Agent-ready. Verified. Kept that way.
Sichta Web tests whether AI agents and assistants can read your website, generates the files that fix what fails, publishes them for you, and tells you when your site drifts back out of shape.
19 deterministic checks · fix · verify · monitor drift
Free, no account. Every finding with its fix.
What Sichta Web is
Sichta Web is a subscription service for website owners. It checks whether AI agents and assistants — ChatGPT, Claude, Perplexity, Gemini and the browsers acting on their behalf — can find, read and correctly quote your website.
It runs 19 deterministic checks against your site's machine-readable surface, generates the files that fix what fails from your site's own content, serves those files for you, and re-checks daily so you hear about it when something drifts.
You sign in with an email link or a Google account, add the site you own, and prove you control it. Nothing is published to your site without your say-so.
How it stays fixed
Sichta Web checks your site against what AI agents actually need, fixes what fails, verifies the fix is live, and keeps watching.
check
19 deterministic checks across four pillars. Every failure has a typed error and a fix — never a score you have to interpret.
fix
We generate llms.txt, llms-full.txt, JSON-LD and security.txt from your own pages, and serve them for you.
Generated from your content. Anything we can't source, you approve — we never invent.
verify
After every change, the checks run again against the live site. Fixed means verified, not deployed.
monitor
A daily check. If something that passed starts failing, you hear about it once — not every day.
The 19 checks
This is the actual registry the engine runs — not a marketing summary of it.
Six requests. That is the whole surface — the checks are what we read in the answers.
ADiscovery surface
Whether an agent can find the map of your site: llms.txt exists, parses, and its links resolve.
| C-LLMS-01 | llms.txt exists |
| C-LLMS-02 | llms.txt structure is valid |
| C-LLMS-03 | llms.txt links resolve |
| C-LLMS-04 | llms-full.txt exists when referenced |
| C-SITE-01 | sitemap.xml exists and parses |
| C-SITE-02 | llms.txt is not blocked by robots.txt |
BAccess policy
Whether AI crawlers are actually let in: robots.txt parses, treats them consistently, and your CDN doesn't refuse them.
| C-BOT-01 | robots.txt parses |
| C-BOT-02 | AI crawler policy is consistent |
| C-BOT-03 | AI user-agents are not refused |
CStructured data
Whether a machine can read what your business is: valid JSON-LD, a complete organisation entity, resolving sameAs links.
| C-SD-01 | JSON-LD present and parses |
| C-SD-02 | Organisation entity is complete |
| C-SD-03 | sameAs links resolve |
| C-SD-04 | Schema type matches the page |
DMachine endpoints
Whether the newer agent surfaces answer honestly: /.well-known/, UCP manifest if you publish one, markdown negotiation.
| C-WK-01 | /.well-known/ is reachable |
| C-WK-02 | UCP manifest is valid |
| C-WK-03 | security.txt present |
| C-MD-01 | Content negotiation for markdown |
| C-FWD-01 | MCP server card is valid |
| C-FWD-02 | WebMCP tools declared |
What a finding looks like:
C-SD-01 FAIL · NO_JSONLD
JSON-LD is how a machine reads what your business is without interpreting your layout.
Add a <script type="application/ld+json"> block to the homepage and check that it is valid JSON.
Managed serving
Set once, never stale: your conformance files are served from our edge and regenerated when your site changes.
Your files keep being served, unchanged, for 30 days after the subscription ends, and you get an email containing the static bundle so you can host them yourself. After that the managed routes return 404 and fall through to your own site. Nothing else about your site is touched.
The moving target
Agent-readiness is not a file you generate once — the target moves on three fronts.
2026-01-11
UCP announced by Google and Shopify — manifest version 2026-01
2026-01-28
Chrome auto browse preview: Gemini agents start browsing real sites
2026-05-07
Lighthouse 13.3 ships the agentic-browsing audit — llms.txt, WebMCP, CLS
- today
Your site changes
New pages, renamed products, a redesign — every deploy can silently break what passed last month.
2026-05-07 Lighthouse 13.3 ships the agentic-browsing audit — llms.txt, WebMCP, CLS
The agents change
New crawlers, new user-agents, new browsing behaviours ship monthly. What agents read tomorrow is not what they read today.
2026-01-28 Chrome auto browse preview: Gemini agents start browsing real sites
The standards change
llms.txt, UCP, WebMCP — the conventions themselves are still moving. The checks move with them.
2026-01-11 UCP announced by Google and Shopify — manifest version 2026-01
Proof, not a promise
Every verified site gets a public badge that reads "verified until" a date, and a page behind it showing the score and what was checked. Each passing day's check rolls the date forward; a fortnight without one lets it go.
Lapses if verification stops. That's the point.
Embed it:
<a href="https://web.sichta.gigliotti.software/verify/cmt2wm3xc0004p222xky8deiq">
<img src="https://web.sichta.gigliotti.software/verify/cmt2wm3xc0004p222xky8deiq.svg"
alt="Sichta verified" width="200" height="44" />
</a>Callable from where you already work
The free scan is an MCP server as well as a page. Ask an assistant whether agents can read a site and it answers with the findings and the fixes — no account, no key, nothing stored.
In Claude Code
claude mcp add --transport http sichta https://web.sichta.gigliotti.software/api/mcp
Also machine-readable
- One tool, scan_site — a hostname in, every finding and its fix out.
- A server card at /.well-known/mcp.json, for clients that discover rather than get told.
- An OpenAPI document for the Scan API, generated from the same registry the engine runs.
- Every public page serves its markdown source to a request that asks for it.
Pricing
Monitor
€15
per site / month
Know the state of your site, every day, with proof you can show.
Conformance
€29
per site / month
The whole lifecycle: checked, fixed, served, verified, watched.
One-time fix
€149
per site, once
One fix cycle, no subscription. For a site that changes rarely.
Agency
€23
per site / month
Conformance for a portfolio, with the report in your own name.
Questions
The ones we are asked most, answered plainly.
Can I run the check from Claude or Cursor?
Yes. The free scan is an MCP server: add https://web.sichta.gigliotti.software/api/mcp to any MCP client and ask it about a site. It needs no account and no key, it stores nothing about the site it checks, and it returns the same findings and fixes the website shows.
How is this different from llms.txt generators?
File generators keep one file fresh; Sichta verifies the whole agent surface and proves it. llms.txt is one of 19 checks — access policy, structured data and machine endpoints are the rest, and the badge shows the whole surface passing.
Do you guarantee AI visibility or rankings?
No, and nobody honestly can. Sichta guarantees something checkable instead: that the machine-readable surface of your site exists, is valid, and stays that way — with a public badge as proof.
Does the free scan really show everything?
Yes. Every finding, its explanation and its fix description are free — diagnosis is free, delivery is paid. The subscription generates the files, serves them, verifies them and monitors drift.
Do you write my content with AI?
Descriptions are drafted from your own pages and flagged for your review; nothing deploys unapproved. Where there is no source content at all, you get an explicit TODO instead of an invented sentence.
What happens when I cancel?
Your files keep being served, unchanged, for 30 days, and you get the static bundle by email so you can host them yourself. After that the managed routes return 404 and fall through to your own site.
Anything else — hello@gigliotti.software.