AI Citability Audit

Summary

The site is unreachable to answer engines for two independent reasons — and neither one shows up in a browser, and neither one shows up in a search-ranking report.

Net: no crawler that does not execute JavaScript can read either page. The content itself is fine — it is simply never delivered.

Full report

AI CITABILITY AUDIT
========================================================================
source: ./fixtures/bad  (captured snapshot)

HIGH=2  MEDIUM=9  LOW=3  INFO=1

[HIGH] (access) R-BLOCK — AI crawlers are blocked outright: amazonbot, anthropic-ai, applebot-extended, bytespider, ccbot, chatgpt-user, claude-web, claudebot
    robots.txt disallows the whole site for one or more AI crawlers. A
    blocked crawler cannot cite you no matter how good the page is. This
    is very often accidental: a wildcard rule written for a different
    purpose now covers them.
    FIX: Inspect the rule, decide whether blocking is intentional, and
    if not, add an explicit Allow group for the crawlers you want.
    Blocking is a legitimate choice — but make it a choice.

[HIGH] (readability) P-JS — Content looks JavaScript-rendered (page.html)
    The raw HTML carries 0 characters of text against 10622 characters
    of script. A crawler that does not execute JavaScript receives
    essentially an empty page. This is the single most common reason a
    site that ranks well in search is never cited by an answer engine:
    search engines render your JS, most answer-engine crawlers do not.
    FIX: Serve the primary content in the initial HTML (server-side
    render, static generation, or a pre-rendered fallback). Then verify
    by fetching the page with JS disabled and reading what comes back —
    not by looking at it in your browser, where it always looks fine.

[MEDIUM] (discovery) L-MISSING — No llms.txt
    llms.txt is a proposed convention: a single markdown file at your
    root that tells a language model what your site is and which pages
    matter. Adoption is uneven, so treat it as a cheap, low-risk
    addition rather than a requirement.
    FIX: Publish /llms.txt: an H1 with your site name, a one-line
    summary as a blockquote, then a markdown list of your most important
    pages with a short description each. Keep it curated, not a full
    dump.

[MEDIUM] (page) P-NODESC — No meta description (page.html)
    A page with no description leaves the summary to be guessed from
    whatever text the crawler can find.
    FIX: Add a one-sentence description that states what the page
    establishes, not what the company is.

[MEDIUM] (page) P-NOTITLE — Missing <title> (page.html)
    The title is the strongest short signal a crawler has about what the
    page answers.
    FIX: Write a title that names the question the page answers.

[MEDIUM] (quotability) P-NOJSONLD — No structured data (page.html)
    Structured data is the machine-readable version of your claims.
    Without it, a crawler has to infer that a number is a price, a fact
    is a fact, and an answer is an answer.
    FIX: Add JSON-LD: Article or FAQPage where it fits, Organization on
    the home page. Describe what is genuinely there.

[MEDIUM] (quotability) Q-FEWSPECIFICS — Almost no verifiable specifics (js_partial.html)
    Found 0 numbers and 0 dates across 247 characters. An answer engine
    cites specifics — a figure, a date, a named result. Pages of
    well-argued generalities have nothing quotable in them.
    FIX: Add the concrete figures you already know: how many, how long,
    how much, compared with what, as of when.

[MEDIUM] (quotability) Q-FEWSPECIFICS — Almost no verifiable specifics (page.html)
    Found 0 numbers and 0 dates across 0 characters. An answer engine
    cites specifics — a figure, a date, a named result. Pages of
    well-argued generalities have nothing quotable in them.
    FIX: Add the concrete figures you already know: how many, how long,
    how much, compared with what, as of when.

[MEDIUM] (quotability) Q-THIN — Thin content (js_partial.html: 247 characters)
    Only 247 characters of text were extractable. Answer engines prefer
    sources they can quote from at length; a thin page gives them
    nothing to lift.
    FIX: Either deepen the page with specifics or let a stronger page
    carry the topic.

[MEDIUM] (quotability) Q-THIN — Thin content (page.html: 0 characters)
    Only 0 characters of text were extractable. Answer engines prefer
    sources they can quote from at length; a thin page gives them
    nothing to lift.
    FIX: Either deepen the page with specifics or let a stronger page
    carry the topic.

[MEDIUM] (readability) P-JSRENDER — Part of the content is assembled in JavaScript (js_partial.html)
    About 848 characters of prose appear only inside script blocks and
    never in the HTML itself, against 247 characters of visible text.
    The page is not an empty shell, so a shell check passes — but a
    crawler that does not run JS receives the frame without the
    substance. This is the quiet version of the JavaScript problem, and
    it is harder to notice because the page looks complete in a browser.
    FIX: Move the substance into the served HTML — render it at build
    time or server-side, and keep JavaScript for behaviour only. Verify
    by fetching the page with JS disabled and reading the raw response,
    not by opening it in a browser.

[LOW] (access) R-NOSITEMAP — robots.txt declares no sitemap
    Without a Sitemap: line, discovery depends on links alone. Crawlers
    that see fewer links see less of your site.
    FIX: Add `Sitemap: https://your-domain/sitemap.xml`.

[LOW] (page) P-NOCANON — No canonical link (page.html)
    Without a canonical, duplicates across http/https, trailing slash
    and query parameters compete as separate sources.
    FIX: Add a canonical link to the preferred URL.

[LOW] (page) P-NOH1 — No H1 (page.html)
    A heading hierarchy tells a crawler which text is the document's
    claim and which is support.
    FIX: Add exactly one H1 stating the page's subject.

[INFO] (quotability) Q-NOQUESTIONS — No question-shaped headings (page.html)
    Headings phrased as the user's actual question are the cheapest
    match between a page and an AI query.
    FIX: Where it reads naturally, phrase some H2/H3s as questions.

========================================================================
This audit covers whether an answer engine CAN read and cite the
page. It does not measure whether they DO. That needs probing real
answers and counting mentions — a different job.

What to do, in order

Fix robots.txt first. It is a one-line change and everything else is blocked behind it. Add explicit Allow groups for the crawlers you want, plus a Sitemap: line.
2.

Move the substance into the served HTML — render at build time or server-side. Then verify by fetching the page with JavaScript disabled and reading the raw response, never by opening it in a browser.
3.

Open with the answer. State the conclusion in the first two hundred characters instead of building up to it, and put one number or date in roughly every 500 characters, so a quotation carries weight.
4.

Publish a curated /llms.txt — an H1, a one-line summary, and a short list of the pages worth reading. Cheap and low-risk.
5. Re-run the audit and watch the counts fall. It exits 1 while findings remain, so it drops straight into CI.

Next steps

• Re-run after each fix; the report is deterministic, so the diff is meaningful
• Add --json to keep a dated record and track the trend
• Ask for a second pass on the pages that matter most commercially, not just the homepage