AI marketing news September 2026: ChatGPT ads, Gemini, agents and the tools marketers adopted

The biggest AI marketing story of the past five weeks is money: OpenAI said on 31 August 2026 that ChatGPT Ads passed a $1 billion annual run rate in under 200 days and opened self-serve buying to India and 31 European markets. Behind it sit new frontier models from OpenAI, Anthropic and Google, EU labelling rules now in force, and ad platforms handing controls to AI agents.

ChatGPT Ads passes $1 billion and opens to India and Europe

31 August 2026 · Source: ppc.land, Adweek

What happened. OpenAI announced that ChatGPT Ads, which began testing in February 2026, crossed a $1 billion annualised revenue run rate in under 200 days. Adweek reports the company is targeting $2.3 billion in ad revenue for 2026, with a $100 billion goal by 2030, and that self-serve buying now covers 31 European markets. In India, Outlook reports Ads Manager opened on 4 September with daily budgets from INR 725, more than 50 launch brands, and WPP and Omnicom as the first agency partners. Ads show on the free and Go tiers only.

The product also grew up. ppc.land lists cost-per-click bidding, a conversions API, pixel measurement, carousel ads for product feeds and a LiveRamp integration, and counts roughly 1,200 unique US advertisers in August, up 43% month on month since May. The category mix moved fast. Financial services went from 2% to 13% of US spend between April and August, travel from 5% to 11%, and shopping slipped from 37% to 21%.

Why it matters. A run rate is a projection, not banked revenue, and the State of AI Marketing points out OpenAI released no CTR, conversion or CPA data. One agency test it cites, Choice OMG, saw a 0.65% CTR on 8,940 impressions and zero leads. But the vertical shift is the signal. Banks, brokers and travel brands are buying because ChatGPT users ask long, high-intent questions, and those advertisers can afford CPCs that OpenAI reportedly suggests start at $3 to $5.

What to do. Run a small, tightly measured test rather than a launch. Install the conversions API before spending, cap it at two weeks, and judge it against your paid search benchmarks, not against Meta. Our PPC news for September 2026 covers the bidding and measurement changes in more detail.

OpenAI ships GPT-6 Astra and cuts GPT-5.6 Sol prices

3 September 2026 · Source: Fortune, Al Jazeera

What happened. OpenAI released GPT-6 Astra to a limited set of enterprise customers on 3 September, with Plus, Pro and Enterprise users, the API and AWS following “in the coming days”, according to Fortune. The headline capability is computer use: Greg Brockman demonstrated the model moving through spreadsheets, forms and web pages on its own. OpenAI claims 99.9% on ARC-AGI-3 with tools, against 7.8% for GPT-5.6 Sol. Al Jazeera notes the public version rejects some prompts in areas such as cybersecurity, and that the launch landed amid political scrutiny after a July incident involving OpenAI agents at Hugging Face.

Two weeks earlier, on 21 August, OpenAI cut GPT-5.6 Sol API pricing by more than 20%, to $4 input and $20 output per million tokens through at least 21 November, per Technology.org.

Why it matters. Computer use is the feature that turns a chatbot into a junior media buyer. A model that can open Ads Manager, pull a report and change a bid does not need a native integration. That cuts both ways for platforms, which is why so many are shipping official connectors (see the MCP section below). And the price cut is a reminder that model costs keep falling, so any AI content or analysis workflow you priced in June is now cheaper.

What to do. Do not give an agent write access to ad accounts yet. Give it read access, let it draft, and keep a human on the approve button. Our AI in digital marketing guide has a practical stack by team size.

Anthropic releases Claude Fable 5.1 and signs Claudeforce with Salesforce

1 September 2026 · Source: VentureBeat, Salesforce

What happened. Anthropic released Claude Fable 5.1 on 1 September, priced at $10 input and $50 output per million tokens with cache reads cut from $1.00 to $0.25, according to VentureBeat. Anthropic positioned it for multi-hour agentic work and cited a 38-hour unattended Ramp workflow and an 82% completion rate on Browserbase’s hardest browser-agent benchmark. A restricted sibling, Mythos 5.1, goes only to vetted security and life-science organisations.

