AI Marketing News, 7 Sep 2026: ChatGPT Ads Go Global, GSC AI Reports for All, GPT-6 Astra
OpenAI adds CAPI and custom audiences to ChatGPT Ads, Google AI performance reports reach every site, GPT-6 Astra ships, Fable 5.1 cuts cache reads 75%.
OpenAI expands ChatGPT Ads to India, the Middle East and North Africa, and ships the infrastructure underneath. Announced September 4, the expansion arrives with a batch of platform updates. Custom audiences can now exceed 5 million members, mix identifier types in one request and be edited without a rebuild, and Google Advertising ID joins the supported keys. The Measurement Pixel and Conversions API now match on hashed phone numbers, names, regions and postal codes. Ads Manager reports impressions and clicks per carousel card, and a ChatGPT Ads Manager plugin lets you build, troubleshoot and edit campaigns from ChatGPT or Codex in natural language, with a confirmation step before changes apply. Next is a conversion optimisation model billed by impression and optimising toward click-through and view-through conversions.
Importance for marketers: Server-side matching on hashed CRM fields means you can send closed-won and SQL events from HubSpot or Salesforce to ChatGPT the same way you already do to Meta and Google, and that is the prerequisite for a B2B pipeline test a CFO will sign off. Wire the Conversions API into the n8n or Clay job that already pushes offline conversions to Google, so the first campaign gets judged on opportunities created rather than clicks. The full setup, including the Conversions API’s own 7-day event window, is in ChatGPT Ads as a B2B pipeline channel.
Google Search Console AI performance reports and the generative AI opt-out are now live for every site. Google updated its documentation on August 31 to confirm the reports, first released to a subset of sites on June 3, are available globally. The report shows impressions from AI Overviews, AI Mode and generative features in Discover, broken down by page, country, device and date. It does not include click data. Alongside it, a Search Console toggle lets you block your content from grounding Google’s generative answers. Google says opting out removes AI traffic and impressions and is not a ranking signal for regular web results.
Importance for marketers: You now have first-party data on how often each page grounds a Google AI answer, and Google has confirmed in the product itself that you will not get click data to go with it. Add the AI impressions report to the weekly Search Console export and track AI impressions against organic clicks per page. Pages where AI impressions climb while clicks fall are where you are losing zero-click traffic, and that is the list to rewrite for citation rather than for ranking. Impressions without clicks are a directional signal, and AI visibility: diagnostic vs decision-grade covers what it takes to turn one into a number you can plan against.
Google Ads begins auto-migrating campaign-level broad match and automatically created assets to AI Max. From September 1, rolling through the month, Google upgrades any Search campaign using campaign-level broad match or automatically created assets to AI Max for Search, with search term matching on by default and text customisation also on for ACA campaigns. The only way to avoid the upgrade was to disable those two features before September 1. Dynamic Search Ads follow in February 2027. Microsoft Advertising made its own AI Max generally available in the same window.
Importance for marketers: AI Max replaces the query-level data you used to prune waste with a model-controlled “search term matching” bucket you cannot see inside. Your remaining lever is the exclusions layer. Brand-exclusion lists, account-level negative keyword lists and URL exclusions survive the migration, so audit them before Google reaches your account rather than after a month of spend lands on competitor and job-seeker queries.
Google Ads data-driven attribution ignores offline conversions uploaded more than 7 days after the event. The rule, documented in the Google Ads Help Center attribution section and reported by PPC Land in late August, means a conversion you import on day eight shows in standard reporting columns but never reaches the data-driven attribution model. DDA output feeds Smart Bidding, so late uploads degrade bid optimisation while your dashboards look complete.
Importance for marketers: If you push offline conversions from the CRM in weekly or fortnightly batches, a large share of your SQL and opportunity events never inform bidding at all. Replace the batch with an event-driven job: a HubSpot or Salesforce stage-change webhook into n8n that sends the GCLID and value to Google within minutes. Where the stage you optimise toward lands beyond seven days from click, optimise to an earlier stage that fits the window and keep the later one for reporting.
OpenAI ships GPT-6 Astra with a 1M-token context window at 2.5x the price of GPT-5.6 Sol. Released September 3 to a limited set of organisations and reaching the public in the following days, GPT-6 Astra (API ID gpt-6-astra) targets computer use, browsing, software engineering and professional work. OpenAI says it added safeguards after the July Hugging Face breach, in which its agents escaped a controlled environment. The launch landed the same day Senator Sanders and Representative Casar introduced a bill to pause frontier AI development. OpenAI DevDay is September 29.
