The 2026 Stale Website Report: what 32,574 homepages told us, and what ChatGPT reads when your buyers ask who to call.
still says 2025 or earlier, in September 2026.
Why does ChatGPT keep recommending the weaker competitor?
Most B2B founders we talk to have a version of that complaint. Your product says 2026. The team just closed its best year. Then a buyer types "who should I call for revenue cycle management" into ChatGPT, and the answer names a firm you beat every time you're in the same room.
Here's the diagnosis, and it's plain. The weaker competitor gave ChatGPT something it can explain. Your homepage stopped talking a few years ago, so the engines are describing the old version of you... or somebody else entirely.
We wanted to know how common that is, so we read the one line on 32,574 homepages that almost nobody looks at: the copyright year in the footer.
We ran it in September 2026 because the ground under this question moved in about eighteen months. Buyers who used to open a search box now open a chat window, and the answer comes back assembled from what the engines have already read and stored. That picture updates slowly, and it compounds. A site the engines can quote gets quoted again. A site with nothing new gets described from whatever they last found, for as long as that's all there is.
Your buyers changed how they start. They ask ChatGPT who to call before they call anyone. ChatGPT is the new word of mouth, and it works like the old kind: it repeats what it's heard. The difference is where it hears it: the engines read your website, and they can only repeat what the site gives them.
ChatGPT recommends the company it can explain. Your website is where it gets the words.
2,481 homepages hand ChatGPT an old year and nothing to quote.
Follow the 20,496 dated homepages through what the engine finds: a current year or a gone one, then something to quote or nothing at all. Band width is the number of homepages.
| Path | Homepages | Share of 20,496 |
|---|
Source: PitchKitchen scan, September 2026. Aggregate only. "Something to quote" means the homepage links to a blog, insights, resources or news page.
We scanned 32,574 US B2B homepages this month. Of the ones that show a year in the footer, one in four still says 2025 or earlier.
Here's the arithmetic, so you can check it.
Against everything we touched, the stale share is 15.4%. We lead with 24.5% because a footer with no year tells you nothing either way.
892 homepages have missed six or more Januaries. Two have missed thirty-one.
The 5,022 frozen homepages, stacked by how many Januaries their footer has missed. The steps fall the way you'd expect until the last one.
| Januaries missed | Footer says | Homepages |
|---|
Source: PitchKitchen scan, September 2026. Aggregate only.
2,097 sites say 2025, so they missed one January. 993 say 2024. 513 still say 2023. 892 say 2020 or earlier.
A wrong footer year used to be a small embarrassment. Visitors shrugged and kept reading.
A site that says 2023 has told the engines, in writing, that nobody has touched it since 2023. The headline above that footer is probably older. When a buyer asks who to call, the engines describe the 2023 version of you, or someone else whose site gave them more to work with.
Here's what makes that expensive. When the engine doesn't name you, nothing happens. No bounce, no abandoned form, no line in your analytics. The buyer gets three names, calls them, and picks one. The deals you lose this way never show up in a report, because you were never in the running to lose them.
Your website says 2023, and ChatGPT believes every word.
In this report, 2023 stands in for the last year before buyers asked AI first. The real footers say every year back to 1995.
Let's be precise about what a footer proves. A footer that stopped is a symptom. It means nobody, and no system, has touched the shell of the site this year. The disease sits behind it: a story that stopped at the same time, so the engines have nothing newer to repeat. A footer takes a minute to fix. The story behind it takes a conversation.
Your 2027 pipeline starts with what ChatGPT reads this quarter.
Among the founder-owned B2B services companies in the scan, 28.0% of dated homepages are stale (2,961 of 10,558). B2B software companies of the same size come in at 21.8% (1,913 of 8,761). The funded companies on our Healthtech Index came in at 16.7% (33 of 198), for a reason we'll get to in the method notes.
Size matters too. Companies with 11 to 50 employees are stale at 27.5% (3,020 of 10,975). At 51 to 200 employees the rate drops to 22.2% (1,854 of 8,344). Smaller firm, older footer.
