Summarize with AI
A website messaging framework is a one-page document that settles six things before anyone writes a page: the category noun, the buyer, what you replace, three jobs, one proof for each job and the one call to action. Every surface, from the homepage hero to the pricing page to your LinkedIn tagline, is then written from those six lines instead of from a blank page.
Most AI startups skip it. The hero gets written the weekend before the raise, pricing gets written by whoever had a free afternoon, and the LinkedIn page gets written for a conference booth. Each one was true in the week it was written, so each one describes a slightly different product. You know which version is current. A stranger only has the pages, and when they disagree he has no way to tell which one to believe.
Vision AI, unlimited storage and compliance agents: one company, three pitches
Picture a startup, used here as an example and not a client, that reviews dashcam footage for trucking fleets of 50 to 500 trucks. The cameras are already in the cabs. Every hard brake or swerve saves a clip. A safety manager at a 200 truck fleet gets around 300 clips a week. The product watches them first. It tags each one (following too close, phone in hand, rolling stop), drops the false alarms and drafts a short coaching note the manager can send to the driver. On Monday morning the queue reads 12 of 287 clips need you.
Now open the company's public pages in three tabs. The homepage headline says Vision AI for safer roads, and the subhead says The intelligence platform for modern fleets. The pricing page says Simple pricing for every fleet, charges per camera per month, and lists Unlimited video storage as the first line under the Starter plan. The LinkedIn page tagline says AI agents for transportation compliance.
That is three companies. The first is a computer vision lab. The second sells cameras and cloud storage. The third automates paperwork for regulators. None of the three pages says safety manager, and none says coaching, which is the thing a buyer actually pays for. The demo form adds a fourth reading. Its first question is How many cameras do you need?
Why every page was right on the day it was written
Nobody made a bad call here. The founder wrote the hero before the seed round, when investors liked the words vision AI. A contractor built the pricing page from a competitor's layout, so it copied that competitor's per camera model and its storage line. The cofounder wrote the LinkedIn tagline for a trade show where compliance was the theme of the year. Each person wrote one section, alone, on a different week, with no shared page to check against.
Watch the founder's week and it shows up. On demo calls he answers the same question first, roughly whether they sell the cameras. He says we work with the cameras you already have on most calls this month. He has said that sentence more often than any line on his own homepage, and it appears nowhere on the site. When a friend says the site feels off, he rewrites the headline again. Pricing and LinkedIn stay as they are, because nobody owns them.
This is the part most advice on website copy gets wrong. It treats the problem as a weak headline and offers formulas for a better one. A better headline on one page makes the mismatch louder, because now the homepage is clear and the pricing page still sells storage. The fix sits one level up, in a document no visitor ever sees.
What three stories cost a sales-led startup
For a company that sells through demos, the site has one job: book calls with the right buyers and turn the rest away. A page that sells cameras books fleet owners shopping for cameras. A page that says compliance books people who want help with an audit next month. The founder takes those calls himself, because at ten people there is nobody else, and each one is half an hour that could not have closed.
The cost compounds in places you do not see. A buyer who checks your LinkedIn page after the homepage now has two answers and trusts neither. An AI assistant asked what your company does builds its answer from the pages it can read, so if those pages disagree, the answer is a blend of all three. And time is short. Across all sectors the median gap from seed to Series A is around 616 days, and roughly 70 percent of seed companies never raise a Series A. Weeks of demos spent explaining what you are not come out of that window.
The one-page messaging framework, line by line
Here is the whole document for the dashcam example. It fits on one screen, and that is the point. A messaging doc longer than one page stops getting opened by the second month.
- Category noun. Dashcam video review for fleet safety teams. Use a noun a buyer already searches for, not one you invented.
- Buyer. Safety managers at trucking fleets of 50 to 500 trucks that already run cameras. Name the role, the size and the condition that makes them ready.
- What it replaces. Watching every flagged clip by hand. One line is enough here; what to write on it is a post of its own, naming what your product replaces.
