Blog
Insights for AI advertisers
Guides, strategies, and comparisons for creating AI video ads that convert.
Gemini Agentic Video Understanding Cuts Long-Form Analysis Costs Up to 66%, Google Says
Google launched agentic video understanding on September 1, 2026, letting Gemini hunt through a video instead of sampling every second. The savings land on hours of footage, not on six-second ads, and the numbers are Google's own.
Google's Official Ads MCP Server Is Read-Only: What Agent Connectors Can and Cannot Do
Google's own Google Ads MCP server ships three tools and lists its mode as read-only, so an agent can query your account through it but cannot pause a campaign. Third-party connectors already write, and the difference is the thing to check before you hand one an account.
Reddit's 15-Second Engaged Video Views: When to Optimize for 15 Seconds, Not 6
Reddit opened 15-second Engaged Video Views to every advertiser in public beta on September 1, 2026. Here is what the counting rule changes about how you cut a video ad, and when the 6-second goal is still the right call.
FTC Sues Amazon Over $20 Billion in Alleged Ad Surcharges: What Advertisers Can Control
The FTC and 22 states say Amazon quietly rewrote its ad auction in 2019 and never told buyers. The transferable lesson is not about Amazon: you cannot audit an auction from inside the platform that runs it.
Adobe Firefly Adds 3 AI Audio Tools for Ads: What Commercial Safety Actually Covers
On August 20, 2026 Adobe made Generate Music, Generate Speech and Generate Sound Effects generally available in Firefly, and sold the licence as the product. Here is what Adobe's own indemnification terms say in writing, and where the covered and uncovered outputs sit side by side in one interface.
ChatGPT Ads Reaches 31 European Markets With No Custom Audiences: What You Can Buy
ChatGPT Ads went live across 31 European markets on August 24, 2026, and OpenAI's developer documentation blocks custom audiences for any campaign targeting the EEA or Switzerland. Here is what those campaigns can actually buy, and who is allowed to buy them.
Microsoft Ads Removes Max CPC October 1: The 3 Bid Strategies That Keep It
From October 1, 2026, new non-portfolio campaigns on Microsoft Advertising lose the Max CPC field, while AI Max rolls out globally and widens what a campaign is allowed to match. Portfolio, Target Impression Share and Enhanced CPC keep the ceiling.
Nano Banana Pro for Ad Creative: Legible Headlines, Price Badges, and the 2K Sweet Spot
Nano Banana Pro is no longer the only model in its family that reaches 4K, and for ad creative that was never the interesting part. Here is what still makes it the right pick for text-heavy statics, and how the resolution you choose decides what a variant set costs.
AI Video Credits Explained: What 1 Credit Buys, and How to Compare Vendor Prices
A credit is a private currency. The vendor defines it, it changes per feature, and it converts into video only through a table the vendor controls. Here is how to turn any credit price into a cost per finished ad.
Google's New AI Ads and Analytics Tools Are Gated: Check These Four Account Settings
Google's August 10, 2026 release footnotes three of its new AI tools as beta, English-only. Here are the four account-level gates that decide whether any of it reaches you, and which ones you can actually open yourself.
A 22B Open-Weights Video Model Now Renders Multishot Ads on Hardware You Already Own
LTX shipped a new open-weights release on August 11, 2026, and the genuinely new capability is multishot: one generation that holds character, scene and voice across cuts. Here is what changed, what was already there, and what it costs.
Mirage Avatar X costs 71% more per second on fal than from the vendor
Mirage lists Avatar X at $0.175 per second on its own API while fal lists the same model at $0.3 per second, and the billing increment underneath those numbers changes the answer again.
How Long Does AI Video Generation Take? 60 Days of Measured Production Renders
Across 1,045 production renders in 60 days, the median end-to-end wait ran from 45 seconds for a still image to 435 seconds for the slowest video model. Here is the measured table, with the sample size printed beside every number.