Cutting Cards: Snipe, Automatic & Agent

The three card-cutting modes, what each is for, and how to pick the right one.

DebateCardAI gives you two places to cut cards: Single Card Snipe for one card at a time, and Batch Input for a whole stack. Batch Input then runs in one of two styles, Automatic or Agent. All of them produce the same kind of card from the same AI. They differ only in how many cards you cut at once and how hands-on you stay.

If you just want to cut one card right now, follow the quickstart. This page is the full reference for all three.

Which one should I use?

Single Card Snipe Automatic (Batch) Agent (Batch)
Best for One card A handful (about 3 to 10) you want to check A big stack (10 or more)
What you do Paste one source, generate Add cards, then generate and approve each one Add cards, select them, hit Start, walk away
Control Full Full, one card at a time Hands-off; review at the end
How you reach it Its own tab Batch Input, then Process Cards Batch Input, then Process Cards

Rule of thumb: one source is Snipe; a handful you want to vet is Automatic; a pile you want cut while you do something else is Agent. Automatic and Agent work on the exact same queue you build in Batch Input. The only choice is whether you approve each card yourself (Automatic) or let the app run them all (Agent).

Single Card Snipe

The one-card mode, and the default tab when you open the app. Paste a Source URL, a User Claim, and the Source Text (or click Fetch to pull the text from the URL), then click Generate Card. The finished card lands in the editor on the right, where you can edit it, Copy it (Word- and Docs-ready), or Save to Library.

Two things are unique to Single Card Snipe:

  • Custom tagline. Expand it under the claim to write the exact tag yourself. The other modes do not offer this. Whatever you type also steers what the AI highlights.
  • Add to Case. This drops the current card into a Case Collection list right below the editor, which has its own Copy All button and drag-to-reorder. This is a scratch list inside the tab; it is not the same as saving to your Library.

Full step-by-step is in the quickstart.

Batch Input

Use this when you have more than one card to cut. You build a queue first, then process it.

  1. Open Batch Input (under the Card Cutting menu on desktop, or the More sheet on mobile).
  2. Add a card: fill in Source URL, Claim / Argument, and Source Text (the Fetch button works here too). Card title is optional and defaults to your claim. Click to add it to the queue.
  3. Repeat for every card you want to cut. The queue holds all of them.
  4. Click Process Cards. You will see a "Choose Your Card Cutting Style" screen with two options: Automatic Mode and Agent Mode.

Automatic Mode (approve each card)

Marked Recommended, and best when you want to read every card before keeping it.

  1. From the style screen, click Automatic Mode.
  2. Your queue is on the left. Click a card to make it active.
  3. Expand Source Details on the right to review or fix that card's URL, claim, or source text.
  4. Click Generate Card, edit the result if needed, then click Save & Proceed. The card is saved to the collection, the queue item is marked done, and the next card loads automatically.
  5. Repeat down the queue. Finished cards collect below, with a Copy All button.

Agent Mode (cut them all at once)

Marked High Volume, and best when you trust your inputs and want to step away.

  1. From the style screen, click Agent Mode.
  2. Each queue card has a checkbox. Tick the ones you want, or use select all.
  3. Click Start. The app cuts each selected card in order, on its own, showing live status and a per-card check or error mark as it goes.
  4. You can Pause and Resume anytime, and you can even add more cards mid-run; the agent picks them up.
  5. When it finishes, review the whole batch and Copy All.

Card settings: control how every card is cut

Open Settings (the gear in the top nav). The Card tab holds the controls that shape every card you cut, in every mode. Set these once:

  • Cutting Mode. Lay highlights words that read as a smooth, grammatical sentence when spoken aloud (it may add small bracketed connectors). Tech packs in as much as possible for spreading (very fast reading), and the highlights do not need to flow. The default is Lay.
  • Highlight Density. How much of the card gets highlighted: Sparse (just the core warrants, roughly 15 to 25 percent), Balanced (the default, roughly 30 to 40 percent), or Dense (warrants plus context, roughly 45 to 60 percent).
  • Highlight Color. Yellow, Purple, Cyan, or Green. The default is Yellow. The color survives the copy into Word or Google Docs.
  • Card Signature. Your initials, added to the end of every citation. You can also set this inline from the nav header.
  • Font sizes. Separate sizes for the tag, body (the shrunken context text), author, and citation.

The Advanced tab adds:

  • H4 Tag Headers. Wraps each tagline as a heading so your tags show up in Word's navigation pane. Off by default.
  • Paragraph Integrity. Keeps the source's original paragraph breaks instead of collapsing the card into one block.
  • Token Display. Shows the input and output token counts under each generated card. Off by default.

Every one of these is global: it applies to Single Card Snipe, Automatic, and Agent alike, and it follows your account across devices.

Improve Card (Recut)

After a card is cut, you can refine how it looks without changing the words. Select text in the card editor and choose Highlight less, Highlight more, Retag, or enter a custom instruction like "make this punchier" or "emphasize the warrants." The AI reworks only formatting and highlighting; it never adds, deletes, or rewrites card text. Server-side guards ensure your source words stay exactly the same, just formatted to fit your needs better.

Your first Improve on any card is free. Additional improves on the same card count against your daily cut limit. (A "card" resets whenever you cut a new one into the editor.)

What happens under the hood

When you generate a card, the app sends your claim, source URL, source text, and your settings to the cut-card function, which runs Google's Gemini 3 Flash model to read the source, pick the strongest passage, write the tag and citation, and highlight it. A card usually takes about 15 to 45 seconds, depending on how long the source is.

One detail worth knowing: this step does not fall back to a second model if Gemini is overloaded; it retries the same model a few times and then reports an error. So an occasional "the model is busy" message means try again in a moment, not that anything is broken.

Limits and cost

  • One generated card is one card cut, in every mode. A batch of ten cards in Automatic or Agent is ten cuts. There is no batch discount.
  • Free: 5 cuts per day (a rolling 24-hour window, not a midnight reset) and 150 per month. Pro: 50 per day and 500 per month.
  • A failed cut or a "no match" still counts. The credit is taken when the AI starts reading your source, so if it errors out or decides the source does not support your claim, that cut is still used. Requests rejected before that point (not signed in, source too large, missing fields) do not cost a cut.
  • Source size cap per card: 40,000 characters on the free plan, 100,000 on Pro. Free users pasting more than 40,000 characters get a clean error, not a cut.
  • Rate limit: 10 cuts per minute on every plan.

Full details, resets, and Pro pricing are in Plans & Limits.

Common problems

  • "No matching evidence." The source text does not actually support your claim. Try a different source or adjust the claim. (This still uses a cut.)
  • Fetch failed, or it is a PDF. The automatic reader could not pull the article (some sites block it, and PDFs are never fetched). Copy and paste the text in by hand.
  • "Source text exceeds 40,000 characters." You are on the free plan and pasted too much. Trim it to the relevant section, or upgrade to Pro for the 100,000-character cap.
  • It is slow, or the app feels laggy. See Troubleshooting & Performance.