Taste: Automated
This week in AI & marketing
If you have been following us for a while now, you might remember our big 4-part piece on taste (start part 1 here). We argued that AI will automate, or at least enhance, marketing workflows to the point where the only moat will be human judgment in aesthetic choices.
AI can make decisions at least as good as humans' but 100x faster on issues supported by data. So humans should only stay in the loop for the input (managing what data AI should look into and why), and for the review of the output (just to safeguard that it is the right thing AI produced). For the rest, it’s all automated.
We thought that taste, design, aesthetics, and grand-level strategic decisions would stay human. AI-powered, but still human-driven. Yes, there are some elements even there where AI will take over, like UX design is mostly data-based, but we thought AI won’t be able to produce amazing frontend, layouts, and full brand designs. Also, AI won’t be able to produce entire campaign strategies based on customer decisions, choices, feelings, etc. Or at least not within this short period of time, which means we marketers still have some time to catch up.
Well, we were wrong.
We wrote our taste series in November last year. Around half a year ago. In January, Claude Code was able to spin up entire sites with good enough design UX. And the months have passed, and passed, and now, we have Fable.
We are careful to say clickbait statements like, groundbreaking release, or something like graphic design and marketing copywriting is dead, thanks to Fable. But to be honest, we have been using it for the most versatile marketing tasks, and at every output we look at, our faces are like Ron Swanson looking at a vegan burger.
— Torsten and Peter
Claude Fable 5 Is the First Model With Taste
Anthropic released Claude Fable 5 this week, the first publicly available model from its Mythos class — the family that gave the cybersecurity world a small heart attack earlier this year. It’s state-of-the-art on nearly every benchmark, which is also what every launch post has said since 2023.
We’ve been putting it through its paces, and three things actually matter for us, marketers.
First, the design work is genuinely creative: original ideas, not remixed Dribbble. Look at this example. No model even comes close:
It also has more taste than any model we’ve ever used when it comes to writing copy. Asking it to generate title ideas sounds like shooting at sparrows with a cannon, but believe us, the difference is enormous.
Lastly, it’s seriously good at strategy. Feed it your positioning, your product info, your competitor intel, and watch it produce a GTM plan or a full marketing strategy that doesn’t read like a consulting template.
Two practical notes before you run off to test it. Fable 5 is included on paid Claude plans at no extra cost through June 22; after that it moves to usage credits, and it is expensive. And because the underlying model is genuinely dangerous in areas like cybersecurity and biology, requests on those topics get handled by the older Opus 4.8 instead — you’ll be told when it happens. For marketing work, you’ll likely never see it.
Facebook Gives Creators an Analyst in the Dashboard
If you’ve ever seen a creator paste analytics screenshots into ChatGPT and ask why a Reel flopped, Meta just productized that workflow. Creator Assistant lives directly in the Facebook creator dashboard and answers performance questions in plain language — when to post, why one Reel outperformed the rest, how your audience has shifted — using the page’s own data. It also brainstorms: trending audio, cultural moments, content angles matched to your niche and goals. It’s rolling out to creators in the US, Canada, and India first, with more countries promised.
The strategic read is simple: ChatGPT never sees Meta’s full first-party signal, and this does.
Every question a creator asks inside the dashboard is a question they’re not asking a third-party tool, and one more reason to keep posting on Facebook instead of TikTok.
Buried in the same announcement: AI Reels translations are expanding to Arabic, Bahasa Indonesia, French, Thai, and Vietnamese, with optional lip-sync. Half a billion people already watch AI-translated videos on Facebook every week. That’s the number we’d pay attention to.
Meta Wants an AI Running Your WhatsApp Storefront
There are whole parts of the globe where entire businesses run on WhatsApp — the catalog, the haggling, the payment-confirmation screenshot, all of it. So when Meta makes its Business Agent globally available, it is more than a chatbot update. The agent answers customer questions, recommends products from your catalog, qualifies leads, books appointments, and closes sales across WhatsApp, Messenger, and Instagram DMs. It learns from your website, chat history, and FAQs, and replies in your tone and the customer’s language.
Getting started is free, then it moves into paid WhatsApp Business Premium tiers, with large enterprises billed by token usage. There’s also an enterprise platform that connects the agent to systems like Shopify and Zendesk so it can take real actions: updating orders, booking calendar slots, writing back to your stack. For anyone running click-to-WhatsApp campaigns, the first responder is now an agent that qualifies and converts at 3 a.m.
One warning: the agent is only as good as the data you feed it — a messy catalog, vague FAQs, a missing return policy, and it will improvise. Confidently.
ChatGPT Gets Its Google Shopping Moment
OpenAI added a “product feed” campaign type to its Ads Manager. Retailers connect a catalog (the specs support anywhere from 1,000 to 2 million SKUs), set filters, and ChatGPT auto-generates sponsored product units from your titles, images, and pricing — shown below the organic answer, clearly labeled, triggered by conversational intent rather than keywords. If you’ve run Google Shopping or Meta catalog campaigns, the mechanics will feel instantly familiar.
This is the update that turns ChatGPT advertising from a curiosity into something resembling a performance channel.
CPC bidding, a conversion pixel, adtech partnerships with Criteo: the stack has assembled remarkably fast over the past few months. The open question is whether conversational intent converts like search intent does, and nobody has real data yet, including OpenAI. But your feed already exists in Merchant Center, so testing this is easy. Being early on a new ad surface has historically been the cheapest traffic you’ll ever buy.
A German Court Just Made Google Own Its AI’s Words
The Munich Regional Court issued a preliminary injunction holding Google directly liable for false claims in its AI Overviews. Two publishers found themselves described as scams running “dubious business practices” — claims that appeared in none of the cited sources. The court’s reasoning: AI Overviews are independent, new statements in Google’s own words, so the liability shields that protect search engines for third-party links don’t apply. The “AI can make mistakes” disclaimer didn’t save them either.
Sure, it’s preliminary, it’s one regional court, and Google can appeal. But the logic generalizes to every answer engine that names companies and people, and the court itself noted its reasoning could carry beyond Germany.
Two takeaways for you.
If an AI Overview spreads falsehoods about your brand, you now have a legal lever, at least in the EU. And monitoring what AI systems actually say about you stops being a side project — barely 1% of users click through to the sources, so the summary is the brand impression.
Share with a colleague. Earn paid membership months.