September in Barcelona always feels like a reset. The city filled back up, the terraces stopped being a tourist-only sport, and we got that run of warm, slightly heavy days where the sea is still swimmable but the light has already turned autumn. Good month for sitting down and thinking about Q4, which is exactly what everyone around me has been doing.
Because here is what actually happened last month, underneath the product announcements: the industry started attaching prices to AI. Google began paying publishers when their content shapes an AI answer. Cloudflare switched on a payment layer so sites can charge AI agents per request. Amazon put its advertisers inside ChatGPT. Meta started charging brands a monthly fee for professional tools and an allowance of AI customer service. IAB Tech Lab published the first standard for agents negotiating media deals with each other. And while all of that was being metered, Anthropic and OpenAI both cut what their frontier models cost to run. We have spent two years talking about AI as capability. September was the month it got a price list.
That matters more than another feature roundup, because prices change behaviour in a way features never do. If access to your content has a number on it, publishing becomes a commercial decision. If an agent can read your catalogue but not your brand story, you find out which one you were really investing in. Most of what follows has a number attached, and several of those numbers land this month.
🔍 Search Marketing Updates
Amazon advertisers can now buy ads inside ChatGPT
Amazon struck a partnership with OpenAI that lets its advertisers run ads inside ChatGPT, starting with selected brands in the United States. The largest retail ad network in the world is now plugged into the biggest consumer AI assistant, roughly 200 days after OpenAI first opened ChatGPT to advertising.
💡 Why it matters: As I wrote last month, ChatGPT Ads had just opened across 31 European markets and the open question was who would actually spend there. Amazon answering that question changes the character of the channel completely: this is no longer an experiment for AI-curious brands, it is retail media’s next shelf. Amazon also brings something OpenAI could not build quickly: purchase data at scale.
🚀 Takeaway: If you sell on Amazon, ask your account team when the ChatGPT placement opens beyond the initial US cohort and put a small test budget aside now, before the CPMs get discovered. If you do not sell on Amazon, treat this as your signal that paid presence inside AI answers is becoming a normal line item, not a 2028 problem.
Source: CNBC
Google started paying publishers when their content shapes an AI answer
Google began testing AI Contribution, a Search Console pilot that pays publishers when their content contributes significantly to an answer in the Gemini app, AI Overviews or AI Mode. Payment attaches to the generation phase, when your content shapes what the answer says, not to being linked afterwards, and publishers can opt out whenever they like.
💡 Why it matters: As I covered last month, Google had just rolled out AI performance reporting to every site owner so you could finally see your AI visibility. Paying for it is the much bigger move, because it concedes that content feeding an answer has value even when nobody clicks. The caveat is real: Google has not explained the calculation, one participant called it quite black box, and for several small and midsize sites the money is coming in under 0.1% of their ad revenue.
🚀 Takeaway: If you publish content at any volume, check whether an AI contributions panel has appeared in your Search Console and read the terms before accepting. Either way, start reporting AI-answer contribution as a separate content KPI this quarter, because it is now a thing with a market price.
Source: Search Engine Journal
Microsoft put a number on the first AI-assisted holiday season
Microsoft Advertising published research finding that 76% of US shoppers, 74% in Australia and 70% in the UK plan to use AI to plan their seasonal spending. It also measured the average holiday conversion journey at 52 days, with 56% of November conversions and 35% of December conversions beginning in October, and 63% of US shoppers starting before Halloween.
💡 Why it matters: Most teams still budget as though peak season starts with Black Friday. If more than half of November’s conversions begin in October, the discovery spend you are holding back for the last week is arriving after the decision was made. The AI figure is the part I would not wave away either, because a shopper who starts in an assistant never sees your category page or your paid search ad.
🚀 Takeaway: Pull part of your November prospecting budget into the first three weeks of October and judge it on assisted conversions rather than last-click, then make sure your product feed answers the comparison questions an assistant would ask (sizing, materials, returns, delivery windows) rather than just listing attributes.
