ContractParser.ai

Frequently Asked Questions

Short answers to the things customers ask before first use.

On this page
  • Supported file formats
  • Size and batch limits
  • Data retention
  • Languages
  • Pricing and charges
  • Quick vs. Verified
  • Source citations
  • Auto-suggest fields
  • Amendment grouping
  • AI training on customer data
  • Failures and cancellation
  • Payment and receipts
  • Cloud integrations

Supported file formats

What file types can I upload?

Documents: PDF, DOCX, DOC, RTF, TXT, HTML, CSV, XML, JSON, Markdown. Images: PNG, JPG, GIF, WebP. Archives: ZIP (we extract and analyze each supported file inside). Native-text PDFs extract with the highest quality; scanned PDFs go through AI vision OCR.

Can I upload a folder of contracts?

Yes — drag-and-drop a folder directly from your desktop, or ZIP it first. Either works. ZIPs of ZIPs are not recursively unpacked.

Size and batch limits

How big can a single file be?

TypeMax sizeWhy
PDF32 MBAI provider hard limit; also up to 100 pages per PDF
Image (PNG, JPG, etc.)5 MBAI provider hard limit
Word, text, HTML10 MBOur sanity cap; realistic contracts are well under this
ZIP archive5 GBTotal upload size

How many documents per upload?

Up to 1,000 documents per batch, across drag-drop and ZIP contents combined. For larger volumes, split into multiple uploads. Total upload size is capped at 5 GB per request.

What if I hit the limit?

We accept what fits, tell you the batch was truncated, and you can upload the rest as a separate batch. You're only charged for the documents that were analyzed.

Data retention

How long do you keep my uploaded documents?

Uploaded documents are automatically deleted within 2 hours after processing completes. This is strict — contracts are the sensitive asset and we don't want to be responsible for them any longer than necessary.

How long are results available?

Extracted results (the CSV and the on-site results view) are downloadable for 24 hours. If you provided an email, the CSV is also attached to that email — your inbox is the durable copy, so you never lose access.

Check your junk / spam folder if you don't see it. ContractParser is a new sender domain, and some mail filters (Outlook.com especially) route new-domain mail to junk for a few weeks until reputation is established. Marking the email as "Not Junk" helps — both for you and for every other customer after you.

Do you use my data for AI training?

No. Ever. Our AI provider runs under a commercial agreement that prohibits training on customer data. Your documents and results are never used to improve any AI model.

Languages

Can I analyze non-English contracts?

Yes — we support most world languages including Spanish, French, German, Japanese, Chinese, Korean, Arabic, and many more. Extracted values come back in the source language of the document. Full details on the Languages page.

Pricing and charges

Quick answers below. The full breakdown is on the pricing page.

How is pricing calculated?

Pay per page. PDF pages are counted directly. Other formats (DOCX, text, HTML) are estimated at roughly 3,000 characters per page.

Two minimum-charge tiers apply, so you can cheaply iterate on small samples before committing to a large batch:

  • $0.50 minimum for small batches of 10 pages or fewer — meant for iteration on a sample before a big run.
  • $2.00 minimum for anything larger, to cover small-volume overhead.

Example costs

BatchVerified ($0.15/page)Quick ($0.10/page)
One 2-page contract (sample iteration)$0.50 (small-batch min)$0.50 (small-batch min)
One 15-page contract$2.25$2.00 (min)
One 50-page contract$7.50$5.00
50 × 20-page contracts (1,000 pages)$150.00$100.00
500 × 30-page contracts (15,000 pages)$2,250.00$1,500.00

When am I charged?

Your card is authorized (funds placed on hold) at checkout. The actual charge happens after processing completes — and we only capture the amount for documents that were successfully analyzed. Unsuccessful documents are not charged.

Can I tune my fields or prompt and re-run the same documents?

Yes. After a run completes you'll see a Tune & Re-run button on the results page. Change any combination of fields, custom prompt, or tier (Quick / Verified) and run again against the same uploaded documents — no re-upload needed. Each run is a separate charge, but at the small-batch rate ($0.50 minimum for ≤ 10 pages) iteration is cheap. Multiple runs are kept side-by-side on the results page so you can flip between them, and a Compare button shows a merged table highlighting the fields that differ.

