Cannonball Media
Websites

Not screenshots. The actual sites.

Every one of these is running live below, the same build the public gets. Click into one and use it. Switch it to phone width and watch it rearrange itself.

chesapeakeciderco.com
Open
Loading
Chesapeake Cider CoHard cider brand, Baltimore. Opens on an age gate.
mosiekatekollective.com
Open
Loading
Mosie Kate KollectiveSocial media and content agency, nationwide.
wishweaver.ai
Open
Loading
WishweaverAI Disney World trip planning. Talk to Stella on the page.
Built to be read by AI too

A site an AI assistant cannot read is a site it cannot recommend.

People increasingly ask an AI assistant before they open a search page, and it answers from whatever it managed to read. That turns out to be a build problem more than a writing problem, and most site builders get the default wrong.

Zero
of the major AI crawlers run JavaScript

ChatGPT’s and Claude’s crawlers fetch your JavaScript files and read them as plain text, but they never execute them. Anything your site draws in the browser after load is, to them, not there.

Vercel, crawler analysis
~35%
of AI crawler fetches land on pages that do not exist

ChatGPT 34.8%, Claude 34.2%, against Googlebot’s 8.2%. They are worse at guessing than Google is, so stale links and missing redirects cost you more here than they do in search.

Vercel, crawler analysis
28%
of Googlebot’s crawl volume now comes from AI crawlers

939 million requests in a month between ChatGPT and Claude alone, against Googlebot’s 4.5 billion. Not a rounding error, and not something you have to opt into for it to be happening.

Vercel, crawler analysis

Figures current as of September 2026.

So this is how we build
Rendered on the server

The words are in the HTML before the browser does anything. This is the single biggest difference between a site an AI assistant can quote and one it cannot see, and most template builders get it wrong by default.

Marked up so a machine knows what it is

Structured data for the business, the services, the people and the FAQs, so an answer engine does not have to infer what your homepage is claiming.

Answers written as answers

Plain statements a model can lift and attribute, headings that match how people actually ask, and facts in text rather than baked into an image.

Clean URLs and honest redirects

Since a third of AI crawler fetches already miss, nothing we build leaves a dead link behind when a page moves.

A deliberate robots.txt

Every AI company publishes the name of its crawler. Which ones may read you, and which may train on you, becomes your decision instead of an accident.

Fast, because it is small

Fewer bytes and less script is better for a person on a phone and better for a crawler with a budget. The same work pays twice.

None of this is a bolt on, and there is no monthly fee attached to it. It is what a site should have been built like anyway. The sites above are built this way, which is part of why they are quick enough to run inside another page.