Source: Microsoft Advertising
Google’s September spam update is the slowest one yet
Google began rolling out the September 2026 spam update on 24 September at 9:15 a.m. PDT, globally and across all languages, and said the rollout could take up to two weeks. It is the fourth confirmed spam update of the year, and by some distance the longest.
💡 Why it matters: As I flagged last month, August’s update hit harder than its predecessors despite finishing in under three days. This one is budgeted for two weeks, which means the ranking movement you are seeing right now is probably still the update and not something you broke. Google has not said what it targets, so resist the urge to diagnose from a single week of data. Worth noting too: there was no September core update, whatever a handful of SEO blogs told you.
🚀 Takeaway: Freeze any significant site or content changes until the rollout is confirmed complete, so you can actually attribute what moved. Then compare landing pages that lost visibility against Google’s spam policies line by line rather than rewriting on instinct.
Source: Google Search Status Dashboard
⚙️ Marketing Technology Updates
Google pushed Data Manager into Analytics and DV360, and shipped geo-experiments to everyone
Google released a measurement update built on three ideas: stronger first-party data foundations, more signals, and causal proof. Data Manager is now integrated directly into Google Analytics and Display and Video 360, a new Data Strength Uplift metric quantifies the extra conversions your first-party setup recovers, and Meridian GeoX went generally available worldwide.
💡 Why it matters: Google says advertisers who connect offline and app data to Data Manager see an average 26% lift in incremental ROAS, and that the tag gateway alone delivers around 14% conversion uplift, over 20% on Demand Gen. Those are vendor numbers, so discount them, but the direction is not in doubt. The more interesting release is Meridian GeoX, which puts proper geo-based incrementality testing within reach of teams who do not employ a data scientist.
🚀 Takeaway: Run one geo holdout this quarter on your single largest always-on channel, ideally branded search or Advantage+, and find out what it is actually contributing before you set 2027 budgets on last-click.
Source: Google Ads & Commerce Blog
Amazon merged its DSP and ads console into one conversational platform
At unBoxed 2026 Amazon collapsed the Amazon DSP and the Ads Console into a single surface called Amazon Ads Agent, retiring the DSP name in favour of DVA for display, video and audio. New Full-Funnel Campaigns hand channel mix, creative, audience and placement to Amazon’s models, and the company says beta advertisers saw 67% higher long-term ROAS.
💡 Why it matters: Amazon rebuilding its buying surface in the last week of September, days before Prime Big Deal Days, is bold timing. The deeper shift is the same one Google and Meta already made: the console stops being a set of controls and becomes a conversation, with the model choosing. That is a real efficiency gain and a real loss of diagnostic visibility, and the 67% figure is Amazon’s own.
🚀 Takeaway: Do not let Full-Funnel Campaigns take your whole peak-season budget on a 67% claim. Run it against a manually structured campaign on matched budgets through October so you have your own number before November.
Source: Amazon Ads
Salesforce gave its AI agents names and job descriptions
Salesforce launched a portfolio of named, job-ready Agentforce agents spanning service, IT and HR, commerce, outbound sales, inbound pipeline generation and supply chain. Most are generally available now, with Hunter, the outbound sales agent, in pilot ahead of GA in November 2026.
💡 Why it matters: The naming is marketing, but the capability underneath is not. Salesforce calls it long-horizon runtime, built on memory, durable execution and dynamic steering, which in plain terms means an agent can hold a goal for days and weeks instead of a single conversation. That turns agent adoption into an org-chart question rather than a tooling question, and most marketing teams have not had that conversation yet.
🚀 Takeaway: Before you pilot anything, write down one process your team owns end to end, with its inputs, its decision points and the person accountable for the outcome. If you cannot describe it that precisely, an agent with a two-week memory will not improve it, it will just scale the confusion.
Source: Salesforce Newsroom
IAB Tech Lab published the first standard for agents negotiating media deals
IAB Tech Lab released AAMP 3.0 with OpenProposal, a specification for how sellers describe their advertising products and respond to buyer briefs in a machine-readable format. It targets the RFP-to-buy stage, the last genuinely manual link in programmatic, and it is in public comment until 22 October.
