Reference

Export formats

Every project can be exported in 14 formats. Field names below are exact and stable. Q&A exports can be filtered by minimum confidence, and personal data can be masked on export.

Q&A

Q&A JSONL

qa.jsonl

One question-answer pair per line, extracted from the page text. Each pair keeps the page it came from, the heading context, how it was found and a confidence score.

questionanswercontextsourcetagslangmethodconfidence

Method and confidence: schema_faq comes from FAQPage markup (1.0); heading_question comes from a heading phrased as a question or a collapsible <details> block (0.9); heading_topic pairs a topic heading with the text under it (0.45).

Use it for: evaluation sets, support bots, retrieval tests, or as the source for any of the fine-tuning formats below.

{"question": "How long does a refund take?", "answer": "Refunds are issued to the original card and usually arrive within 5-7 business days.", "context": "Billing > Refunds", "source": "https://acme-docs.com/billing/refunds", "tags": ["billing", "refunds"], "lang": "en", "method": "schema_faq", "confidence": 1.0}
{"question": "Can I change my plan mid-cycle?", "answer": "Yes. Upgrades apply immediately and are prorated; downgrades take effect at the next renewal.", "context": "Billing > Plans", "source": "https://acme-docs.com/billing/plans", "tags": ["billing", "plans"], "lang": "en", "method": "heading_question", "confidence": 0.9}
{"question": "Webhook retries", "answer": "Failed deliveries are retried up to 8 times with exponential backoff over 24 hours.", "context": "API > Webhooks", "source": "https://acme-docs.com/api/webhooks", "tags": ["api", "webhooks"], "lang": "en", "method": "heading_topic", "confidence": 0.45}

Q&A CSV

qa.csv

The same records as a UTF-8 CSV with a header row. tags are joined with a semicolon.

Use it for: reviewing pairs in a spreadsheet, sharing with non-technical teammates, importing into helpdesk tools.

question,answer,context,source,tags,lang,method,confidence
"How long does a refund take?","Refunds are issued to the original card and usually arrive within 5-7 business days.","Billing > Refunds",https://acme-docs.com/billing/refunds,billing;refunds,en,schema_faq,1.0
"Can I change my plan mid-cycle?","Yes. Upgrades apply immediately and are prorated; downgrades take effect at the next renewal.","Billing > Plans",https://acme-docs.com/billing/plans,billing;plans,en,heading_question,0.9

Q&A Markdown

qa.md

A readable document with one section per question and a source link under each answer.

Use it for: reviewing by eye, building an FAQ page, pasting into a knowledge base or a prompt.

## How long does a refund take?

Refunds are issued to the original card and usually arrive within 5-7 business days.

Source: https://acme-docs.com/billing/refunds (confidence 1.0)

## Can I change my plan mid-cycle?

Yes. Upgrades apply immediately and are prorated; downgrades take effect at the next renewal.

Source: https://acme-docs.com/billing/plans (confidence 0.9)

Fine-tuning

Alpaca

alpaca.jsonl

Instruction-tuning format. The question becomes the instruction and the answer becomes the output; input is left empty.

instructioninputoutput

Use it for: supervised fine-tuning with tools that expect Alpaca-style records, such as many LoRA and Axolotl recipes.

{"instruction": "How long does a refund take?", "input": "", "output": "Refunds are issued to the original card and usually arrive within 5-7 business days."}
{"instruction": "Can I change my plan mid-cycle?", "input": "", "output": "Yes. Upgrades apply immediately and are prorated; downgrades take effect at the next renewal."}

ChatML

chatml.jsonl

Chat format with a list of messages per record: one user turn with the question and one assistant turn with the answer.

messagesrolecontent

Use it for: fine-tuning chat models with APIs and trainers that take a messages array.

{"messages": [{"role": "user", "content": "How long does a refund take?"}, {"role": "assistant", "content": "Refunds are issued to the original card and usually arrive within 5-7 business days."}]}
{"messages": [{"role": "user", "content": "Can I change my plan mid-cycle?"}, {"role": "assistant", "content": "Yes. Upgrades apply immediately and are prorated; downgrades take effect at the next renewal."}]}

ShareGPT

sharegpt.jsonl

Conversation format with from set to human or gpt.

conversationsfromvalue

Use it for: trainers and dataset tools that read ShareGPT-style conversations.

{"conversations": [{"from": "human", "value": "How long does a refund take?"}, {"from": "gpt", "value": "Refunds are issued to the original card and usually arrive within 5-7 business days."}]}
{"conversations": [{"from": "human", "value": "Can I change my plan mid-cycle?"}, {"from": "gpt", "value": "Yes. Upgrades apply immediately and are prorated; downgrades take effect at the next renewal."}]}

Retrieval (RAG)

RAG chunks

chunks.jsonl

Clean text split along the page structure into chunks of about 320 words. Each chunk keeps the trail of headings above it, its position on the page and a token estimate.

idurltitleheading_pathtextwordstokens_estpositionlang

Use it for: loading straight into a vector store. Prepend heading_path to text when you embed for better retrieval, and keep url for citations.

