Somewhere, right now, an ai agent is telling a buyer about your business. It's summarising you inside a chat interface. It's ranking you on a procurement shortlist an analyst built with a prompt. It's inside a Gmail plugin that decides whether the email from your sales rep is worth surfacing at all. Its answer might be right. Its answer might be flattering. Its answer might be an amalgam of your competitors' language, your last website refresh from 2022, and a Reddit thread you never knew existed.

You aren't in the room. You never will be.

Welcome to the new default state of your brand.

For twenty years, marketers have chased SEO as the discipline of being findable. Findability was the game because search was the gateway, and rankings were the scoreboard. That game hasn't stopped. But it's been quietly folded inside a bigger one, which is being describable. And describable, in 2026, is a question about machines.

Because your buyer no longer starts on Google. They start in Claude, or ChatGPT, or Perplexity, or an agent inside their CRM that pre-briefs them before a meeting. The output they see isn't a list of links. It's a paragraph. Sometimes a table. Sometimes a recommendation, hedged. Whatever it is, it's a summary, and the summary is written by a machine that has read whatever ended up in its corpus. Including or excluding you. Fairly or otherwise. Recently or not.

Many aren't worried yet. Google still gets 9 out of 10 searches. Those watching closely will know that 68% of searches end without a click. It was 60% last year. Google is serving ai summaries for half of their searches and rolling out ai search across android. The flattening is upon us.

You can test this yourself in about ninety seconds. Open a frontier model. Ask it to describe your company to a mid-market CFO. Read the answer as if you were the CFO. Ask yourself. Is that us. Is that the point of view I've spent years sharpening. Does it reach for the examples I'd have reached for. Or has it flattened me into the mean of my category, polite and generic and exactly the colour of every competitor on the shortlist.

Most companies fail this test badly. Not because the models are stupid. Because the raw material the models trained on doesn't add up to a brand. It adds up to a category. A hundred press releases, forty thousand words of homepage copy, some case studies that all use the phrase "trusted partner", and a founder's LinkedIn post from 2019. The model does what it's supposed to do. It averages. And the average of most companies' output is generic, because most companies have been producing generic output at industrial scale for a decade, and are now discovering, live, what they built.

Here's the shift. The buyer conversations you aren't present for are now the dominant channel through which your brand is understood. Which means the raw material you publish, everywhere, forever, is brand infrastructure. Not marketing collateral. Infrastructure. The models eat it. The agents cite it. The buyer forms an impression from it before you ever get a meeting.

Two implications, both uncomfortable.

The first is that inconsistency is now catastrophic in a way it wasn't before. When a human read your website, they forgave the drift between the homepage voice and the careers page voice and the pricing page voice. A model doesn't forgive. A model averages. If your voice is scattered, your machine-summary is scattered. If your point of view is negotiable, your machine-summary is generic. There is no charitable reader anymore.

The second is that the fix isn't more content. It's coherent content, produced against a spec tight enough that a machine can learn it. Codified voice. Specific positions. Language you defend rather than negotiate. Repetition of the same claims, in the same words, on every surface the models are reading. This isn't SEO 2.0. It's a discipline closer to what engineers call an API contract. Your brand needs to be readable, callable, and predictable across every surface a machine might touch.

Most CMOs are still spending on campaigns. Campaigns can't feed an agent. Campaigns are episodic. Agents are ambient. The asset that matters now is the standing corpus, coherent and specific, that the model consults every time a buyer asks about you.

Decide who you are. Say it the same way every time. Publish it everywhere.

Do it before the machines make up their own minds.

Because they already are. And they aren't going to check with you first.

Plan B helps organisations become legible to humans, machines, and themselves. If you want to know what a model thinks your company is, ask it. Then get in touch.