Last week, Anthropic published a quietly significant piece titled "Claude is a Space to Think." They also ran a Super Bowl ad, which is about as loud a two fingers to the competition as you can buy. The article says: no ads in Claude, ever. The Super Bowl spot says: we're confident enough in that position to spend millions telling the world about it.

The short version: Claude will remain ad-free. No sponsored links. No advertiser influence on responses. No product placements users didn't ask for. Their reasoning is disarmingly simple. Advertising incentives are incompatible with genuine helpfulness.

Read that again if you work in marketing.

I've spent the last year studying how AI is changing brand strategy and customer experience, how brands get discovered, trusted, and chosen. Testing recipes, throwing plenty away. What follows is what I'm finding. Five things on the specials board from someone who's in the kitchen, not reading the menu from the street.

1. Your website was never really the shop window

Let's be honest with ourselves. We've talked about websites as shop windows for twenty years, but they've always been closer to the last stop on a journey, at the end of a funnel. The place people arrived after they'd already done the googling, asked a mate, read a review, and more or less made up their mind. The shop front mattered, sure. The customer had mostly decided before they walked in.

What's changed is that the journey before arrival has been completely rewired.

Your customers are now asking an AI concierge for directions. That concierge has been everywhere, read every review, and has zero commercial incentive to steer them your way. Your lovely window display is irrelevant if the concierge has never heard of you.

The latest Semrush data shows roughly 60% of all Google searches already end without a click to any website. When Google's AI Mode kicks in, 93% of sessions end without a single visit to an external site (Semrush, September 2025). Ahrefs reported in February 2026 that AI Overviews reduce clicks by 58%. ChatGPT processes around 3 billion prompts a month. Perplexity holds 6.4% of the market share and is growing.

If your website was the end of the funnel before, the question now is what's happening to the entire funnel above it. And the answer is: AI is eating it. The commercial impact of that shift is only starting to register.

Chef's recommendation: I'm mapping where brands I work with appear (and don't) in AI-generated responses right now. I'm testing some of the many tools on the market, and watching everyone from Adobe to Similarweb, Jellyfish and Blackbird pile into the space with tools that show how your brand and content surfaces in LLMs. Worth knowing: they are probability engines, not a source of truth. All good, none perfect, hence if budgets allow, run a few in parallel. Or to start, literally type the questions your customers ask into Claude, ChatGPT and Perplexity and see what comes back. It takes an afternoon (or, like me, if you're dabbling in Claude Code, write yourself an agent to do that very task). The results are sobering and immediately useful. I'd start there before touching anything else, but be warned, the inconsistencies can be huge. More on that later.

2. Trust is the new seasoning

Anthropic states it plainly: Claude's "only incentive is to give a helpful answer." No ad revenue to accommodate. No sponsored placement to serve. Just: what's most useful to this person right now?

I've been obsessing over this because it changes the flavour profile of everything we do in content. For years, the recipe has been: research keywords, write to match, optimise the meta, build some links, wait. It worked. The dish was bland, but it sold. We're moving at pace from Keywords to Trust Signals.

Now the judges have changed. AI models are reading your content the way a well-informed, deeply impatient analyst would. They've already read everything else on the subject. They know when you're padding.

Only 12% of URLs cited by ChatGPT rank in Google's top ten results. Ahrefs, August 2025

The old leaderboard and the new leaderboard are completely different games. And the average B2B landing page? The hero banner, the three bullet points, the "book a demo" button? An AI assistant is never going to recommend that to anyone, because there's nothing there to recommend. Where are the humans, the stories?

Chef's recommendation: It's worth auditing every piece of content you own through a single question: "Would an intelligent, impartial advisor recommend this to someone?" If the answer is no, you've found your problem. And it's always a substance problem. You can't season your way out of bad ingredients, and your old best practices aren't going to cut it.

3. Cooking for two palates at once

Here's the thing that's got me genuinely excited about content again for the first time in years. Every piece of content now has to satisfy two completely different audiences at the same table. That's a complex challenge. It's also the most interesting creative brief I've seen in a decade.

Humans want story, emotion, cultural relevance and voice. They want to feel something (priming). Machines want accuracy, structure, authority, and citation-worthiness. They want to verify something (proving). The best content has always done both. The realisation that matters: the machine audience now controls the reservation book.

Anthropic draws a sharp line between user-initiated commerce (the AI working for the customer) and advertiser-initiated commerce (the AI working for someone else). Yext's study of 6.8 million AI citations across ChatGPT, Gemini, and Perplexity (October 2025) found that 86% of sources cited by AI engines are within marketers' sphere of influence: websites, listings, and reviews. The opportunity is sitting right there. The content just has to earn the recommendation.

ChatGPT mentions brands 3.2x more often than it provides clickable citations. Commercial intent queries drive 4 to 8x higher brand mentions than informational ones. BrightEdge, 2025

The old "awareness at the top, conversion at the bottom" separation is dissolving. The AI conversation is doing both at once.

Chef's recommendation: I'm building content frameworks that explicitly serve both audiences and testing them by running them through AI tools to see how they're summarised, cited, and recommended. If the content disappears or gets mangled, the structure needs work. Think of it as mise en place for machines: everything in its right place, clearly labelled, ready for use.

