Picture this: it's a dark Sunday evening in late November. One of your customers is on the sofa, wondering what to buy their dad for Christmas. They don't type anything into Google, and they don't scroll Instagram. They ask an AI assistant, and it gives them three suggestions in seconds. Are any of them yours?
That question sits under almost everything in this week's news. Instacart has brought in Shrek to sell its AI helper, and Microsoft has published a checklist for getting ready for AI-driven Christmas shopping. Here's what happened, and what it means for your online shop.
Shrek, Scooby-Doo and an AI That Fills Your Basket
Instacart, the big US grocery delivery platform, has launched a major campaign for Clementine, its AI shopping assistant. The ads borrow scenes from Shrek, Scooby-Doo, The Office and A Charlie Brown Thanksgiving. In each one, someone sees food on screen, wants to make it and asks Clementine for a recipe.
Clementine suggests meals, fills baskets, finds promotions, recommends cheaper alternatives and handles repeat orders. It draws on 1.6 billion lifetime orders. Instacart says it wants using the assistant to become a habit.
Why should a UK shop owner care about an American grocery ad? Because of that word: habit. When a company pays for prime-time TV to get people used to letting an AI build their basket, it tells you where shopping behaviour is heading. Notice, too, that brands like Hormel and Hillshire appear in the ads. They've made sure the assistant is talking about them.
The lesson for you is simple. More and more buying decisions will be made by an assistant choosing on your customer's behalf. If your products aren't described clearly and accurately, the assistant won't pick them. Your product data is quietly becoming your shop window.
Microsoft's Four-Point Checklist for an AI-Ready Christmas
Microsoft Advertising has published practical guidance on preparing for AI-driven holiday shopping. It boils down to four areas:
- Feeds: upload your full eligible catalogue, not just the products you advertise. Keep stock levels up to date, and use a separate promotions feed for festive offers.
- Measurement: set up Universal Event Tracking and the Conversions API. Send the value of each sale, not just the number of sales.
- Campaigns: merge any campaign with fewer than 30 conversions in 30 days into a larger one. Give new bidding strategies at least four weeks to settle, and change budgets by about 15% at a time.
- AI visibility: use Microsoft Clarity to see how AI tools find your brand, and look at Brand Agents. One retailer, Alexander Del Rossa, saw more than three times the conversion rate in sessions where a Brand Agent helped the shopper.
It's Microsoft's advice, but most of it applies just as well to Google and Facebook. Meta's Conversions API and product catalogue work on the same principle. The better the information you give the machine, the better the decisions it makes with your money.
The timing point matters most. It's late September. If a campaign needs four weeks to settle, anything you restructure now will only just be steady by late October. Leave it until November and you're asking the algorithm to learn during your busiest, most expensive weeks. I wouldn't take that gamble with anyone's Christmas.
Facebook Ads Benchmarks for 2026: Handle With Care
WordStream has published its annual Facebook Ads benchmarks, based on more than 1,800 Meta Ads campaigns. It breaks down click-through rates, costs per click, conversion rates and cost per result by industry, so you can see how you compare.
Benchmarks are useful, but they're easy to misuse. An average cost per click tells you nothing about your margins, your order values or how often customers come back. A shop selling £400 garden furniture can happily pay far more per customer than one selling £12 phone cases.
Facebook and Google also do different jobs. Facebook puts your products in front of people who weren't looking for them yet. Google catches people who are already searching. Comparing their numbers side by side rarely tells the full story.
So treat these figures as a rough check, not a target. If your costs are far out of line with your industry, find out why. But the real question is whether Facebook brings you profitable customers, which leads neatly to the next story.
Would Those Sales Have Happened Anyway?
PPC Hero has a solid piece this week on incrementality testing. It sounds technical, but the idea is simple: how many of these sales would have happened without the ads?
That matters because every platform wants the credit. Google, Facebook and your analytics can all claim the same order. Incrementality testing gets past that by running a controlled experiment. The two main approaches are:
- Geo holdout tests: switch ads off in certain regions for a set period. Then compare results with regions where the ads kept running.
- Lift tests: show ads to one group of people but not to another, then compare the two. Google's Conversion Lift studies work this way, and Meta offers its own lift testing too.
The article's advice is sensible. Agree what you're measuring before you start, and avoid promotions and seasonal peaks. Work out the smallest change you'd need to see before you can trust the result. And don't ignore a disappointing outcome.
My take? Don't run one in November. But in January, when things calm down, it's one of the most useful things an e-commerce business can do. That's especially true for Facebook, where a lot of the value comes from people who see an ad and later search for you on Google.
Google Starts Paying for Content Used in AI Answers
Google is piloting payments to publishers whose content "contributes significantly" to answers in Gemini, AI Overviews and AI Mode. Publishers in the pilot see their monthly earnings in Search Console, though one executive described the calculation as "quite black box".
The same round-up covered two other AI stories. Google's John Mueller admitted it's hard to report positions in AI results in a way that's useful. And Cloudflare has added a setting that lets search engines crawl your site but stops your content being used to train AI.
You're not a news publisher, so you won't be paid under this pilot. But it shows Google accepting that AI answers are built on other people's content, and that the rules are still being written.
Think carefully before using the Cloudflare setting. Blocking AI might feel tempting, but most online shops want AI tools to understand and recommend their products. Decide on purpose rather than just flicking the switch. And if anyone promises you a neat "AI ranking" report, remember that even Google says that data is hard to pin down.
What I'm Watching
One thread runs through all of this. AI is increasingly making choices for your customers, whether it's a grocery assistant advertised by cartoon characters or an answer at the top of a search page. That makes the basics more important, not less.
Get your product data clean and complete. Make sure your tracking tells the platforms what each sale is actually worth. Have your campaigns settled well before the rush. Then, once things calm down in the new year, test honestly to see what's really driving your sales.
Over the next few weeks I'll be watching how Google and Meta push their AI shopping features into peak season. If you'd like a second pair of eyes on your feed or tracking before Christmas, get in touch. There's still time, but not much.