- Three jobs. Sort the week's clips down to the ones that need a person. Write the coaching note for each driver. Keep a clean record per driver for the insurer.
- One proof per job. The review queue showing 12 of 287. A real coaching note with the driver's name blurred. The exported driver file, one page per driver.
- The one call to action. Send us a week of your clips. Every page ends here, and nothing on the site competes with it.
Add one more short line under the six: words we do not use. For this company that list reads vision AI, intelligence platform, agents and storage. Those words were each true of a past version of the pitch, and each one pulls the reader toward a company this is not.
There are no adjectives, brand values or tone of voice here. Those belong in a brand guide. This page holds only the claims a stranger needs.
Writing each page from the six lines
Once the doc exists, the pages stop being writing tasks and become assembly. The homepage headline is the category noun, and the subhead is the buyer line plus what you replace. Our guide to SaaS hero section design covers what the visual and the button then need to do. The three sections below the hero are the three jobs, each with its one proof as the image. The pricing page headline repeats the buyer line, and the price is per truck, because the buyer counts trucks, not cameras. The LinkedIn tagline is the category noun, word for word. The demo form asks fleet size and camera brand, which sorts buyers instead of inviting the wrong ones.
A test for any page: hold the doc next to it and point to the line each section came from. A section that cannot point anywhere is either a new claim that belongs in the doc first, or filler that can go.
This is not only an AI problem. Mi-VAD builds a heart pump about the size of a coin, and its whole case rests on being small. The Mi-VAD site talks to two audiences at once, investors and clinicians, and its homepage makes the case for small before it says a word about specifications. One claim, carried first, gives both readers the same company.
Who writes it, and why it is not the designer
The founder writes the first draft, in one sitting of about two hours, from the words he already uses on calls. Pull up the last three demo recordings and write down the noun you said in the first two minutes. That noun is usually better than anything on the site, because it was tested on buyers who could leave.
A copywriter or designer can sharpen the doc. They should not write it from scratch, because the doc is a set of decisions about who you sell to and what you leave out, and those are the founder's calls. When an agency starts a website project without one, it writes its own version on the fly, section by section. That is how the three company problem gets rebuilt in a nicer layout.
The 20 page templates are the other trap. Message houses and brand pyramids suit a company with a marketing team. At ten people nobody opens a 20 page document before changing a pricing table. One page gets opened, so pin it where site changes get discussed.
When to change the doc after a pivot
Change the doc when line one or line two changes: a new category noun, or a new buyer. A pivot from fleet safety to insurance underwriting changes both. A new feature usually changes nothing, unless it adds or replaces one of the three jobs. If a job changes, swap it and its proof together.
The order matters. Edit the doc first, then rewrite every surface from it in the same week, so the old and new stories are never live side by side for long. Our post on what changes on your site when you reposition your ICP covers which pages go first and how to keep the search rankings the old pages earned.
Five days to write it and roll it out
- Monday. Copy the first line of five surfaces into one doc: homepage headline, pricing headline, LinkedIn tagline, demo form question one and the second slide of your deck. Circle the noun in each and count the companies.
- Tuesday. Listen to three recent demo calls. Write down the noun, the buyer and the first job you mention when nobody is reading from a script.
- Wednesday. Write the six lines and the words we do not use. Stop when it fits on one screen.
- Thursday. Rewrite the homepage hero and the LinkedIn tagline from lines one to three. These two are read most often by people who have never met you.
- Friday. Fix the pricing headline and the demo form so they name the same buyer. Then ask someone outside the company to read the homepage and say what you sell. If the answer matches line one, you are done.
A messaging doc is a page of text, and no buyer will ever read it. They read the site that gets built from it. Studio Maydit does that second part: website design for AI founders in the US, UK and Europe, built in Framer, Webflow or custom code. Dualite reached 100,000+ users in seven months after design work that supported a repositioned ICP. Projects run as a fixed scope of three to four weeks, or as a monthly retainer with no long lock-in. If your six lines are written and your pages still describe three companies, book a 45 minute call.
