Three Deadlines This Week, and the First One Lands in a Fortnight

No big promises this week, three dates instead. That is more useful. Meta changes its commercial terms on October 30, and the new wording lets it collect the content of your website. OpenAI set January 14 as the end of GPT-4o and GPT-4.1 in its programming interface, the door your automations use to call a model. And Google launched a shopping cart in the US that lives inside its search and its assistant, not on your store.

Two of these three ask you for an inventory before December. The third just asks you to look at where the purchase now gets decided. None of them asks for budget.

Meta Can Collect Your Website Content From October 30

Meta is updating its commercial terms, the contract you accept by running a business account on Facebook, Instagram or Ads Manager. They take effect on October 30. Lawyers who read the new wording, speaking to Nieman Lab, see in it permission for Meta to send crawlers to collect content from sites associated with a business account, then to use that content to train its products. And the contract defines products to include Meta's AI. There is no way to decline that clause and keep the account: refusing the terms means stopping business use of the platform. Meta did not answer the detailed questions put to it, and says the wording already existed, spread across several documents.

Why this matters to you. If you buy advertising on Facebook or Instagram, you run a business account, so the clause covers you whether or not you publish articles. What leaves is your site: product pages, case studies, method pages, everything your team wrote to set you apart from your competitors. You cannot keep the advertising on one side and the content on the other either, it is one contract. For a company whose content is the product, a publisher, a consultancy, the question stops being theoretical on October 30.

What I would do. I would have your legal counsel read the clause this week, not in November. I would not leave Meta over it: if the platform drives your acquisition, leaving costs far more than being trained on. I would look hard at what is actually exposed, though, and move behind a form anything valuable that has no reason to be public: a white paper, a price grid. As for blocking it through robots.txt, the file that tells crawlers where they are allowed to go, some lawyers think the contract could override it: no court has ruled, so do not treat it as your only protection.

OpenAI Pulls GPT-4o on January 14, With Three Months of Notice

On October 14, OpenAI posted a shutdown notice on its deprecations page. Five snapshots of the GPT-4 family, GPT-4o and GPT-4.1 among them, leave its programming interface on January 14, 2027. The notice covers automated calls only, not the ChatGPT app itself. The recommended replacement is the small model of the next generation, and fine-tuned versions stop on the same date. One detail caught developers' attention: OpenAI's own policy promises at least six months of notice for a generally available model, and this one grants three.

Why this matters to you. Any automation that names one of these models explicitly stops working that day, and this is not a developer problem. Scenarios built in Make or n8n, automatic summaries of sales calls, qualification of inbound forms, draft replies in support: marketing teams have been building these for two years without telling anyone. You probably know neither how many there are, nor who built them, nor which ones still run.

What I would do. The inventory this week, not in December: one table listing every automation that calls a model, its owner, and what breaks when it stops. That is half a day of work, and it will serve you again at the next announcement of this kind. Then test the replacement on your own cases: this is not a name to change in a parameter, the new model reasons differently and handles tool calls differently, so compare outputs on thirty real examples. At Skello, what saved us on migrations like this was never speed, it was knowing exactly which automation touched what.

Google Puts the Shopping Cart in Its Search, Not on Your Store

On October 15, Google launched Universal Cart in the US, a single shopping cart that lives inside Google Search, AI Mode included, and inside the Gemini app. Shoppers drop in products from different merchants, come back to them later, and get an alert when a price drops or an item is back in stock. They either pay directly with Google Pay or finish on the merchant's own site. The merchant stays the seller of record: order, confirmation and after-sales support stay with them. Nike, Sephora, Target, Walmart, Wayfair and a share of Shopify merchants are in at launch, and YouTube and Gmail will follow with no date announced.

Why this matters to you. Product discovery already ran through Google. Now the add-to-cart does too, and it happens without the shopper opening your site. What represents you in that cart is no longer your page, it is your product feed: title, price, availability, images, category. If that data is approximate, you are not eligible, and you will find out by watching competitors show up there. The other consequence is less visible: part of the journey escapes you, so the abandoned-cart email and your own cross-sell never fire.

What I would do. If you do not sell in the US, there is nothing to buy and nothing to integrate, and I would not start a project over it yet. I would still have the product feed audited, because it is exactly the work that already serves your Shopping campaigns and your organic search: full titles, accurate stock, clean images, attributes filled in. At Le Dressing de Paolina, the 38% conversion lift did not come from a brilliant idea, it came from product pages that were finally correct. The day this cart reaches Europe, that work will already be done, or it will not.

Everything Else, in Three Lines

Google opened its most powerful model, Gemini 4 Argon, to paying subscribers and developers, with a launch rate that doubles on December 1, so do not build a budget on the price shown today. Claude's editable documents and decks, the ones I wrote about in September when they were still in beta, have been on by default since October 15, which is worth a look at the setting if you administer an enterprise account. And the European AI Office sent its first information requests to OpenAI and Anthropic under the EU AI Act, which are questions rather than penalties, with nothing for you to do this week.

See You Next Friday

Three dates, and none of them asks for money. October 30 calls for a legal reading, January 14 for an inventory, the American cart for a clean product feed. Three pieces of plumbing, which is exactly why they get pushed back until the day they turn urgent.

If one of the three speaks to you directly, the Meta clause, the inventory of your automations or the state of your product feed, let's talk. See you next Friday.