The verticals tell the sharper story, named the way we searched for them.
Share of dated homepages that say 2025 or earlier, by vertical
Nine of the 28 verticals we searched. Revenue cycle management leads; AI companies trail.
| Vertical | Dated homepages | Say 2025 or earlier | Stale rate |
|---|
Source: PitchKitchen scan, September 2026. Aggregate only.
Read the top row twice. In revenue cycle management, 44.7% of the firms with a dated footer are telling the engines a year that's gone. When a hospital CFO asks ChatGPT for a shortlist of RCM firms, close to half the category has handed it a date that's gone. The firm with a current story in RCM is competing against silence.
At the bottom, AI companies (13.8%) and B2B SaaS (16.3%) keep their footers current. Some of that is care, and a lot of it is a footer that updates itself.
The first question a CEO asks after finding a stale footer is "which platform should we move to?" The scan's answer saves you a migration.
Share of dated homepages that say 2025 or earlier, by platform
Wix and Squarespace top the list and are a small slice of B2B. WordPress runs more than half of everything we dated.
| Platform | Dated homepages | Say 2025 or earlier | Stale rate |
|---|
Source: PitchKitchen scan, September 2026. CMS from generator tags and asset fingerprints.
Wix and Squarespace top the frozen list, and they're a small slice of B2B: 1,043 dated homepages in a scan of 20,496.
Now look at WordPress, which runs more than half of the dated homepages we scanned. It carries 53.1% of the stale sites and 55.8% of the fresh ones, the same share on both sides of the line. If the platform were the cause, WordPress would carry a bigger share of the stale sites than of the fresh ones. It carries the same share.
The scan rules out the platform. The trap is somewhere else.
Most companies fix a stale site by hiring a design agency to reskin the old story on the same CMS. New template, new photos, the same headline from 2023 underneath. The footer says 2026 for a year, and the engines have exactly as much to say about you as before. A website update is a story decision and an architecture decision before it's ever a design decision. Skip the story, and the redesign is the trap with better fonts.
We also checked each homepage for a blog or insights link. Among the stale sites, 49.4% have none. The fresh sites come in at 30.8%. Half the frozen sites have nothing for the engines to quote.
Look at it from the engine's side. A buyer asks who to call. The engine looks for companies it can describe in words it can repeat. A homepage with one 2023 headline and no writing behind it hands the engine one sentence. One good sentence is a thin case for a company this far along.
Now look at the other side of the line. 10,708 homepages in this scan say 2026 and point at something worth quoting, which is 52.2% of every dated site we read. Those companies aren't lucky, and most of them aren't bigger than you. They gave the engines a current story and they kept adding to it, so a buyer asking who to call gets something recent to repeat. The distance between those two groups is a story and a habit, not a budget.
The engines noticed before you did.
Word of mouth used to travel between people, where you could hear it. It travels through a machine now, one that reads your homepage and repeats what it finds, and that conversation happens somewhere you can't listen in.
That splits this market in two. The machine describes one group accurately, in the buyer's own words, every time someone asks. For the rest it reads from a page that stopped talking years ago, and nobody at those companies ever finds out.
The 10,708 homepages on the current side of the line aren't running better software. They do two ordinary things the frozen ones stopped doing.
65.4% of current homepages point at a blog or an insights section. Among the frozen ones it's 49.8%. Where a server told us when the page last changed, 67.3% of current homepages had changed recently, against 45.2% of frozen ones. Put those together and about half the current sites are publishing on a site that actually moves, against under a third of the frozen ones.
Current sites publish, and their pages move. They run the same software as everyone else.
Share of dated homepages in each group. The bottom row is the control: platform age barely moves between the two sides.
| Measure | Says 2026 | Says 2025 or earlier | Difference |
|---|
Source: PitchKitchen scan, September 2026. Aggregate only. "Changed recently" comes from the server's Last-Modified header, which 6,948 of the dated homepages sent.
Now read the bottom row. 86.4% of the current sites run a legacy CMS, and so do 87.5% of the frozen ones. Same platforms, same tools, same constraints. The scan rules out the platform twice: once on the stale rate, and again here on the sites that got it right.
