit $100M. Self-Serve Opens in April.

OpenAI crossed $100 million in annualized ad revenue from its ChatGPT pilot, launched just six weeks ago. More than 600 advertisers are on the platform, but fewer than 20% of eligible free-tier and Go users are currently seeing ads. The current revenue represents a fraction of eventual capacity.

Wired's Reece Rogers asked ChatGPT 500 questions on mobile and found roughly one in five triggered an ad, shown as a tailored link button at the bottom of the response. Targeting is topic-based and draws from conversation context plus whatever ChatGPT has stored in memory about the user. Travel questions triggered ads most frequently. Brand mentions triggered competitor ads. Ask about DoorDash, get served Uber Eats. Anyone who's managed search campaigns will recognize the poaching dynamic.

OpenAI says ads don't influence answers and that full conversation content isn't shared with advertisers. Early signals show low ad dismissal rates. But Sam Altman called ads "a last resort" just two years ago. The company is now expanding to Canada, Australia, and New Zealand.

What this means: Self-serve access in April shifts ChatGPT from managed pilot to open advertising channel. For paid teams, evaluate the inventory now, especially in travel, productivity, and consumer tech where early targeting appears strongest. For organic teams, your brand name is already triggerable ad inventory for competitors. Neither Gemini nor Claude currently runs ads, but Google hasn't ruled it out.

🔗 Search Engine Land | Search Engine Land (frequency) | Wired​

📊 AI Visibility Research

Gemini Doubles Referral Traffic, Passes Perplexity

SE Ranking analyzed data from 101,000+ sites with Google Analytics and found Gemini more than doubled its referral traffic between November and January. That's a 115% increase in two months. The acceleration lines up with Google's Gemini 3 rollout: Pro in November, Deep Think in December, Flash (now the default in the Gemini app and AI Mode) in December.

In August, Perplexity sent roughly 3x more referral traffic than Gemini. By January, Gemini had flipped it. 29% more visitors globally, 41% more in the US. ChatGPT still drives about 80% of all AI referral traffic, but its lead over Gemini narrowed from 22x to 8x in three months. All AI platforms combined represent about 0.24% of global internet traffic. The absolute numbers are small. The trajectory is not.

Why it matters: If you're only monitoring ChatGPT for AI visibility, you're missing the fastest-growing referral source. Gemini's surge suggests that ecosystem integration (Search, Chrome, Android) gives it a distribution advantage standalone tools can't match.

🔗 Search Engine Journal | SE Ranking​

AEO: How AI Actually Selects Content for Responses

Search Engine Journal published part two of a five-part series on optimizing for the agentic web. It's got some good tactical guidance. The big idea: AI doesn't rank pages. It selects fragments. Microsoft's Bing team described how AI assistants "break content down into smaller, structured pieces that can be evaluated for authority and relevance," then assemble answers from multiple sources.

The academic research backs this up. The foundational GEO paper (Princeton/IIT Delhi/Georgia Tech) found that citing credible sources produced a 115% visibility increase for lower-ranked sites. One surprising finding: writing in an authoritative or persuasive tone did not improve AI visibility. AI systems respond to verifiable information, not rhetorical style. The University of Toronto's large-scale analysis found AI overwhelmingly favors earned media over brand-owned content.

The takeaway: Structure content in extractable fragments with clear claims backed by cited sources. AI is picking the best pieces from many pages, not the best page.

🔗 Search Engine Journal​

Stop Chasing Reddit and Wikipedia for AI Visibility

Search Engine Land makes the contrarian case against the Reddit/Wikipedia obsession sweeping marketing teams. Yes, they're the most-cited domains across major LLMs. But that's because they're massive sites with topical footprints spanning millions of categories, not because they're the right strategic target for your brand.

The problem: CMOs are seeing aggregate citation data and pivoting their entire GEO strategy toward Reddit. But high-ranking Reddit threads on bottom-of-funnel queries can't be reproduced. They're the result of years of authentic peer reviews, not something you can reverse-engineer with a "Reddit SEO agency." Wikipedia's editorial guardrails make it even harder to influence.

