Make your product
legible to AI agents.

Kamil Choiński · KC-IT — llms.txt · schema.org JSON-LD · AI-crawler posture · 13-check audit

People are starting to find software by asking an assistant instead of a search box. Assistants don't run your JavaScript and they don't guess — they read llms.txt, structured data and robots.txt, then cite whoever made the answer easy to extract. I do that work on your own domain, and hand you a score you can re-run yourself, for free, any time.

sites audited

12every site I run, swept in one pass on 2026-07-28

average score

68.6 → 97.5out of 100, on the same 13 checks

perfect scores

9 / 12at 100/100 after the sweep; none before

re-audit cost

$0.00the grade is a public endpoint — verify me

Four artefacts, and the score that proves they landed

This is not a dashboard subscription. It is a set of files on your origin, written for the way agents actually read a site — plus an audit you own and can re-run without me.

llms.txt and llms-full.txt

The agent-facing index of your site: an H1 with your name, a one-paragraph summary an assistant will quote almost verbatim, then the pages worth reading, linked and described. llms-full.txt inlines the content so an agent gets everything in a single fetch instead of crawling you.

schema.org JSON-LD

Typed facts instead of prose an extractor has to interpret — Organization, Product, Service, Offer, FAQPage, wired together in one @graph with stable @ids. This is the single heaviest check in the audit, and the one most sites fail outright.

robots.txt and AI-crawler posture

A deliberate position on GPTBot, ClaudeBot, PerplexityBot, CCBot, Google-Extended and the rest — allowed where you want to be cited, blocked where you don't, and nothing left ambiguous for a crawler to resolve as "no". Includes the edge-level blocks your CDN may be injecting above your file without telling you.

The audit, and the head that passes it

13 weighted checks — llms.txt shape, sitemap, agent card, HTTPS, and the per-page work: title, meta description, canonical, Open Graph, machine-readable alternates. You get the A–F grade, the failures by name, and for each one the fix and a paste-ready snippet.

audit_onlyThe 13 checks, the grade and the ranked fixes — you implement.

done_for_youI write and ship the files, then re-score until the grade is where it should be.

re_scoreFree and public, at index.kc-it.pl — no lock-in, nothing to cancel.

I ran it on everything I own first

Twelve deployed sites — games, shops, scrapers, a Slack app, the registry itself — audited on the same 13 checks, fixed, and audited again. This is the whole fleet, not the flattering half.

★ before / after · 12 sites · same 13 checksaverage 68.6 → 97.5

site	before	after
index.kc-it.pl	A 100	A 100
kc-it.pl	E 48	A 100
110kc3.github.io	C 79	A 100
…/polish-sweepstakes	A 93	A 100
…/rentgen-ofert	C 78	A 100
…/bankier-street-bets	C 73	A 100
przetargimiejskie.pl	D 69	A 100
…/puzle	D 67	A 100
…/pool-gliwice-automation	D 60	A 100
stareaparaty.com	E 54	A 90
protocolindex.eu	D 64	A 90
overtimelog.com	F 38	A 90
average	68.6	97.5

// the three sites at 90 fail exactly one check: their CDN injects a managed AI-crawler block into robots.txt
// above the file in the repo. That is a dashboard setting, not a commit — reported as such rather than papered over.

Two things that sweep taught me, and that you are paying me to already know:

Unmeasured sites are the worst sites. Three of the twelve were invisible to my own collector — Cloudflare Pages projects with nothing in the repo naming the domain — and those same three were the three lowest scores in the fleet. Anything nobody audits drifts.

Most origin-level checks are inherited, so the per-page work is the <head>. llms.txt, robots.txt, sitemap.xml and /.well-known/agent.json resolve against the origin — fix them once and every page on the domain benefits. What is left per page is title, meta description, canonical, Open Graph and JSON-LD. Knowing which is which is the difference between an afternoon and a fortnight.

Start with the free grade — it is the same audit I sell

There is no lead-capture form and nothing to install. Paste a URL, get a letter grade and the name of every failing check. Then decide whether you want the fixes or want me to ship them.

$0 · instant scorePaste your URL into the box at index.kc-it.pl. A–F grade, 13 checks named pass or fail, no signup.

$0.05 · per-call APIFor agents: POST /api/audit, paid per call over x402 in USDC on Base. Full per-check detail, ranked fixes and paste-ready snippets. Machine-payable — no account, no invoice.

quoted · done for youEmail [email protected] or open a [hire] issue on the registry repo. I quote per site against your current score.

On price. The AI-visibility tools people already subscribe to run roughly $29–489 per month for dashboards that estimate how often a model mentions your brand. This is the concrete version of that idea: the files that make you extractable, written into your repo, and a grade anybody — including you, including a prospect's agent — can verify in a single request. When the number stops moving you stop paying.

On dogfooding. Every claim here is checkable. The audit endpoint is public, the registry that hosts it is open source at github.com/110kc3/seo, and the payment rail behind the per-call audit settled its first real on-chain payment on 2026-07-29. Run the score on this page before you email me.

Make your productlegible to AI agents.