Quick vs. Verified

What's the difference?

Verified ($0.15/page, the default): full extraction using Claude Opus 4.7, plus a second Opus pass that audits every field across the document. The audit explicitly checks cross-field arithmetic (does total contract value reconcile with rate × duration?), date math (does start + duration equal stated end?), cross-reference validity (do referenced sections actually exist?), self-contradictions within a single clause, and — when citations are on — whether each cited quote actually supports its value. Flags use hedged language and quote the conflicting numbers verbatim.

Quick ($0.10/page): Claude Sonnet 4.6 single-pass extraction, no audit. Source citations are still available as an opt-in. Great for bulk analysis and straightforward contracts where individual errors aren't consequential.

Three things set our Verified tier apart from what similar tools offer:

  • Included at the tier price, not an add-on. Some parsers offer numeric confidence scores but charge an extra per-page credit for them. Verified bakes the entire audit pass into the $0.15 rate, including citation validation when citations are enabled.
  • Cross-field reasoning, not just per-field confidence. The audit checks the math across values. Example flag: "totalValue $220,000 appears inconsistent with rateOrPricing $8,500/mo × 12 + $9,000/mo × 24 = $318,000 over the 3-year Initial Term, which does not appear to reconcile with the $220,000 cap." Confidence scores can't tell you that.
  • Narrative reasoning with quoted evidence. Instead of a bare "confidence: 0.67," you get the actual concern in plain English, with every conflicting number quoted verbatim so a reviewer can spot-check in seconds.

General-purpose chatbots (ChatGPT, Claude, Gemini) can be prompted to self-check, but they won't return structured, per-field audit results you can drop into a spreadsheet or workflow.

Which should I pick?

Verified is the default and the right call for most contract work. Quick is there for bulk runs on simple documents where individual errors aren't consequential. You can switch per batch.

Source citations

What are source citations?

An opt-in feature (checkbox in step 2 of the upload flow). When enabled, every extracted value comes back with the page number it was found on and the verbatim quote from the document that supports it. On the results page, each cell shows a small p.N chip; click the chip to open a popover with the full quote and a copy-to-clipboard button. In the CSV, each field gets two extra columns alongside the value: <Field> (page) and <Field> (source quote).

Why turn citations on?

  • Auditable extraction. A value without a citation is the AI's claim; a value with a citation is a verifiable assertion. Critical for legal review, M&A diligence, regulated workflows, or anything where a wrong number has cost.
  • Hallucination detection (on Verified). The audit pass cross-checks whether each cited quote actually contains or implies the extracted value. If the AI fabricated an answer that isn't supported by the cited text, the audit flags it.
  • Source-of-truth on long contracts. On a 60-page master agreement, knowing that "total value: $4.2M" came from page 17, section 3.2, saves your reviewer the page-flip.

Are citations always available?

Yes, on both Quick and Verified tiers. There's no per-page surcharge. The CSV gets wider when citations are on (roughly triples the column count), which is why it's opt-in — if you want a slim CSV for downstream automation, leave citations off.

What about derived values like the Executive Summary?

Some fields are synthesized across the whole document and don't have a single supporting quote — for example, an executive summary or a computed contract status. For those, the page and quote columns are intentionally left blank. The audit pass treats this as expected, not as a flag-worthy issue.

Auto-suggest fields

What does “Auto-suggest fields” do?

After you've uploaded files but before you run extraction, click the ✨ Auto-suggest fields button (top-right of the field-selection step). ContractParser reads your first uploaded document, identifies what kind of contract it is, and recommends a tailored set of fields for that contract type. It also proposes additions to your custom prompt for anything specific to the document.

How is this different from just clicking the defaults?

The default field selection is broad — what most contract reviewers want, across all contract types. Auto-suggest is narrower and document-aware. For an NDA, it'll skip fields like “Total Contract Value” and “Insurance Requirements” that don't apply, and emphasize confidentiality, term, and remedies. For an MSA, it'll include indemnification, warranty, SLA, and liability cap. For a lease, it'll prioritize parties, dates, rent, and insurance while skipping non-compete and IP ownership.