💡 Why it matters: Agents already handle discovery and evaluation. Standardising the proposal stage means a buying agent can compare far more publisher offerings than a human planner ever could, which sounds like pure upside until you realise the flip side: anything not expressed in a machine-readable format simply does not enter the comparison. Being unlisted is the new being unranked.
🚀 Takeaway: If you sell any inventory, sponsorship or partnership placements, read OpenProposal during the comment window and ask whoever owns your rate card what a machine-readable version of it would look like.
Source: IAB Tech Lab
Meta shipped an MCP server so an AI agent can set up WhatsApp Business for you
Meta released the WhatsApp Business Tools MCP, letting an AI coding agent such as Claude, Cursor, Codex or ChatGPT handle WhatsApp Business setup end to end: creating the account, verifying the number, registering the Cloud API, writing and editing message templates, running test messages and webhooks, and watching compliance.
💡 Why it matters: This one is quietly important if you sell anywhere in Spain or Latin America, where WhatsApp is not a channel but the channel. The reason most mid-sized brands never launched properly was never strategy, it was the four-tool setup slog across Developer Console, Business Manager, API reference and editor. That excuse is gone, and the brands that move first get to learn the channel before their category crowds into it.
🚀 Takeaway: Pick your two highest-intent lifecycle messages (abandoned cart and back-in-stock are the obvious pair) and have someone stand up a WhatsApp template flow for them this month rather than adding it to the 2027 roadmap.
Source: TechCrunch
📱 Social Media Marketing Updates
Meta turned its apps into a subscription business, brands included
Meta launched Meta One across Facebook, Instagram, WhatsApp and Meta AI, with individual plans from $2.99 a month and creator and business tiers at Essential $14.99, Advanced $49.99, Expert $149 and Max $499 a month. The business plans bundle enhanced profiles, professional publishing, exportable analytics, team access and, crucially, an allowance of Meta Business Agent, the AI that answers your customers around the clock.
💡 Why it matters: More than 50 features launched with it, globally, and the pricing tells you what Meta thinks each is worth. The real story is the Business Agent allowance, because it turns customer service capacity into a metered line on an invoice. Organic reach has been quietly priced for years through ad spend. This is the first time professional presence on Meta has a sticker price, and I doubt it is the last platform to do it.
🚀 Takeaway: Before subscribing, work out what you currently pay per customer conversation across your helpdesk and agency time, then compare it against the tier that covers your actual message volume. Most brands will find Essential or Advanced is cheap and Max is a procurement conversation.
Source: Meta Newsroom
Pinterest launched Visual Search Ads
At Pinterest Presents, Pinterest introduced Visual Search Ads, a dedicated placement and format in search results and Pin closeups that pairs keyword relevance with visual intent. The format supports lower-funnel objectives, works with Performance+, and enters beta for eligible advertisers across every Pinterest ad market.
💡 Why it matters: Pinterest says users run more than 80 billion searches a month, the majority visual and more than half commercial. That is a search channel most e-commerce teams file under social and staff accordingly. A dedicated search placement makes that intent biddable for the first time, and it arrives with Priority Products, self-serve A/B testing and a campaign optimization score alongside it.
🚀 Takeaway: Pull your Pinterest search terms report and compare the top commercial queries against what you are paying for the same intent on Google Shopping. If the gap is as wide as it usually is, request Visual Search Ads beta access before your category fills up.
Source: Pinterest Newsroom
🔒 GPTs, Emerging Tech & Data & Privacy
OpenAI turned ChatGPT into an app platform with always-on agents
At DevDay 2026 OpenAI opened ChatGPT to full applications rather than plugins, with interactive panels, sidebars and file editors running inside the conversation, more than 4,000 apps available and over 30 partners including Adobe, Figma and Salesforce in a new enterprise marketplace. It also launched Dots, always-on agents with their own cloud computer that keep working between conversations, plus Sign in with ChatGPT so users carry their ChatGPT identity and usage allowance into third-party apps. OpenAI put ChatGPT at 1.2 billion weekly users.
