AI Marketing News, 3 Aug 2026: GPT-5.6 Price Cuts, EU AI Act Transparency, Ads in Google AI Mode
OpenAI cuts GPT-5.6 Luna prices 80%, EU AI Act transparency rules take effect, and ads now appear in 29% of commercial Google AI Mode queries.
OpenAI cuts GPT-5.6 Luna API prices 80% and Terra 20%. Effective July 30, GPT-5.6 Luna dropped from $1/$6 to $0.20/$1.20 per million input/output tokens, and mid-tier Terra fell 20% to $2/$12, while flagship Sol stays at $5/$30. Luna’s new input price undercuts DeepSeek, and OpenAI simultaneously introduced a Fast Mode for the Sol API delivering 2.5x faster inference at double the standard price, replacing the old Priority Processing tier. The cuts land days after Anthropic shipped Opus 5 at half the compute cost of Fable 5.
Importance for marketers: An 80% cut on the small model changes which steps in your stack deserve a frontier model at all. Every high-volume classification, lead-scoring, or firmographic-parsing step in a Clay table or n8n flow that was priced out of per-row LLM calls in Q2 is now viable at $0.20 input. Re-run the cost math on your enrichment waterfall this week; the steps you hard-coded as regex or lookup tables to save money may now be cheaper to do properly. Provider order matters as much as price here: see enrichment waterfall provider order.
EU AI Act Article 50 transparency obligations became enforceable August 2. The AI Act’s transparency rules now legally apply: chatbots must make clear people are talking to AI, deepfakes must be labeled, and AI-generated or altered content must carry machine-readable marks under Article 50(2). The May 2026 AI Omnibus agreement gives generative systems already on the market until December 2 to meet the machine-readable marking requirement, and the Commission’s Code of Practice on marking and labeling, assessed as adequate in early July, is the recognized compliance route. Enforcement falls to national market surveillance authorities.
Importance for marketers: Two of the four duties land directly on marketing teams: synthetic content in EU campaigns engages Article 50(2), and any depiction of real people triggers the deepfake labeling duty in 50(4). The practical move is to inventory which assets in your campaign pipeline are AI-generated and confirm your tools embed C2PA metadata, because non-signatories to the Code of Practice should expect information requests and will have to defend whatever alternative they chose. If an AI chatbot or agent fronts your EU demand-gen motion, it needs an explicit AI disclosure now, not at the December marking deadline.
OpenAI is building a full ads business inside ChatGPT. Job listings on OpenAI’s careers page describe engineers building image, video, native, conversational, and interactive ad formats, moving well beyond the sponsored text unit tested with US free-tier users since February. The pilot crossed $100 million in annualized revenue with more than 600 advertisers by late March, and OpenAI is leaning on Criteo and Smartly to onboard advertisers while it builds its own stack, with reported talks with The Trade Desk. Emarketer forecasts meanwhile suggest OpenAI’s 2030 ad revenue target could miss by as much as 90%.
Importance for marketers: Conversational ad formats are the piece that matters for B2B: an ad that can answer follow-up questions inside ChatGPT competes directly with your SDR’s first-touch email, on a surface where roughly 910 million weekly users already are. The Criteo and Trade Desk involvement means buying will likely route through existing programmatic relationships rather than a new self-serve platform, so the near-term action is with your paid media agency, not a new login. Treat the Emarketer miss forecast as a pricing signal: early inventory on a platform under revenue pressure is where negotiating leverage lives.
Google rolls out AI content labels inside Google Ads asset studio. Starting July 28, Google began labeling AI-generated and AI-edited creative directly in asset studio, extending the “How this ad was made” disclosure system it launched across Search, YouTube, and Discover earlier in July. Assets produced with Google’s own generative tools are labeled automatically, while creative made with third-party tools relies on advertiser self-reporting backed by C2PA metadata detection.
Importance for marketers: The asset-studio placement means disclosure status is now visible at the point where creative gets built and trafficked, so agencies and in-house teams can no longer claim they didn’t know which variants carried labels. With Article 50 now enforceable in the EU, Google’s automatic labeling is doing part of your compliance work for you, but only for assets made inside Google’s tools. Anything generated in Canva, Midjourney, or a custom pipeline and uploaded still needs its own marking process, and C2PA detection will catch unlabeled AI edits whether you disclose them or not.
Study: ads appear on nearly 1 in 3 commercial queries in Google AI Mode, but ad spend doesn’t buy citations. SE Ranking analyzed 50,032 commercial keywords across 20 niches and found text ads in 29.45% of AI Mode responses, rising to 53.6% for keywords with CPCs of $10 or more. Where ads appeared, 71.1% of responses carried two ads. Only 11.5% of advertisers who bought an AI Mode ad were also cited as an organic source in the same answer.
Importance for marketers: The 11.5% overlap is the finding to act on: paying for AI Mode placement and earning citation in the answer are separate games with separate mechanics, and winning one does nothing for the other. High-CPC B2B categories are exactly where the 53.6% ad density lives, so your buyers are already seeing competitors’ ads inside AI answers on your money keywords. Split your reporting now, because blending “AI Mode ad impressions” and “AI answer citations” into one AI-visibility metric will hide the fact that most spend produces the first and not the second. Earning the citation side is a content infrastructure problem, mapped in citation infrastructure.
Knak study: 88% of AI-generated marketing output still requires human editing. Research from email and landing-page platform Knak, released July 28, found that 88% of marketers report AI-generated campaign output needs human editing before it ships, spanning tone corrections, factual fixes, and brand-compliance rework. The finding lands alongside a wave of martech vendors, including Algolia and StackAdapt, announcing deeper agentic features that generate more output, faster.
Importance for marketers: The 88% number prices the actual bottleneck: generation is no longer the constraint, review is. If your team is adopting agentic content tools without a defined editing pass, you’ve moved the cost from writing to QA without measuring it. The teams getting leverage from AI content are the ones who templatize the review step, tight brand rules and approved claims fed into the prompt up front, so the human pass is a check rather than a rewrite. Volume without that is just faster drafts.
Google tests AI-generated descriptions in Shopping ads. Google is testing AI-written product descriptions inside Shopping ads, extending an earlier experiment on Search ads. The generated copy draws from the advertiser’s product feed and landing pages, and appears without requiring advertisers to opt in or write the variants themselves. It follows Google’s move to bring AI Max query matching and dynamic ad copy into Standard Shopping campaigns.
Importance for marketers: Google writing your ad copy from your product feed means the feed is now the creative brief, and most B2B and DTC feeds were built for inventory sync, not persuasion. The lever worth pulling this quarter is feed quality: titles, attributes, and landing-page copy that read like positioning rather than a SKU database, because that is the raw material Google’s generation pulls from. Teams already piping enriched product data through n8n into their feed have a head start; everyone else is letting an LLM improvise from a spreadsheet.