ChatGPT started showing brands their own mentions and clicks. ChatGPT ads got a local invoice in 16 CEE markets, though not in Kosovo or Albania. Synthetic customers got cheaper, and Brussels wants agencies to answer for their influencers. AI marketing is moving from guessing to measuring, and the agencies that measure first set the benchmarks.
Happy Friday, friends.
Yesterday Every's Mike Taylor published a piece on what comes after running teams of AI agents. His answer is simulation: build a model of the market your agents work in, and test strategies there before reality does. The part that stayed with me was a miss. He cloned nearly 2,000 Hacker News commenters as AI personas, and they picked which headlines would reach the front page 60% of the time. When he checked the misses, the personas weren't the problem. One extra early upvote sent identical posts down completely different paths.
That's marketing, honestly. Taylor points to a 2006 study from Columbia and the Santa Fe Institute: 14,341 people split into eight separate worlds heard the same 48 songs, and a song could be a hit in one world and a flop in the next. The best songs rarely flopped. Almost anything in between was possible.
I think this week agencies got their first real tools for telling skill from luck. ChatGPT now shows a brand how often it gets mentioned and clicked. Agencies in 16 CEE markets can buy ChatGPT ads on a local invoice, and our two markets aren't among them. Synthetic customers got cheaper and better funded, which is exactly when you need to know what they can't tell you. And Brussels wants agencies answerable for the influencers they hire.
Here's my tension. I like the idea of rehearsing a launch on synthetic customers. I don't trust a panel that has only ever met your current ones. Taylor says it himself: run a bad simulation 50 times and you get a wrong answer you feel sure about.
Agency owners, start with the ChatGPT piece. Send the Brussels one to whoever writes your influencer contracts.
ChatGPT now shows brands what it says about them
On 10 November OpenAI opened Business Hub in beta. A company can claim its profile in ChatGPT, correct what shoppers see, upload a product feed with prices and stock or connect Shopify, and see 28 days of aggregate data: website clicks, mentions and the topics that keep coming up. It's free at launch. Profile claims are open wherever ChatGPT operates and OpenAI supports verification; the Business Agent that answers product questions is US-only for now.
AI visibility just became a line in the monthly client report, and the agency that doesn't own that line will lose it to one that does. The ground under it moves fast. Berlin's Peec AI published data on 11 November showing that after ChatGPT switched to GPT-5.6 on 8 August, product pages rose from 13.8% to 20.1% of its citations while "best X" listicles fell from 16.2% to 7.3%. In Peec's survey of 700 marketers, 89% raised AI-search spend and only 24% can prove a return. Peec sells visibility tracking, so it gains from the alarm, and OpenAI only shows numbers above a minimum threshold, so small brands may see a blank dashboard. Still, a first-party number from OpenAI beats every estimate agencies sold this year.
This week, claim and verify the profiles of your three biggest clients and save the first 28-day reading as a baseline. Then move part of their AI-search budget from listicle placements to their own product pages.
ChatGPT ads now come with a local invoice in 16 CEE markets, not in Kosovo or Albania.
On 12 November OpenAI named Aleph its exclusive reseller for ChatGPT ads in 44 markets. In Central and Eastern Europe that covers Croatia, Serbia, Slovenia, Bulgaria, Romania, Hungary, Czechia, Slovakia and the three Baltic states. Advertisers there sign locally, get invoiced locally and pay in local currency, and Aleph "extends credit", says founder Gastón Taratuta.
The obstacle to a ChatGPT test budget in this region was never the idea. It was the invoice. A Zagreb or Belgrade agency can now put ChatGPT spend on the same kind of bill as Meta and Google, with credit terms, through one partner. The money is there: Netokracija reports IAB Croatia's estimate of €190.9 million in Croatian online ad spend for 2025, up almost 23%, with €202.8 million forecast for 2026. Albania, Kosovo, North Macedonia, Montenegro and Bosnia aren't on the list. The fair objection is that ChatGPT ads are new, nobody has regional benchmarks, and the first campaigns will be expensive lessons. True, which is why the agencies that run the first small tests will own the benchmarks everyone else quotes.
