Vibe-Coding PanicAEO Strategy

How should a B2B company run its website with AI? There are four ways.

Greg Rosner

By Greg Rosner

Founder of PitchKitchen · Author of StoryCraft for Disruptors

· 6 min read

TL;DR

There are four ways to run a B2B website with AI. Keep your CMS and let AI help write. Plug AI into the CMS, so drafts land there and a person publishes. Split it: the main site stays put, and one new section runs separately, written and updated with AI. Or go AI-native: the whole site lives as files in a code library, with no CMS, and your team changes it by asking an AI. Company size doesn't decide which one fits. Who edits the site and who owns running it do. Every way keeps a person approving the page, every way needs an owner for the quiet work a CMS does, and none of them helps if the story going through it isn't true.

Which of the four ways fits your company?

Find the row that sounds like you. (Your CMS is the system your team logs into to edit the website, like WordPress, HubSpot or Webflow.)

The way✓ Pick it if✗ Skip it if
1. Keep your CMS✓Lots of people edit the site and copy changes every day✗You need many new pages live fast
2. Plug AI into your CMS✓You like your CMS and want pages out faster, with your review steps intact✗Your CMS can't connect to AI yet
3. Split it✓You're adding a big new section, like a library of answer pages, and someone will run it✗Nobody will own a second system
4. Go AI-native✓Few people edit, the site is in one language, and a developer or partner owns the site✗Many editors, several languages, or sign-off on every page

Not sure? Start with way 2. It's the safest move for most mid-market B2B teams.

The question that changes the website conversation

This week a client's team sent us one line that reframed a whole project.

"There's a big difference between building a site and running a site."

We'd been talking about how AI changes publishing. A page that used to take an agency three weeks can now be written, previewed and approved in an afternoon. We run pitchkitchen.com that way, and the speed is real.

Their point was everything after launch. Who owns the sitemap, the redirects, the forms, the cookie banner, the security patches? A CMS does that work quietly. Take the CMS away and somebody has to own every line.

That's the question most website conversations skip. Teams argue about platforms when the real choice is who runs the site.

What's actually holding your website back?

We call it the Publishing Ceiling: how fast your company can get a true, current answer onto the internet without booking an agency ticket.

AI made pages nearly free to write. It didn't make them free to publish. The ceiling is set by your process and by who owns each job, and every one of those jobs still has to land somewhere. Call it the quiet-work list:

  • The sitemap, the list that tells Google and AI engines which pages exist
  • Meta titles and descriptions, the title and blurb that show up in search results
  • Share cards for LinkedIn and Slack
  • Redirects when a page moves
  • Forms, spam filtering and the handoff to your CRM
  • Images sized for every screen
  • Cookie consent
  • Analytics and ad tracking (tag manager, LinkedIn and Meta pixels)
  • Who's allowed to edit what
  • Security patches

Any way you run your site has to cover this list. The four ways below differ in who covers it.

What are the four ways to run a website with AI?

Buyers now ask ChatGPT and Claude before they ever visit you, and the engines cite pages that answer the question in front of them. A team that ships one good page a month falls behind a team that ships one a week. The market has settled into four ways to close that gap.

1. Keep your CMS and let AI help write

The site stays exactly where it is. AI drafts the copy, a person pastes it in and publishes.

It fits when lots of non-technical people edit the site, copy changes daily, or you run several languages. Watch for one thing: you've sped up the writing, and the publishing queue is just as long as it was.

2. Plug AI into the CMS

WordPress, HubSpot and Webflow now have built-in connections that let an AI assistant like Claude or ChatGPT create drafts right inside the CMS. Most teams limit the AI to drafts, so a person still presses publish.

This is the right call for a lot of mid-market B2B companies: a CMS your team knows, a web partner you trust, and a review process that already lives there. Regulated teams especially. Give the AI the smallest role that works, and check what it's allowed to change.

3. Split it

The main site stays on its CMS. One new section, say a library of answer pages or resources, runs separately as files in a code library, with no CMS, and AI writes and updates them. A single setting at your web address sends visitors to the right place, so they never notice the difference.

It fits when you're adding a big set of new pages and don't want to move the whole site to do it. Two systems means two owners, so write down who runs each.

4. Go AI-native: the whole site as files

The whole site lives as plain files in a code library, usually GitHub, the same kind of place software teams keep their code. There's no login screen and no Edit button. You ask an AI that's connected to those files for a change, it shows you a preview of the page, a person approves, and it goes live.

It fits when few people edit the site, it's in one language, and a developer or partner owns running it. Everything on the quiet-work list is now yours.

How do you pick the right way?