On 26 August, Salesforce and Anthropic announced Claudeforce. Claude now powers Agentforce Vibes and Agentforce Coworker by default, and a “Salesforce in Claude” plugin with 37 prebuilt sales skills is in pilot with open beta expected in September 2026.

Why it matters. For marketing teams on Salesforce, this decides which model sits behind Marketing Cloud journeys and Agentforce campaign agents. The cache price cut is the more practical news. Long brand guidelines, product catalogues and tone documents are exactly the kind of repeated context that cache pricing rewards, so brand-safe content pipelines get roughly 25% cheaper overnight.

What to do. If you run Salesforce, ask your account team for the open beta date. If you run content ops on any Claude model, restructure prompts so the brand context sits in a cached prefix. The content marketing guide covers AI production with quality control.

Google puts Gemini 3.7 Flash inside AI Mode

14 August 2026 · Source: Search Engine Journal, Google

What happened. Google released Gemini 3.7 Flash on 13 August, three weeks after 3.6 Flash, at an introductory price of half the cost per million tokens, per Google’s August roundup. One day later it appeared as a selectable model inside AI Mode for Google AI Pro and Ultra subscribers, reported by Search Engine Journal. Google has not said whether it becomes the default. The same roundup says the Gemini app passed 1 billion monthly users, 63% of whom use voice, and that Gemini Omni 1.1 Flash added scene extension and 4K upscaling for video generation.

Why it matters. The model behind AI Mode determines which pages get cited. Two models, two citation patterns. Marketers tracking AI Mode visibility can now test the same query under Auto, Pro and 3.7 Flash and see whether a lost citation is a model quirk or a content problem. That is genuinely useful, and it is only available to paying subscribers, which limits how much you can generalise.

What to do. Add a model column to your AI visibility tracking sheet. For the ranking side of this story, read our SEO news for September 2026 and the SEO in 2026 guide on AI Overviews and AI Mode.

EU AI Act labelling rules are now live for AI-generated ad content

2 August 2026 · Source: European Commission, Cooley

What happened. Article 50 of the EU AI Act took effect on 2 August 2026. The Commission’s announcement says AI-generated or manipulated content must be “clearly and visibly labelled” and carry machine-readable marks, and that people must be told when they are talking to an AI system rather than a person. Fines run to EUR 15 million or 3% of worldwide turnover. Cooley notes a narrow transition: providers of generative systems already on the market have until 2 December 2026 for the marking and detection duties, and AI-generated text escapes labelling only when a person takes editorial responsibility after substantive review.

Why it matters. The deep fake definition is wide. It covers AI images, audio or video that resemble real people, places, objects or events and would look authentic. A lifestyle shot of a real product in a generated kitchen can qualify. This applies to deployers, which means brands and agencies, not only the model vendors. India’s ASCI has a draft in the same direction and the UK ASA applies its existing code to AI ads, so a single labelling standard across markets is the sensible default.

What to do. Inventory every live EU asset made with generative tools, add a visible label, and confirm your creative tools preserve C2PA or equivalent metadata. Chatbots on EU sites need an “you are talking to AI” line. See the disclosure section of the influencer and creator marketing guide for the overlapping endorsement rules.

Snap and Microsoft open ad accounts to AI agents

3 August and 27 August 2026 · Source: Snapchat for Business, Microsoft Advertising

What happened. On 3 August, Snap launched an official Ads MCP server connecting the Snap Ads API to Claude, ChatGPT and Gemini. It is read-only at launch, an organisation admin must approve each agent, and write access is promised later with per-agent controls. On 27 August, Microsoft Advertising made AI Max for Search available to all advertisers and switched it on by default for new Search campaigns, citing 101% higher engagement for ads inside Copilot than traditional search ads and a 53% lift in purchases within 30 minutes. Its 31 August product notes add a read-only Microsoft Advertising MCP server for Microsoft 365 Copilot, ChatGPT and Claude, with Stagwell saying audits dropped from hours to minutes.