If you're in one of the 16 markets, pitch one client a capped ChatGPT test this quarter with one question it has to answer. If you're in Prishtina or Tirana, ask Aleph now how a client of yours could buy, before your clients ask you.
Synthetic customers got cheaper
On 11 November Plurai raised a $13 million seed round led by Team8 and opened its platform to everyone, with a free tier. It builds a replica of a company's AI agent from its knowledge base, tools and policies, then runs it against realistic and hostile simulated users to catch made-up answers, broken tool calls and policy breaches before real customers do. The same day the ANA, the US advertisers' association, published The Synthetic Respondent, its analysis of AI-generated panels for ad research.
Testing on synthetic people is becoming a normal step, and the agencies selling it will have to sell its limits too. Mark Himmelsbach, co-founder of the agency Episode Four, argued in Marketing Dive that a panel needs "evidence about the people they're supposed to represent", so it's good for testing on existing customers and weak for reaching new ones. Constant scoring, he wrote, makes a team "more conservative and risk-averse". Agent teams have their own failure mode: in Anthropic researchers' Worse Together study, one coordinator agent beat a group of per-user agents in all four test settings.
Before you sell a client synthetic testing, write down what data the panel is built on and run one small real test next to it. Keep a slice of budget for an idea the panel scored low. If you build chatbots for clients, run the next one past simulated users before launch.
Brussels wants agencies to answer for their influencers
On 11 November the Commission proposed the Digital Fairness Act, a Regulation that would replace the 2005 Unfair Commercial Practices Directive with one rulebook across the EU. Influencer content must disclose its commercial intent "clearly and prominently", and the text treats "advertisement" and "commercial collaboration" as clear enough. A brand using influencers must ensure compliance "by contract or other appropriate means". A Commission official told Agence Europe that "brands and advertising agencies" must do their part.
Influencer compliance is moving from the creator's problem to the contract you signed. The same proposal bans dark patterns, requires a reminder before a free trial converts and an online way to cancel, and sets national fines of at least 4% of turnover in the countries concerned, according to Freshfields. This reaches agencies in Prishtina and Tirana too, because the rules follow the consumer, not the agency's address. It's a proposal: Parliament and the Council still have to agree it, and it would apply 12 months after entry into force, so nothing changes this year. But client legal teams rewrite templates long before a law bites.
Pull your current influencer contracts and add three things: the exact disclosure label, an approval step before posting, and a record of every post. Then look at your gifting and seeding programmes, which usually run on a DM and no contract at all.
Short Signals
Creative: Adobe is letting ChatGPT, Claude and Gemini drive Creative Cloud. At MAX on 10 November Adobe announced a local MCP server that exposes features in Photoshop, Illustrator, After Effects and Premiere to outside assistants, working on editable files. No pricing or release dates yet. Pick the one deliverable your studio repeats every week, such as resizes or cut-downs, and plan to script it first.
Testing: Google's Decisions API grades options instead of writing text. The Gemini Decisions API answers up to 1,000 yes/no, multiple-choice or rubric questions in one call, each with a confidence score, at $0.10 per million input tokens with output free. Google pitches it for routing and policy checks; we'd use it to score 200 headline variants against a client's brand rules before a human reads the top ten.
Production: Ogilvy says AI can cut its cost per asset by up to 75%. With CAA, Ogilvy launched ECHO, which turns one 8 to 10 asset shoot into 30 to 50 assets from approved talent likenesses, rolling out across WPP this month. The 65 to 75% saving is Ogilvy's own figure. Expect clients to quote it at you in the next production pitch.
Payments: Kosovo's instant payments start on 23 November. The central bank set the launch of TIPS-K, with transfers between accounts in seconds, around the clock, and every bank connected by 23 February 2027. Fees aren't public yet. If you run an online shop or build checkouts in Kosovo, ask your bank when it joins.
If you run an agency, hit reply and tell me whether a client has asked you yet what ChatGPT says about them.
Next edition soon,
Çelik