Does it cost extra?

No. The pre-flight suggestion is one quick call to a smaller model (Claude Sonnet) and the cost is absorbed — we don't charge for the suggestion itself, only for the pages you actually run.

What if it suggests fields I don't want?

The suggested set is rendered as a banner with field chips above your normal checklist. Click Apply suggestions to replace your current selection with the AI's recommendation, or Dismiss to keep what you have. You can also manually tick or untick individual fields after applying.

Amendment grouping

What is amendment grouping?

If your batch contains an original contract along with one or more amendments or addenda that modify it, ContractParser detects the relationship and groups them together. The results page gets a Group by parent view that shows each original contract with its amendments indented underneath, sorted by amendment sequence (Amendment 1, Amendment 2, etc.).

How do I turn it on?

It's on by default. Three fields under Amendment & Hierarchy — Document Type, Modifies Contract, and Amendment Sequence — are pre-checked in the field-selection panel. After a run completes, if any documents reference another contract, a Group by parent button appears alongside Tune & Re-run.

What if the parent contract isn't in my batch?

The amendment still gets grouped. It just renders under a "Referenced" header showing the parent contract's identifier (extracted from the amendment text). Useful when you're processing only amendments to a known master agreement.

Can I export the hierarchy?

Yes. Downloading the CSV from a run with grouping data pops a choice: Flat (one row per document, the standard layout) or Indented hierarchy (parents and amendments grouped, with a Hierarchy column showing the structure for use in Excel pivot tables and similar).

What about Master Services Agreements with SOWs and SOW amendments?

Today's grouping is two-level (parent + children). It's accurate for the typical case where amendments reference the original agreement directly. Deeper chains like MSA → SOW → SOW amendment are flattened — the SOW is treated either as a child of the MSA or as its own root, depending on how the documents reference each other. If you have a case where deeper nesting is important, let us know — we have a path to extend this.

AI training on customer data

Are my documents used to train AI?

No. See retention and our Privacy Policy for details.

Where does my document go during analysis?

Your document is sent over encrypted HTTPS to our AI provider's API, analyzed, and the result is returned. Under the commercial API agreement we use, the provider is contractually prohibited from retaining or training on your data.

Failures and cancellation

What if a document fails to analyze?

You're not charged for it. The results CSV marks it with an error note. Common causes: corrupt PDF, password-protected file, heavily degraded scan.

Can I cancel a batch mid-processing?

Yes. Click Cancel during processing. Documents already analyzed are kept and delivered; you're charged only for those. No refund is needed — the unprocessed portion is never captured.

What if everything fails?

If zero documents analyze successfully, the card authorization is released and you pay nothing.

Payment and receipts

Who processes my card?

All payments are processed by Stripe. We never see or store your card number, CVV, or full card details. We receive only a transaction confirmation.

Do I get a receipt?

Yes. Stripe sends an email receipt to whatever email you enter at checkout. Separately, if you provided an email in ContractParser, you also get the results email with the CSV attached.

Can I get a refund?

Because of how authorize-and-capture works, the normal case never needs a refund — you're only charged for successful documents. If something goes wrong anyway, email support@grovestreams.com and we'll handle it.

Cloud integrations

Can I import contracts from Salesforce?

Yes. Click the Salesforce button on the upload page and log into your Salesforce account. You'll see your Contract records and can pick PDFs attached to them. You log in with your own credentials; we never see your Salesforce password and can only access what you explicitly select.

What about Google Drive, OneDrive, Box, Dropbox?

Planned, not enabled yet. For now, drag-drop from your desktop or upload a ZIP. When those integrations ship, you'll see matching buttons on the upload page.

Still have questions?

Email support@grovestreams.com. We reply within one business day. Support is handled by the GroveStreams team — same people, same company, shared infrastructure.

ContractParser.ai · About · FAQ · Compare · Use cases · Pricing · Terms · Privacy · Languages

A product of GroveStreams LLC · Operating production SaaS since 2011