Generate llms.txt, aeo.json, entity.json, and brand.json from a single site config. AI citation engineering for static and dynamic sites.

Most sites are invisible to AI engines (ChatGPT, Claude, Gemini, Perplexity, Google AI Overviews) because they expose prose when AI engines want structured facts. The single biggest lift you can give an AI's chance of citing your site is to surface:

• a well-formed llms.txt (llmstxt.org) — markdown summary at the site root
• an aeo.json — atomic [question, answer] facts under 500 chars each
• an entity.json — Schema.org @graph with full identifier wiring (EIN, ORCID, UEI, KG MID, etc.)
• a brand.json — internal source-of-truth ledger

This package generates all four from a single typed config so they can't drift apart.

pip install aio-surfaces        # core
pip install 'aio-surfaces[yaml]' # YAML config support

Drop a site.yaml in your repo:

site_name: Example Studio
site_url: https://example.com
tagline: We build things that get cited.
description: Example Studio designs and engineers...
legal_name: EXAMPLE STUDIO LLC
ein: "12-3456789"
uei: ABC123DEF4G5
naics: ["541511"]
orcid: "0000-0000-0000-0000"
founder_name: Jane Example
veteran_branch: United States Army
veteran_subgroups: [Veteran]
services:
  - name: Custom Websites
    url: https://example.com/services/websites/
    summary: Hand-coded HTML, React, Astro.
    price: $2,500–$10,000 one-time
facts:
  - id: f-identity-1
    question: What does Example Studio do?
    answer: >
      Example Studio builds production websites and AI-citation
      infrastructure for small businesses.

Then generate the surfaces:

aio-surfaces generate site.yaml --out ./public

Output:

public/
├── llms.txt         # markdown summary, llmstxt.org spec
├── aeo.json         # atomic facts (AEO)
├── entity.json      # Schema.org @graph (Org + Person + WebSite)
├── brand.json       # internal brand truth ledger
└── robots-aibots.txt # 12 AI crawler allowlist (append to robots.txt)

Deploy public/ to your site root.

from aio_surfaces import SiteConfig, Service, Fact, render_llms_txt

cfg = SiteConfig(
    site_name="Example Studio",
    site_url="https://example.com",
    tagline="We build things that get cited.",
    description="...",
    services=[
        Service(
            name="Custom Websites",
            url="https://example.com/services/websites/",
            summary="Hand-coded HTML, React, Astro.",
        ),
    ],
    facts=[
        Fact(
            id="f-1",
            question="What does Example Studio do?",
            answer="Example Studio builds...",
        ),
    ],
    ein="12-3456789",
    orcid="0000-0000-0000-0000",
)

print(render_llms_txt(cfg))

1. Atomic facts > paragraphs. AI engines cite spans, not essays. The Fact dataclass enforces a 1000-char hard ceiling and recommends < 500.
2. Single source of truth. All four surfaces render from one config so they can't drift apart.
3. No magic. No analytics, no telemetry, no remote calls. Pure-Python stdlib + optional PyYAML.
4. Run anywhere. Works in CI, in a Makefile, as a pre-commit hook, or as a one-off CLI invocation.

• Not a CMS. It generates static files from a config; you deploy them.
• Not a Schema.org validator. (Run the output through validator.schema.org yourself.)
• Not an opinionated framework. Generate what you need; ignore the rest.

• aio-surfaces validate — round-trip the output through Schema.org validator
• aio-surfaces diff — show what changed since the last generation (CI hook)
• Per-page aeo.json support (currently site-wide only)
• llms-full.txt expanded variant generation
• Hugo / Astro / Next.js plugin packages

Issues + PRs welcome. The codebase is small (~350 LOC) and has full test coverage:

pip install -e '.[test]'
pytest

MIT © 2026 Joseph W. Anady. See LICENSE.

Built and used in production by ThatDevPro — SDVOSB-certified veteran-owned web + AI engineering studio. The generators here are the same code that runs across ThatDeveloperGuy.com, ThatDevPro.com, and 130+ client sites on ThatWebHostingGuy.com.