{"id": "8f3a91c2e4d0-0002", "url": "https://acme-docs.com/api/webhooks", "title": "Webhooks", "heading_path": ["API", "Webhooks", "Retries"], "text": "If your endpoint returns a non-2xx status or does not respond within 10 seconds, the delivery is marked as failed. Failed deliveries are retried up to 8 times with exponential backoff over 24 hours. Each retry carries the same event id, so make your handler idempotent and ignore events you have already processed.", "words": 56, "tokens_est": 74, "position": 2, "lang": "en"}

Pages JSONL

pages.jsonl

One record per crawled page with metadata, a per-page readiness score and the full cleaned content as Markdown.

urltitledescriptionlangwordsscorefetched_atmarkdown

Use it for: your own chunking pipeline, full-text search, or keeping an archive of what the site said and when.

{"url": "https://acme-docs.com/billing/refunds", "title": "Refunds", "description": "How refunds work, how long they take and how to request one.", "lang": "en", "words": 412, "score": 84, "fetched_at": "2026-11-22T09:14:07Z", "markdown": "# Refunds\n\nRefunds are issued to the original card and usually arrive within 5-7 business days.\n\n## How do I request a refund?\n\nOpen Billing, choose the invoice and click Request refund..."}

Pages Markdown

pages.md

The cleaned content of every page as plain Markdown, each page under its own title with the source URL.

Use it for: reading the site the way a model will see it, or feeding a long-context model directly.

# Refunds

Source: https://acme-docs.com/billing/refunds

Refunds are issued to the original card and usually arrive within 5-7 business days.

## How do I request a refund?

Open Billing, choose the invoice and click Request refund...

---

# Plans

Source: https://acme-docs.com/billing/plans
...

For your site

llms.txt

llms.txt

A short Markdown index of your site following the llms.txt proposal: a title, a one-line summary and links to key pages with descriptions.

Use it for: publishing at the root of your site. It is cheap to add, but whether AI assistants actually read it is still unproven, so treat it as optional.

# Acme Docs

> Documentation for the Acme payments API: billing, refunds, webhooks and SDKs.

## Billing

- [Refunds](https://acme-docs.com/billing/refunds): How refunds work and how long they take.
- [Plans](https://acme-docs.com/billing/plans): Changing, upgrading and cancelling plans.

## API

- [Webhooks](https://acme-docs.com/api/webhooks): Event delivery, signatures and retries.

llms-full.txt

llms-full.txt

The same index followed by the full cleaned text of each listed page.

Use it for: giving a model the whole site in one file, or publishing next to llms.txt.

# Acme Docs

> Documentation for the Acme payments API: billing, refunds, webhooks and SDKs.

## Refunds
URL: https://acme-docs.com/billing/refunds

Refunds are issued to the original card and usually arrive within 5-7 business days.
...

FAQPage JSON-LD

faq.jsonld

Schema.org FAQPage markup built from your high-confidence Q&A pairs. The file is a list with one entry per page: its url and the jsonld block shown below.

Use it for: pasting into a <script type="application/ld+json"> tag on the page that shows those questions. Only mark up questions that are actually visible on that page.

{
  "@context": "https://schema.org",
  "@type": "FAQPage",
  "mainEntity": [
    {
      "@type": "Question",
      "name": "How long does a refund take?",
      "acceptedAnswer": {
        "@type": "Answer",
        "text": "Refunds are issued to the original card and usually arrive within 5-7 business days."
      }
    }
  ]
}

Report and bundle

Report JSON

report.json

The project summary: crawl statistics, the AI-readiness score with each check and its recommendation, or for uploaded files the quality score and row flags.

Use it for: tracking scores over time, building your own dashboard, or attaching to a client report.

{
  "project": {"name": "acme-docs.com", "source": "https://acme-docs.com/", "kind": "site"},
  "stats": {"pages": 212, "qa": 486, "chunks": 1940, "words": 164210},
  "audit": {
    "score": 78,
    "checks": [
      {"id": "ai_bots", "label": "AI crawlers allowed in robots.txt", "status": "pass",
       "value": 1.0, "weight": 14, "detail": "No AI crawler is blocked"},
      {"id": "schema", "label": "Structured data (JSON-LD)", "status": "warn",
       "value": 0.42, "weight": 14, "detail": "42% of pages"},
      {"id": "llms", "label": "llms.txt", "status": "fail",
       "value": 0.0, "weight": 3, "detail": "not found"}
    ],
    "recommendations": ["Add schema.org JSON-LD (Organization, Article, Product, FAQPage)."]
  }
}

ZIP bundle

datasetsdb-<project>.zip

Every format above in one archive, with the same confidence filter and PII masking applied to all files.

Use it for: handing a complete dataset to a teammate or archiving a snapshot of a crawl.

qa.jsonl
qa.csv
qa.md
alpaca.jsonl
chatml.jsonl
sharegpt.jsonl
chunks.jsonl
pages.jsonl
pages.md
llms.txt
llms-full.txt
faq.jsonld
report.json

Try it on your own site

Free for up to 300 pages per site. No card needed.