AI Tidbits – AI Grows Up: Ads, Agents, and Personal Intelligence

AI Tidbits – AI Grows Up: Ads, Agents, and Personal Intelligence

🤖 Welcome to 2026: Where AI Gets Busy and Gets Paid

If 2025 was the year AI proved it could work, 2026 is shaping up to be the year it learns how to monetize, remember, repeat itself, and quietly run entire projects without asking permission.

This week’s updates hint at a future where AI agents don’t just assist you — they stick around, learn your life, and occasionally sell you something along the way.

Let’s dive in.

 

💰 OpenAI officially brings ads to ChatGPT

OpenAI has confirmed it will begin testing targeted advertisements inside ChatGPT for Free and Go-tier users in the U.S., marking a major shift in how AI products may sustain themselves at scale.

Ads will appear below responses as Sponsored Recommendations, tailored to conversation context — but explicitly excluded from health, politics, and underage users.

This move coincides with the global rollout of the $8/month ChatGPT Go tier, with ads helping subsidize the lower price point. Importantly, Plus, Pro, Business, and Enterprise plans remain ad-free, and OpenAI has reiterated it will not sell user data or let ads influence answers.

Sam Altman once called ads a “last resort.” Turns out, scale has a way of redefining last resorts.

 

🧠 Do Anything launches AI agents that work autonomously for months

Startup Do Anything just unveiled a new class of long-running AI agents designed to operate independently for months at a time.

Each agent gets:

• Its own name
• Its own email address
• The ability to take actions as itself — not impersonating you

Powered by a framework called MEESEEKS, these agents aim to manage entire projects, not just tasks. Think less “assistant” and more “junior operator who never sleeps.”

Explore it here:
👉 doanything.com

 

🧩 Google launches Personal Intelligence with Gemini

Google has introduced a beta feature that turns Gemini into a proactive, personalized assistant by connecting it to Gmail, Photos, YouTube, and Search.

Ask for recommendations in a new city, and Gemini can now:

• Pull travel dates from Gmail
• Infer interests from Photos
• Factor in your actual schedule

This isn’t just smarter search — it’s contextual AI grounded in your life.
Rolling out to Google AI Pro and Ultra subscribers and turned off by default (for now).

 

🔁 When in doubt… say it twice!!

Prompt repetition improves “non-reasoning” LLMs.
Yes, that’s real research — and it’s surprisingly effective.

Google researchers found that simply copy-pasting your prompt twice improved accuracy across Gemini, GPT-4o, Claude, and DeepSeek (with reasoning mode off).

Results:

• 47 wins out of 70 benchmarks
• Zero losses
• Accuracy jumps of up to 76 percentage points

This doesn’t replace well-structured prompts — but it adds a ridiculously simple trick to your AI toolkit.

Repeat it.

 

😊 When in doubt… say it twice!!

Prompt repetition improves “non-reasoning” LLMs.
Yes, that’s real research — and it’s surprisingly effective.

Google researchers found that simply copy-pasting your prompt twice improved accuracy across Gemini, GPT-4o, Claude, and DeepSeek (with reasoning mode off).

Results:

• 47 wins out of 70 benchmarks
• Zero losses
• Accuracy jumps of up to 76 percentage points

This doesn’t replace well-structured prompts — but it adds a ridiculously simple trick to your AI toolkit.

Repeat it.
😊😊😊

 

📉 Important news for Digital Media Planners

Global publisher traffic from Google dropped by nearly one-third in 2025, with declines expected to continue into 2026.

Since May 2023:

• Google Search referrals: -21%
• Google Discover: -18%
• All external referrals: -24%

The message is clear: distribution strategies built solely on search are aging fast.

Full report:
👉 Google-Traffic-Down-2025-Trends-Report-2026

 

🏥 Health AI snippets

A quick pulse check on healthcare AI this week:

• Google quietly removed certain AI health summaries from Search after concerns about misleading medical interpretations
• Anthropic is pushing Claude into real healthcare workflows with HIPAA-ready tools for admin and clinical operations
• OpenAI launched ChatGPT Health — a separate, encrypted space that helps users understand medical records (not diagnose)

Healthcare remains AI’s hardest test: high stakes, high impact, zero room for hallucinations.

 

😄 To sign off this week…

“Easier to match human intelligence if the human isn’t intelligent.”

— Dilbert, by Scott Adams

Sometimes satire explains the AI era better than whitepapers ever could.

See you next Tuesday. Stay tuned 😊