💡 Why it matters: An app surface in front of 1.2 billion weekly users is a distribution channel, and distribution channels develop their own discovery rules, which is what you will be optimising for next. Sign in with ChatGPT is the sneakier one, because an identity layer that follows people into other products is how platforms stop being tools and start being infrastructure. Notably, OpenAI shipped no revenue share, so this is not an app store yet.
🚀 Takeaway: If you run any kind of SaaS, tool or configurator alongside your commerce, scope what the thinnest useful version of it looks like inside ChatGPT. Early presence on a new surface has always been the cheapest acquisition any of us ever bought.
Source: TechCrunch
Cloudflare put a payment layer between AI agents and the open web
Cloudflare moved two products into beta: Monetization Gateway, which lets a site charge AI agents per request to an API, MCP tool, dataset or page, and Pay Per Use, which pays publishers when AI companies report actually using their content rather than merely fetching it. The Gateway runs on the x402 protocol and HTTP 402, settling in USDC, with four implementations live including Ceramic.ai and Stocktwits.
💡 Why it matters: As I covered back in July, Cloudflare had just started pushing AI companies to pay for publishers’ content, and from 15 September its defaults began blocking training and agent crawlers on pages that host ads. Beta payment rails are the logical next step, and the consequence for marketers is strange and new: which AI assistants can cite your brand becomes partly a commercial decision made by your infrastructure provider. Both products are US-only sellers and buyers for now.
🚀 Takeaway: Find out this week which AI crawlers your site currently allows and whether anyone chose that configuration deliberately. If you are blocking training and agent bots by default and you want to be cited in AI answers, you have just made an expensive decision by accident.
Source: Search Engine Journal
The EDPB set one methodology for GDPR fines across Europe
The European Data Protection Board adopted guidelines harmonising how national regulators decide whether to fine, replacing the old WP29 guidance, alongside final guidelines on how the Digital Services Act and the GDPR interact. Regulators now follow a five-step test: confirm a fine is available, identify the liable controller or processor, assess intent or negligence, weigh aggravating and mitigating factors, and check the fine is effective, proportionate and dissuasive. Fourteen worked examples are included and consultation runs to 13 November.
💡 Why it matters: Harmonisation is mostly good news for anyone operating across several EU markets, because it narrows the range of outcomes you have to plan for. It also removes an excuse. Once there is a published methodology with worked examples, “we did not know how this would be assessed” stops being a defence, and the intent-or-negligence step is where a sloppy consent banner or an undocumented tracking pixel starts to cost real money.
🚀 Takeaway: Ask whoever owns your tag manager for a current list of every pixel firing before consent, with an owner’s name next to each one. That single document is what the negligence assessment turns on.
Source: European Data Protection Board
Anthropic and OpenAI both cut frontier prices in the same week
Anthropic released Claude Opus 5.5 at $4 per million input tokens and $20 per million output, 20% below Opus 5, with cache reads 60% cheaper and overall running costs around 40% lower. The same day, OpenAI added GPT-6 Sol and GPT-6 Luna, cutting API prices roughly 50% against GPT-5.6 promotional pricing, with Sol aimed at complex work and Luna at high-volume extraction and summarisation. Anthropic followed with Sonnet 5.5 on 28 September.
💡 Why it matters: As I noted last month, Gemini 3.7 Flash had just arrived at half its predecessor’s price, and the pattern has held every month since. The practical consequence is that the automation business case you rejected in spring on unit economics may now clear the bar, and the cheap high-volume tiers are aimed squarely at the boring work most marketing teams still do by hand: product description generation, feed enrichment, review summarisation, support triage.
🚀 Takeaway: Go back to whatever AI project you shelved this year because the per-unit cost did not work, and re-run the maths at today’s prices on the cheapest capable tier rather than the flagship. A surprising number of those will now pencil out.