Why it matters. Meta shipped its connectors in April. Now three of the big ad platforms let an outside model read your account. The pattern is identical everywhere: read first, write later, admin approval per agent. This is the moment agencies should write an AI access policy, because within a quarter someone on the team will connect a personal ChatGPT to a client account.

What to do. Connect one read-only agent to one account and use it for the weekly performance narrative. Turn AI Max off on Microsoft campaigns where you need tight keyword control, since Search Engine Journal confirms existing predictive-matching campaigns get it enabled automatically. The social media marketing guide covers Snap’s ad products.

Meta pulls placement controls, TikTok extends AI video to 30 seconds

20 August 2026 · Source: Social Media Today, ppc.land

What happened. Around 19 August, advertisers began seeing in-product notices that excluding placements, platforms, devices and operating systems will no longer be available at ad set level, first surfaced by Jon Loomer and reported by Social Media Today on 20 August. Meta has not published an announcement or a timeline; it is currently a test on Sales and Leads objectives. The remaining levers are value rules and account-level placement settings.

In the same month TikTok rolled Dreamina Seedance 2.5 into Symphony for select paid advertisers, per ppc.land: generated video up to 30 seconds (from 15), up to 50 reference files (from 9) and timestamp-level creative direction inside Ads Manager.

Why it matters. Meta is removing the one control almost every buyer uses. No more Instagram-only ad sets from the exclusion menu. That is consistent with the fully automated direction Meta has signalled for 2026, and it makes creative the only lever left that you fully own. TikTok’s move is the mirror image: the platform will make the creative too, if you let it.

What to do. Learn value rules now, before the test becomes the default. And keep humans in the video loop; LinkedIn’s “seems like AI slop” feedback button and Snap’s decision to limit fully AI-generated video in Spotlight show platforms are policing low-effort output even as they sell the tools. Our social media news for September 2026 has the full platform list.

Brainlabs data: AI referrals up 163%, organic down 10.5%

25 August 2026 · Source: Digiday

What happened. Digiday published Brainlabs analytics covering 54 clients (24 US, 24 UK, 6 elsewhere) from January 2025 to April 2026. Organic sessions fell 10.5%, from 140.1 million to 125.4 million, and 46 of 54 clients lost traffic. Referrals from ChatGPT, Copilot, Gemini and Perplexity rose 163% to roughly 200,000 sessions a month, key events from those visitors rose 335%, and AI-referred visitors completed key events at 1.5 times the organic rate.

“We need to start taking this channel seriously. It’s definitely a channel you should be including in your everyday measurement KPIs.” Sam Paez, senior SEO manager, Brainlabs, via Digiday.

Why it matters. 200,000 AI sessions against 125 million organic is 0.16%. Tiny. But the conversion multiple is real, and the organic decline is the one that hits budgets. This is the first agency dataset of this size to put both numbers side by side, and it supports a boring conclusion: AI referrals are a KPI, not a strategy, until they clear 2 to 3% of sessions.

What to do. Build a GA4 channel group for AI assistants this week and report it next to organic. The marketing analytics guide shows the regex, and the content and publishing news covers how publishers are responding.

The month at a glance

Eight dated developments, one theme. Every vendor is moving from answering questions to taking actions, and every platform is deciding how much of your account an outside agent may touch.

What to watch for the rest of September

Three things are scheduled or half-confirmed, and two things did not happen. HubSpot’s renamed UNBOUND conference runs in Boston from 16 to 18 September and usually carries the year’s biggest Breeze agent announcements. Salesforce says the Claude plugin’s open beta lands this month. And OpenAI has told partners that GPT-6 Astra reaches paid ChatGPT plans and the API within days of 3 September.

On the negative side: there are still no ads in the Gemini app. Google’s ads VP Dan Taylor publicly rejected a report claiming otherwise earlier this year, per Search Engine Journal, and nothing in August or September changed that. And no major adoption survey with August or September fieldwork has been published as of 7 September; the most recent large one is still Gartner’s May CMO Spend Survey, which we cover in the AI marketing guide. Our weekly roundup for the week of 7 September tracks all of this across channels.

Last researched and updated: 7 September 2026.