The real play: Focus on the sources AI actually cites for your specific category, not the sources that dominate aggregate charts. Your AI competitive map is different from the macro data.

🔗 Search Engine Land​

🚀 Platform Updates

Google-Agent: The Agentic Web Is Here

Google announced a new user agent called Google-Agent, built specifically for AI agents browsing on behalf of users. When an agent using Google infrastructure (like Project Mariner) visits your site, it identifies itself with this new tag.

Marie Haynes maps the broader ecosystem: MCP (Model Context Protocol) lets agents access your backend data. A2A (Agent2Agent) enables bot-to-bot transactions. UCP (Universal Commerce Protocol) lets machines buy products from the SERPs. WebMCP lets agents use your website's functionality natively, not by parsing pixels, but by interacting with tools and forms directly in real time.

Google's head of search, Liz Reid, described a near-future where "a lot of agents are talking with each other." If AI agents can fill out your forms, browse your catalog, and negotiate with your backend, the question stops being whether your site is optimized for humans. It becomes whether it's optimized for machines acting on behalf of humans. This extends the AAIO framework we covered in VR #9.

🔗 Marie Haynes | Semrush​

YouTube Tests Replacing Video Titles With AI Summaries

Some YouTube users on Android are seeing video titles replaced entirely by AI-generated summaries. Thumbnails remain, but the title is gone. Your keyword signal, brand voice, and click-through driver, replaced by an expandable AI-written synopsis. This mirrors Google's recent testing of AI headline rewrites in Search results.

Why it matters: If this rolls out broadly, YouTube titles become a signal for the algorithm rather than a message to the viewer. Creators should make sure content is clearly structured and front-loads key information in the first 30 seconds. That's likely what the AI summary draws from.

🔗 Search Engine Land​

The Death of the Static Google Business Profile

The 2026 Local Search Ranking Factors report shows behavioral and engagement signals climbing fast in importance. Posts, photos, clicks, calls, direction requests, review cadence. Google is rewarding businesses that "look alive." One finding worth pausing on: being open when users search is now the #5 local pack ranking factor. A BrightLocal study found rankings drop when a business is listed as closed.

Action item: Post weekly, respond to reviews within 24 hours, upload fresh photos monthly, audit hours quarterly. Businesses doing this are outperforming competitors who optimized once and walked away.

🔗 Search Engine Journal​

Reddit Declares War on Bots

Starting March 31, Reddit is rolling out mandatory [App] labels for automated accounts, human verification for suspicious behavior, and expanded bot reporting tools. Reddit already removes about 100,000 accounts per day for spam. The new rules formalize what was already happening and add visibility: if an account is automated, users will know.

For brands relying on Reddit for AI citation strategy (see above), this reinforces the same message: authentic participation wins. Gray-area automation is about to be exposed. Legitimate bot accounts must register by June or risk removal.

🔗 Foundation Inc​

📝 From the Tool Blogs

• Scrunch:The Half-Life of AI Citations (3.5M Events) · ChatGPT cycles through sources fastest (3.4-week half-life). Perplexity holds citations nearly 70% longer (5.8 weeks). Google clusters in between (4.3-4.8 weeks). Citation strategy requires playing both offense and defense.
• Scrunch:Content Gaps: Find and Fix Your AI Blind Spots · New feature auto-detects when your site is missing content for tracked prompts, prioritized by topic, persona, and urgency.
• Scrunch:From Keywords to Prompts: Right-Size Your AI Search Tracking · Practical formula for determining how many prompts to track: topic clusters × 12-15 questions per cluster. Prompts are about 5x longer than keywords with far more intent specificity.
• Scrunch:The ROI of AEO (and How to Prove It) · Why traditional attribution models fail for AEO, and how to report on impact metrics that hold up in budget meetings.
• Otterly AI:Does Schema Markup Really Impact AI Search? · Controlled experiment across 7 AI platforms. Only Gemini correctly fetched JSON-LD. Google AI Mode hallucinated schema types that weren't even implemented. But schema still delivered +377% SERP features and +1,500% AI Overviews within Google's ecosystem.
• AirOps:Prompt Mining: Turn Customer Questions Into AI Search Tracking · Connect Gong, Zendesk, or Intercom to extract the actual questions buyers ask, then pipe them into tracking via MCP. Your competitors use generic keyword lists. You have first-party data they don't.
• Profound:We Reverse-Engineered ChatGPT's Shopping Trigger · Prompts naming shippable consumer goods trigger Shopping cards; software, services, and travel almost never do. Their classifier reproduced ChatGPT's behavior with 95-97% accuracy.

