We wanted AI to introduce us accurately, too

Ask ChatGPT or Perplexity about a company and the answer is sometimes wrong. Often the information is on the site; it just exists only in a form made for people to read.

We worked on this while rebuilding lightoflife.work. Along the way we found that search engines and generative AI look for different things on the same page. A search engine decides which searches the page should appear for. Generative AI looks for sentences it can carry into an answer.

Sentences that make sense on their own

The sentences AI quotes well still make sense when lifted out. They don't start with words like "we" or "here," and they say who does what within a single sentence.

"We run three businesses" means nothing once it is taken out of context. "LIGHT OF LIFE is a creative venture company that plans and runs its own products, spaces, and experiences" works on its own.

So our FAQ is kept as data, not just as text on a page. The same sentence goes to the screen and to the structured data search engines read, and we edit it in one place. That way the page and the search result never say different things.

llms.txt, a summary for AI

llms.txt is a text file that sits at the root of a site, like robots.txt. It sets aside everything designed for human eyes and describes, in markdown, what the site is and which pages it has.

# LIGHT OF LIFE

> A creative venture company building products, spaces, and experiences.

## Pages
- [Business](https://lightoflife.work/en/business): ...

## FAQ
### What is LIGHT OF LIFE?
...

It isn't an official standard yet. Still, it takes little effort to make, and it lets us say first how we would like to be summarized. Without it, an AI reads the HTML and cuts it up its own way.

Kept by hand, a file like this eventually drifts away from the site. Ours is generated from the same source as the page copy, so when a sentence changes, llms.txt changes with it.

Saying clearly that crawlers are welcome

AI crawlers come under names that are different from the usual search bots. When robots.txt says nothing about them, some hold back, so if you want to be read, it helps to list them by name.

• OpenAI: GPTBot · OAI-SearchBot · ChatGPT-User
• Anthropic: ClaudeBot · Claude-SearchBot · Claude-User
• Perplexity: PerplexityBot · Perplexity-User
• Google: Google-Extended (a separate setting for generative AI training)

If you would rather not have your content used for AI training, this is also where you can say no. Either way, what matters is deciding and writing it down.

Structured data, tied together

Instead of scattering JSON-LD across pages, we put it into a single @graph and let the pieces point to each other by @id. When Organization, WebSite, and WebPage aren't connected, a search engine has to guess how they relate. Writing the connections down removes the guesswork.

Our official accounts are listed under sameAs, so search engines can confirm they belong to the same organization. Each profile also needs to link back to the site before the two are connected. It's an easy step to miss.

What remains after the setup

Even with all of this in place, a new domain takes a while to show up in search. Search engines need time to get to know a site, and longer still when nothing else links to it.

The technical setup takes about a day. What lasts is having something worth reading. That is why we keep writing here about the problems we run into while building, and how we work through them.