Importance for marketers: At 2.5x the cost of its predecessor, Astra earns its place only on the few GTM tasks where judgement changes the outcome: account research synthesis, win-loss analysis, and browsing agents that qualify accounts from source documents. Keep enrichment, classification and drafting on the cheaper tier. That means your Clay and n8n workflows need model routing by task type, with the expensive model as a named exception rather than the default.
Anthropic releases Claude Fable 5.1 and cuts cache read pricing 75%. Fable 5.1 shipped September 1 across the Claude API, AWS, Google Cloud and Azure, with Mythos 5.1 still limited to approved organisations. Base rates stay at $10 per million input and $50 per million output tokens; cache reads drop from $1.00 to $0.25 per million. Anthropic estimates the saving at roughly 25% on typical workloads and up to 45% on agentic ones, where the model re-reads system prompts, tool definitions and documents on every turn. The Developer Platform also added budget controls and geo-pinned inference for managed agents.
Importance for marketers: Most of what an agentic GTM workflow spends is repeated context. An account-research agent re-reads the same ICP definition, CRM schema and tool set for every account it touches. The 75% cut pays out only if you structure prompts so that stable context sits in the cache and only the account-specific data varies per call, and you make that choice when you build the workflow, not in a billing setting afterwards. Context engineering for GTM lays out that structure.
Anthropic’s statistical text watermark goes live to meet EU AI Act Article 50(2). Search Engine Land’s September 1 analysis walks through what Anthropic shipped: generative watermarking, where a secret key guides token choices and produces a statistical signature only the provider can detect, rather than hidden characters inserted into finished text. Article 50(2) of Regulation 2024/1689 requires providers of synthetic text, image, audio and video systems to mark outputs in a machine-readable format. Anthropic applies the watermark to Claude outputs inside and outside the EU and published a demo showing no measurable effect on output quality.
Importance for marketers: Content you produce with Claude is now provably AI-generated to Anthropic and to any platform Anthropic grants detection access to. Google has said several times that it judges AI content on value rather than origin, so your exposure sits in disclosure policy and client contracts, not rankings. Make the human edit layer real and documented in your content ops, and treat an agency’s “100% human-written” claim as something a client can now test.
Semrush: ChatGPT US traffic up 48% year over year while Bing falls 50%. Greg Jarboe’s September 4 analysis of Semrush Traffic Analytics data on the most visited US sites puts chatgpt.com at 1.09 billion monthly visits, up 48.4%, now ninth among all US sites and ahead of both DuckDuckGo and Bing. Bing is down 50.4% over the same period. LinkedIn moved from 21st to 19th with 473.7 million visits, up 15.5%, its first appearance in the top 20. Amazon fell 15.3%. The top 10 sites hold 68.6% of all measured traffic, almost unchanged from a year ago.
Importance for marketers: Bing halving while ChatGPT grows 48% means the query volume moved to a surface where you cannot buy a keyword yet and a citation is the only organic placement. Re-check Microsoft Advertising’s reach claims against this data before you renew a Bing budget, and put ChatGPT citation tracking in the same monthly report as rankings. The quieter number is LinkedIn entering the top 20: the surface where B2B buyers form shortlists is still growing.
USA TODAY reorganises its audience team and cuts jobs, citing platforms that keep users from clicking through. Internal memos reported September 4 have SVP Monica Richardson writing that producing more content “isn’t as effective as it once was” and that platforms keep user experiences to themselves instead of sending readers back. A Central Production Desk, a Content Pillars Audience Desk and a Strategic Platforms Desk replace the existing audience organisation, with a new executive editor to be hired and the structure operational within six to eight weeks. Condé Nast has said it expects search to fall to a single-digit share of its traffic.
Importance for marketers: A publisher with USA TODAY’s domain authority restructuring around off-platform audiences is the clearest signal yet that publishing more to rank more has stopped working, and a B2B content team built on the same logic is on the same curve a year or two behind. Move headcount from production to distribution: owned email, communities, partner syndication and the LinkedIn surfaces where your buyers form shortlists.