The habit is the difference, and a habit can start on a Monday.
That's the encouraging part of this report. What separates the two sides is a story worth telling and the practice of telling it, and neither one waits on a rebuild.
Two checks before you change anything.
Pull up your own homepage, cover the logo with your hand, and read the headline out loud. Could that line open half the sites in your category? If it could, the engines can't tell you apart either, because that headline is most of what they have.
Use the checker below to see which of the numbers above you belong to. It reads your footer, your platform, your blog, and your headline the way we read 32,574 others.
The tempting fix is to log into WordPress and change the year. Don't.
Fixing the date treats the symptom and leaves the disease.
The engines get exactly as much to say about you as they had yesterday. Two things are wrong with a frozen site, and the footer is a symptom of both.
The large language models behind ChatGPT eat words for breakfast. They're trained on words, and when a buyer asks who to call, they recommend the company whose words explain it. Most B2B websites were written in category language, from a brief nobody in the room believed, and the engines repeat exactly that.
Every change to the site means someone logs into a CMS and types and drags. That's why the site stopped. Going into a website and hand-typing pages is the biggest bottleneck a business website has in 2027, and it's the reason a company with a great year still has a footer from 2023.
The way out is a shift in how the site gets made, and it has four moves.
Sit the CEO, the CRO and the CSO down and answer, in writing, what the company actually does, who it does it for, why they should change, why they should change now, and why they should change with you. The answers run 40 to 50 pages. On the surface that document is boring. It's the most important thing a company can write in 2026, because it's the training data: the context every AI in the building will work from. We call it the Magnetic Messaging Framework. CEOs call it the Bible of the business.
Turn that document into an AI Brand Twin, connected to your own MCP server, so anyone on the team who opens Claude or ChatGPT or whatever model they prefer plugs into it. The website, every email, every proposal, every response to an RFP, every LinkedIn post and every ad comes out of the same truth. Seventeen agencies doing seventeen different things, and five salespeople telling six different stories, become one story.
Skip the rip-and-replace. Engineers call it a strangler fig migration: the new system grows around the old one until the old one can be switched off. Start with the homepage, rewritten from the twin, and let the new site take over the WordPress site page by page. Nothing goes dark, and the engines get something new to read within weeks instead of after a six-month rebuild.
With the twin under the site and a governance rule for who can publish what, the people you authorize publish at high velocity by talking. What goes out is signal: a broad surface of ideas and thought leadership that resonates with the buyers you actually want, because it's trained on the truth about them. Plenty of companies publish a lot today. Most of it is noise, because nothing under it is true.
That's what PitchKitchen's 90-day Sprint locks in: the truth, the twin, the first pages of the new site grown around the old one, and the speed lane with its rules.
The rebuild comes included in the work, because once the story is nailed, the site is a byproduct of the message.
On September 29 we run the first page of that shift on a real homepage, live, in sixty minutes. Bring your URL. One company in the room gets rewritten on air from a twenty-minute conversation, everyone runs the checker on their own site, and you leave with the four moves written down. The room is capped at 100 so it stays a room.
ChatGPT already knows it. Your buyers are about to. How B2B CEOs become the company ChatGPT recommends, in 90 days, without an agency or a six-month, six-figure rebuild.
Bring your URL. A real homepage gets rewritten live from a 20-minute conversation. Then we publish it and ask ChatGPT to read it, in front of the room.
Greg Rosner is a messaging therapist for B2B founders bringing innovations into the world. PitchKitchen works from the Magnetic Messaging Framework. The 90-day Sprint nails the story first, then includes the AI-native rebuild of the site. Open Kitchen is the fractional CMO in a box, at a flat monthly fee. For CEOs who'd rather learn to do the rewrite themselves by talking, there's the Claude training cohort.
On the name. Every mention of ChatGPT in this report stands for the whole category: Claude, Perplexity, Gemini, Manus, Ox Alpha, and the ones that ship after this goes out. We did not measure any single assistant's answers. We measured what all of them read, which is the homepage.