No neutral source gives you a page count or a company size where one way beats another. The analysts mostly say "it depends on readiness." Here's what it depends on, as three tests you can run this week:

  1. 1Count who edited your site in the last month. Then ask whether each of them would be comfortable changing a page by asking an AI, with no Edit button to click.
  2. 2Write an owner's name next to every job on the quiet-work list. Every blank is a risk.
  3. 3Time one page change, from the request to the page going live.

If test two leaves the owner column blank, ways 3 and 4 aren't for you yet. Way 2 probably is.

Two more things tip the choice. Several languages, or legal and clinical review that needs a record of who approved what, both point toward keeping a CMS in the loop.

What we see across B2B websites

In October we scanned 32,574 B2B domains for out-of-date homepages. About half the out-of-date homepages ran on WordPress (53.1%). So did about half the up-to-date ones (55.8%).

Same platform, both sides. The platform isn't what keeps a site current.

The moves to way 4 tell a similar story. Sentry, a software company, moved about 2,500 pages to files in a code library and cut the time it takes to rebuild the site after a change from 14 minutes to under four. Another company that moved over 1,000 pages thought it had lost 80% of its traffic, and most of that drop turned out to be a measurement change. A change in how its cookie banner handled tracking meant fewer visitors were being counted. The visitors hadn't gone anywhere. Measure before and after the same way.

One more pattern. Every whole-site example we found is a software, AI or agency company. We couldn't find a regulated or Fortune 500 company running way 4 yet.

A real example

A healthtech company we work with needed a large library of answer pages, written for the questions its clinical and financial buyers ask AI engines.

Moving the whole site was the wrong trade. Their web partner runs the main site well, and their review process lives there. Rebuilding all of it to add one section would have spent months on technical work.

The plan is way 3. The main site stays with their web partner. The answer pages run separately, and one setting at the web address sends visitors to them. If that setting ever fails, visitors simply see today's site. We wrote down who owns each job on the quiet-work list in month one, before a single page ships.

What this means for you

The right way to run your website with AI is the one your team can actually own. Pick it with whoever runs your site in the room, before anyone picks a platform.

Three things to do this week:

  1. 1Run the three tests above with your web lead or partner.
  2. 2Pick the way that matches the owner column, not the one that sounds most advanced.
  3. 3Write down the story every page has to tell before you speed anything up.

Every one of these four ways carries whatever story you feed it. That's the part we work on. The Magnetic Messaging Framework writes down the story your team actually agrees on, so whichever way you pick, the pages going through it say something true.

If you're running one of these four and something broke, tell us. That's the next post.

Questions People Ask

FAQ

Should a B2B company leave its CMS for an AI-native website?

Only if few people edit the site, a developer or partner will own running it, and you'll measure traffic the same way before and after. For many companies, plugging AI into the CMS they already have is the right call.

Can AI publish directly to WordPress or HubSpot?

Yes. WordPress, HubSpot and Webflow offer connections that let an AI assistant like Claude or ChatGPT create drafts inside them. Most teams limit the AI to drafts, so a person still presses publish.

Does moving to an AI-native website hurt SEO?

Not if the move is done with care: keep every page address, or point the old one to the new one, keep the meta titles and the sitemap, and measure before and after with the same tracking. The best-known scare story turned out to be mostly a tracking change.

What's the hybrid way to run a website with AI?

The main site stays on its CMS, and one new section runs separately, written and updated with AI. A single setting at your web address sends visitors to it. It lets you add a big set of new pages without moving the whole site.

We rebuild websites with AI so AI actually recommends you.

See how the 90-Day AI Website Rebuild works →

This article is part of

How do we find the one story that unlocks everything?

The short answer, plus every article we've written on this problem.

Want this kind of thinking shipping for you?

Whichever way you run your site, it can only publish the story your team has agreed on. If that story isn't written down yet, start there.

That's the 90-Day AI Website Rebuild. One quarter, one fixed price: we extract your story, build the Magnetic Messaging Framework and your AI Brand Twin, then ship the website and sales enablement that run on it. $15K–$45K fixed for the quarter, and you own all of it at the end.

About the Author

Greg Rosner

Greg Rosner

Founder, PitchKitchen · Author of StoryCraft for Disruptors · Creator of the Magnetic Messaging Framework™

Greg is a B2B messaging therapist for growth-stage CEOs ($5M-$75M). He helps founders extract the truth they've been hiding from themselves, name the villain in their industry, and build the messaging infrastructure that scales their voice through AI. PitchKitchen has worked with 100+ B2B companies across SaaS, healthtech, fintech, cybersecurity, and AI-driven solutions.