💡 Strategy & Frameworks

Wikipedia Bans AI-Generated Content

Wikipedia published new guidelines prohibiting editors from using LLMs to write or rewrite articles, with two narrow exceptions: basic copyediting (with human review) and LLM-assisted translations. The reasoning ties to Wikipedia's core policies (verifiability, no original research, neutral point of view), all of which LLMs routinely violate through hallucination and synthesis without citation.

Why this matters for AI visibility: Wikipedia is one of the most-cited sources across every major AI platform. A policy change that constrains what content appears there has downstream implications for every AI system that draws from it. If you're thinking about Wikipedia as a citation target, the rules just got stricter. And more human.

🔗 Search Engine Journal​

Query Fan-Out: The Invisible Layer Driving AI Citations

Foundation Inc published a deep primer on query fan-out mechanics and what they mean for B2B marketing. When a buyer asks "what's the best CRM for a mid-market SaaS company?", the AI doesn't search that phrase. It breaks it into dozens of sub-queries (pricing comparisons, integration capabilities, reviews from specific verticals) and synthesizes the answer from everything it finds across all of them.

The critical data point: nearly a third of AI citation opportunities come from sub-queries invisible to any keyword tool. Your content may answer the head query perfectly but miss the sub-queries that actually drive the citation. The piece includes specific B2B playbook recommendations for restructuring content around the full query tree.

🔗 Foundation Inc​

Google's March 2026 Core Update + Spam Update

Google launched the March 2026 core update on March 27, the first core update of the year. It's expected to take up to two weeks. Google describes it as "designed to better surface relevant, satisfying content from all types of sites." Meanwhile, the March 2026 spam update launched on March 24 and finished in under 20 hours, the fastest spam update completion on record. The speed suggests Google's spam detection systems are increasingly confident.

What to do: Monitor your traffic through mid-April. If you see movement, don't panic-publish. Review whether your content actually satisfies the queries you're targeting. The "from all types of sites" language is worth noting. It echoes Google's ongoing effort to diversify results beyond the same dominant domains.

🔗 Search Engine Journal | Search Engine Roundtable​

🔮 What to Watch

1. ChatGPT self-serve ads launch in April. The moment the platform shifts from pilot to open channel. Early movers in search and social advertising saw the same window. Evaluate now, before competition drives up costs.
2. Google-Agent and the agentic protocol stack. MCP, A2A, UCP, WebMCP. Google is building plumbing for a web where AI agents browse, transact, and negotiate on behalf of users. This isn't a 2027 problem. It's Q2 2026.
3. Citation decay is faster than you think. Scrunch's half-life data shows ChatGPT churns through sources every 3.4 weeks or so. If you're not continuously refreshing your content, your AI citations are quietly expiring.
4. OpenAI killed Sora. Six months after launch, the video generation tool is dead. If you're evaluating AI tools for your marketing stack, bet on capabilities with proven product-market fit, not impressive demos.

The Visibility Report #10 | Will Scott
This newsletter is produced collaboratively by Will Scott and Bob, an AI agent. Human oversight, AI efficiency.
​