4. Taste the dish before you serve it

Anthropic's February 2026 article references their anonymised conversation analysis tool, Clio, and a separate study on how people use Claude for support, advice, and companionship. They know an appreciable portion of conversations involve sensitive or deeply personal topics. They know what people actually do with these tools, because they looked. They tasted the dish. They made billion-pound product decisions based on empirical evidence about real behaviour, gathered at scale.

I see the opposite in most marketing teams. Content strategies built on old assumptions and habits. The old playbook about what customers want, how they search, and what they trust. It's like opening a restaurant based on what you reckon people want to eat, having never actually spoken to anyone walking past in the last year.

I've worked with mid-market B2B brands, large enterprises, and embryonic scaleups looking for an edge. Let's take a business. Call them GreenMonkeyFlights (pretty confident that doesn't exist IRL). They have decent Google rankings, a reasonable blog and a content calendar that mostly consists of whatever the product team wants to announce. Nobody had actually researched how their target audience (operations directors at companies with 200 to 500 employees) makes partner decisions. Nobody had mapped that customer experience from the first question to the signed contract. Nobody had asked what those operations directors type into ChatGPT at 10pm when they're trying to solve a problem. That gap is the entire ballgame now. And unlike before, the nuance of those queries is endless and the potential answers also seemingly without limit.

There's less than a 1 in 100 chance that ChatGPT will give the same list of brands in any two responses to the same prompt. SparkToro, January 2026

The recommendations shift constantly. The brands with the deepest understanding of their audience will be in the mix most consistently. But be warned, or at least approach this with eyes wide open. Reading SparkToro's research, you could be persuaded this is a futile race, or rather that the current tools used to measure, optimise and rank are at best probability engines. Sobering reading indeed, but don't lose sight of what makes great content: relevant, trustworthy and useful to an audience who wants their problems answered by an agent.

Chef's recommendation: I'm seeing the best results from teams that treat insight like a seasonal ingredient. It's expensive, it goes off quickly, and it makes everything better. Qual research. Ethnography. Conversation analysis. Surveys. If you're still working from a persona document somebody wrote in 2022, it's worth asking what's changed since then. Probably everything. Don't rely on one source of truth as you approach your GEO strategy, follow your nose, and don't skimp on insights. You wouldn't build any other strategy without them, and this is no different.

5. The most expensive thing on the menu is doing nothing

There's a particular kind of paralysis that affects large organisations when they face unfamiliar territory. It looks like caution. It feels like prudence. It is actually the most expensive item on the menu, because you're paying for the time you lose while everyone else is learning.

We've seen it before. It's the Blockbuster response. The "let's wait and see what happens with this streaming thing" meeting. The "our customers still prefer to come into the store" PowerPoint. And then one morning, you wake up to find Netflix has eaten your lunch and dinner, and is eyeing your breakfast.

Every month spent debating whether AI changes your brand strategy and content approach is a month your competitors spend figuring out what works. Gartner reported that generative AI usage for content discovery grew 156% year-on-year. The kitchen is already cooking. The question is whether you've got a table.

Anthropic's article carries an implicit challenge for every brand: if the most sophisticated AI systems are optimising for genuine helpfulness, what are you optimising for?

Chef's recommendation: Permit yourself to get it wrong. I'm constantly running small, fast experiments. Testing content in AI environments, measuring what gets picked up, and feeding learnings back into the next piece. The strategy is coming from the doing. I'd rather iterate on a bad first attempt than perfect a plan that'll be out of date by the time it's signed off. It's baloney to think mastery exists in this arena when models are advancing and we don't have the data sets to fully understand the long tail. Get experimenting. The ingredients are at your disposal.

What's the difference between me and you

Content strategy has been a bit boring for a while, if I'm honest. My interest wavered. A declining-returns game of SEO formulae, gated PDFs and webinars nobody actually watches. The craft and creativity got squeezed out by the process.

That's over. What's emerging demands genuine skill. Creative vision and commercial rigour in the same breath. The ability to create content that satisfies humans and machines simultaneously, that carries emotional weight and structural rigour, that's distinctive enough to stand out and credible enough to be cited. That's the difference.

The tools got more sophisticated. The craft came back. And the people who'd been quietly caring about the fundamentals all along suddenly had a massive head start. I'm finding this genuinely exciting. The menu just changed, and it favours the cooks who care.

Trust Signals (n.): The qualities that earn a brand's content recommendation by an AI assistant with no advertising incentive: substance, authority, structure, citation-worthiness and demonstrable usefulness. The successor to keywords as the unit of discoverability. See also: Priming and Proving.

Asked at the counter

What is GEO, and how is it different from SEO? Generative Engine Optimisation is the practice of making a brand visible and citable inside AI-generated answers (ChatGPT, Claude, Perplexity, Google's AI Mode), rather than ranked in a list of blue links. SEO optimised for the index; GEO earns the recommendation. Only 12% of URLs cited by ChatGPT rank in Google's top ten, so they are different games with different winners.

How do I find out whether AI assistants recommend my brand? Start manually: type the questions your customers actually ask into Claude, ChatGPT and Perplexity and record what comes back. It takes an afternoon and the results are immediately useful. Visibility tools exist (and multiply monthly), but treat them as probability engines rather than truth.

Should B2B brands optimise content for ChatGPT? Yes, but not by gaming it. 86% of sources cited by AI engines are already within a marketer's influence (Yext, 2025). The work is substance: content an intelligent, impartial advisor would genuinely recommend, structured so a machine can parse and cite it.