Every Nova site now speaks llms.txt

A growing share of local customers never see a search results page. They ask an AI assistant — "which dental clinic near me is good with kids?" — and get an answer with somebody's name in it. We've written about why that door matters; one of our clients gets measurable weekly traffic from a single AI assistant.

So here's a small, concrete thing we shipped for it: every website Nova hosts now serves /llms.txt — ours included (novabuildbot.com/llms.txt).

What llms.txt is

llmstxt.org is a simple convention: a markdown file at the root of your site that gives language models a curated map — who you are in one blockquote, then your key pages with one-line descriptions. It's sitemap.xml for readers that understand prose:

# NovaBuildBot

> NovaBuildBot gives Indian small businesses — clinics, dental practices,
> doctors, local stores — an AI growth team ...

## Product

- [NovaChat](https://novabuildbot.com/novachat/): the 24/7 AI receptionist
  for websites and WhatsApp
- [Pricing](https://novabuildbot.com/pricing/): plans and what's included

An AI assistant crawling your site can reconstruct this from your HTML — sometimes. A file that says it plainly removes the "sometimes."

The design decision: mirror, don't author

The tempting way to build this is to have an LLM write a nice description of the site. We deliberately didn't. Our generator mirrors what each page already says — the real <title>, the real meta description, the pages your sitemap actually lists. No second copy of the truth that can drift from the first; if a page changes, its llms.txt entry changes to match, never the other way around.

That's a lesson we keep re-learning in every part of the product: two sources of truth is one source of bugs.

The part the business owner never sees

"Supports llms.txt" would be a checkbox. The actual feature is that nobody has to think about it:

• Existing sites — every site we host was backfilled in one sweep, generated from each live site's own sitemap and page metadata.
• New sites — the scaffold that builds a Nova site now ships llms.txt from day one, wired into the build like robots.txt.
• Every edit after that — when the website agent adds, removes, or retitles a page, updating llms.txt in the same commit is part of its job description. Blog posts are deliberately excluded — the blog index covers them, and a weekly post shouldn't churn the file.
• Tests pin the promise — the rules above are asserted in our test suite, so a future refactor can't quietly drop them.

We also updated our own robots.txt to explicitly welcome the major AI crawlers (with our internal pages excluded). If your business's answer to "can AI read my site?" is a shrug, that's now a solved problem for every Nova client — and if you're building something similar, we'd encourage the same rule we followed: generate from what's live, keep it in the deploy pipeline, and never let it become a second thing to maintain by hand.