Source: TechCrunch
🧰 Tool of the Month: Cloudflare AI Crawl Control
AI Crawl Control answers a question half of this issue raises and almost nobody can currently answer: what is being taken from your site right now. If your site sits behind Cloudflare you already have it, a dashboard showing exactly which AI crawlers are hitting you, how often, which ones respect your robots.txt, and controls to allow or block them by category. Monitoring and blocking are on every plan, while the paid tiers add the ability to return a custom HTTP 402 pointing bots at a licensing contact. The reason I would spend twenty minutes in it this month is Cloudflare’s default change on 15 September, which started blocking training and agent crawlers on ad-bearing pages for new domains. If nobody on your team has looked, your AI visibility strategy may currently be a setting you never chose.
📅 Upcoming Dates
- Early October Google’s September spam update finishes rolling out. Hold your attribution conclusions until it does.
- 6 to 7 October Amazon Prime Big Deal Days, 48 hours across 22 countries including Spain, the UK, Germany and France. The unofficial start of peak season.
- Mid-October Google Ads API blocks creation of shared ads, ahead of the feature being retired entirely in 2027. Google Ads API v22 is also projected to sunset this month.
- 16 October Public comment closes on IAB Tech Lab’s Programmatic Standard Practices.
- 22 October Public comment closes on AAMP 3.0 and the OpenProposal specification for agent-to-agent media buying.
- All month Microsoft’s own data says 56% of November conversions and 35% of December conversions begin in October. This is the build month.
- In person in Europe this month, if you want to actually be in the room:
- 30 September to 2 October Search Central Live Deep Dive Europe, Barcelona. Three days of Google engineers on crawling, indexing, serving and ranking. Free, application only.
- 7 to 8 October World Summit AI, Taets Art & Event Park, Amsterdam. Tenth edition, around 10,000 attendees.
- 8 to 9 October brightonSEO, The Brighton Centre. 190-plus speakers, 4,500 marketers, and MeasureFest is bundled into the same ticket if measurement is your problem.
- 10 October MeasureCamp Berlin at SumUp, Koppenstrasse 8. Free unconference, and the tickets have already gone, so join the waiting list.
- 19 to 21 October PI LIVE Europe, London, across Old Billingsgate and The Steel Yard. The one to pick if affiliate, partnerships, creator commerce or retail media is where your revenue sits.
- 24 October MeasureCamp Spain at Talent Garden, Madrid. Free, Saturday, analytics and CRO. The easiest one to justify on a Spanish budget.
- Worth the flight, in the Americas:
- 5 to 8 October Advertising Week New York, The Penn District. The broadest of the lot, more brand and media than performance.
- 13 to 15 October MAICON, Cleveland Convention Center. The one conference built entirely around AI applied to marketing.
- 15 October eRetail Day Latam, Auditorio BB, Mexico City. Run by eCommerce Institute with AMVO, free with registration, and the best single room for LATAM digital commerce.
- 20 to 23 October ANA Masters of Marketing, Rosen Shingle Creek, Orlando. Enterprise brand side, useful if that is who you sell to.
- Looking ahead: E-SHOW Madrid at IFEMA on 4 to 5 November, Web Summit in Lisbon from 9 to 12 November, Singles’ Day on 11 November, Black Friday on 27 November, Cyber Monday on 30 November, and the EDPB fining-methodology consultation closing on 13 November.
👋 Sign-off
If I had to compress September into one line, it is that the free period ended. Not dramatically, and not everywhere at once, but the pricing of AI attention started in earnest: paid for by Google to publishers, charged by Cloudflare to crawlers, bought by Amazon inside ChatGPT, billed by Meta to brands. When something gets a price it also gets a budget owner, and that is usually the moment a trend becomes a workstream.
So I am curious about one thing in particular this month. Has anyone on your team actually been offered the Google AI Contribution panel, and if so, what did the number look like? I would genuinely like to hear from someone who has seen it rather than read about it. And as always, tell me which of these mattered most on your desk and which one I underweighted.
Good luck with October. It is the month that decides November.
Ata Gür
Co-Founder & CEO, Frizbit
Barcelona
📩 Want to see how you can recover your lost sales at your online business through Marketing Automation? Book a free demo with Frizbit and let’s explore it together.
