How to Build a Website That AI Can Find: A Filmmaker’s Guide to AI Search

A while back I wrote about why I tore my old website down and rebuilt it, and what I think most filmmakers are getting wrong about how their work gets found. That post was the argument. This one is the how.

People kept asking me the same question after they read it. Okay, but what did you actually build.

So here is the answer, in plain language. No jargon, and nothing you need to be technical to follow.

If you’re interested in the more technical aspects, I created a guide for you here.

If you make films, or anything else, and you want the internet to be able to find your work, this is what the machinery looks like.

Every page says itself twice

The whole thing rests on one simple idea. Every page on my site quietly says itself twice. There is the version you see, with the photos and the words and the spacing. And there is a second version, written underneath in the page’s code where no visitor ever looks, that spells out in plain terms what the page is.

Think of it like a can of soup. The front of the can is designed to catch your eye. The back lists the ingredients in a flat, boring format a machine at the checkout can scan in an instant. My pages have both. The front is for you. The back is for the search engines and the AI assistants, which cannot read a beautiful layout the way a person can. They need the ingredients listed plainly.

So a film page, underneath, simply states the facts: this is a film, directed by this person, made in this year, starring these people, reviewed by these writers. Nothing fancy. Just the facts, written the way a machine can read them.

And each of those little fact lists is connected to the others. The fact list for a behind-the-scenes photo points to the film it came from. The fact list for a person points to the films they worked on. Once everything is linked up like that, a machine can answer a question nobody ever wrote out on the page, like who shot a particular film, because it can follow the trail of connections itself.

The files I wrote just for the machines to read

Those hidden labels live inside each page. There is also a small set of files that sits on top of the whole site, written for the AI assistants on purpose. Almost no website bothers with these, and they are a big reason an assistant can describe my work accurately instead of guessing.

There are a few, and each one does a different job.

One is a short summary of everything. Who I am, the films, the press, what’s new, all on a single page an assistant can read in a few seconds instead of clicking through the entire site.

One is the long version. The full body of work written out as plain text in one place, for the assistants that would rather read it all at once.

One is a plain biography built for machines. Just the basic facts, including the official ID numbers that confirm which James Kicklighter is being asked about, since I’m not the only one, and even a note on how to say my name, so a voice assistant pronounces Kicklighter correctly out loud.

And one is a short, friendly note written straight to the AI assistants, telling them they’re welcome to read my site and mention it. Most websites accidentally shut those readers out, or never say anything to them at all. I decided to roll out the welcome mat instead.

The reason I keep these as separate files, rather than stuffing it all onto the regular pages, is that different readers like their information in different shapes. A person wants a nice-looking page. A search engine wants the plain fact list. An AI wants a clean text file it can read top to bottom. Hand each one the shape it prefers, and you stop losing the ones you were quietly ignoring.

How the listings write themselves

Every page needs two short bits of text. The headline you click on in a search result, and the one-line summary sitting under it. There are thousands of pages on my site. Writing all of those by hand, keeping every one correct, and updating them forever is simply more than a person can do. So that job runs itself, with me checking the work.

Here is how it goes, step by step. The system first gathers what the site already knows to be true about a page: the real cast and year of a film, the real words of a journal post. It hands those true facts to an AI and asks it to write the headline and the summary, following the rules search engines like, such as keeping the length right. The AI can only use the facts it was given. It can’t make up a credit or hand me an award I never won, because it is working from the record and nothing else. Then it shows me the draft, and I approve it, change it, or lock the page so nothing ever touches it again.

A few safeguards keep this calm instead of chaotic. The pages that should never change, like the film titles, are locked into a set format, and I have to deliberately unlock one to edit it. For the lower-stakes pages, like journal posts and photos, I can let the system write and post on its own the moment I publish. It skips anything I’ve already finished. And once a month it checks, quietly, whether the way people search for a page has actually changed, and it only rewrites the handful where that’s true. Everything stable, it leaves alone.

The whole idea there is restraint. A system that rewrites everything all the time is a headache. One that only steps in when a real fact or a real search habit changes is a help.

How the search engines find out, in minutes

A good headline does no good if the search engines don’t know the page changed. Most websites just wait, sometimes for weeks, for the search engines to come back around on their own. I didn’t want to wait.

So the moment I change something, the site tells the search engines directly: come read this again. It also keeps a kind of table of contents for the whole site, a single list of every page, and stamps each one with the exact time it last changed, so an update looks fresh rather than old. There are separate lists just for the photos and just for the trailers, so image search and video search can find those too. The result is that something I post at noon can be picked up by the search engines that same afternoon, not next month.

The loop that tells me whether any of it worked

This is the part I’m proudest of, because it closes the circle. Everything so far is the site sending information out. This is the site listening to find out whether any of it stuck.

On a regular schedule, the site quietly asks the major AI assistants a set of real questions about me and my films, the kind a stranger might type, and writes down whether each assistant actually names me in its answer. Do that over and over, and you get a simple picture over time: is the number of times I get named going up or staying flat, and did a change I made move it.

It pays attention to a few other things, too. It notices which other filmmakers get mentioned in the same breath as me, which tells me where the assistants think I belong. It collects the follow-up questions the assistants suggest, the “people also ask” kind, as new things to keep an eye on. And it spots the gaps: questions the assistants clearly get asked about my line of work that my site doesn’t answer yet. Those gaps turn into a simple to-do list. Instead of guessing what to write next, I can write the exact thing people are already asking about and not finding on my site.

That last part is the real prize. Most people improve their site blind. This one shows me precisely where the holes are.

The part that keeps a big collection honest

None of this works if the collection underneath is a mess, and mine is big. The behind-the-scenes photo archive alone runs into the thousands, on top of years of press and a long list of collaborators. Keeping all of that tidy by hand isn’t realistic, so a separate system handles the boring upkeep.

For the photos: when I upload a burst of nearly identical shots, the system spots the near-twins and keeps the best one. It writes a caption and a short description for every image, plus a written account of what the photo actually shows. That written account is the same thing that lets people find the photo in image search and lets someone using a screen reader know what’s in it. The system even rates each shot for how striking it is, so the best images move to the front on their own.

For the people: every collaborator and cast member is gathered into one clean list, with no duplicates, each with a checked biography, and the connections between people and projects spelled out, so a machine can see who did what. For the press: years of coverage are sorted into one organized library, each article matched to the film it’s about and credited to the outlet that ran it. And new entries start with their basic facts already filled in from trusted sources, so the collection can grow fast without going wrong.

Why it’s built to be fast

Speed isn’t separate from being found. Google and the AI assistants both prefer pages that load quickly, so the whole site is tuned to be fast in ways you never see. The most important part of the styling loads first, so the page appears right away. The connections to outside services, for images and video and tracking, only switch on at the moment they’re actually needed. The tracking waits until you’ve started using the page, so it never slows down that first moment. And any code a page doesn’t need is simply left off. You get a site that feels instant. The search engines notice, and they reward it.

The loop, end to end

Put it all together and these aren’t separate gadgets. They’re a loop. I publish a film, a post, a batch of photos. The listing writes itself from the true facts. The search engines get told to come look. The work turns up across ordinary search, AI assistants, voice, and image search. The tracker checks whether it actually landed, and hands me the next gap to fill. Then the whole loop runs again, a little stronger each time.

The “why” lives in the original essay on filmmaker discovery. This was the how.

None of it is magic, and none of it replaces the work itself. It’s plumbing. But it’s the plumbing that decides whether the thing you spent years making ever reaches the person, or the machine, that goes looking for exactly it.

— James