Draft with AI: troubleshooting common problems
Fixes for common Draft with AI problems: the disabled Generate button, daily and monthly AI limits, timeouts, empty drafts, Counsel Mode Ollama errors, and characters rendering as question marks.
Last updated
Fixes for the problems people most often hit in Draft with AI.
The Generate button is grayed out
Cause: your prompt is under 10 characters — the button's tooltip says "Describe what you need (at least a sentence) to generate a draft."
- Type at least one full sentence describing the document. Names, dates, amounts, and jurisdiction help the most.
- If you pasted a very long brief and got "Prompt is too long (max 2000 characters)", trim it to the essential facts — the template supplies the structure and boilerplate, so your prompt only needs the specifics.
"You've used your N drafts for today"
Cause: the daily drafting cap — 5 drafts per day when signed out, 30 per day when signed in. The message tells you roughly how many hours until it resets.
- If you're signed out, sign in to move to the 30-per-day allowance.
- Otherwise wait out the window in the message, then try again.
"You've used your N AI actions this month"
Cause: drafting (along with AI auto-fill, AI redaction suggestions, and clause review) consumes your monthly AI-action allowance: 25 on Free, 200 on Pro, 500 on Max by default. It resets at the start of each calendar month (UTC).
- Once paid upgrades are live, the message includes an Upgrade button pointing at a tier with a larger allowance. During the private beta, checkout is off, so the message instead notes you've reached this month's free-beta AI limit — the cap applies even while the beta is free.
- Already on Max? Its limit is a fair-use cap rather than an upsell — contact support if you regularly need more.
- Or turn on Counsel Mode (see below): local drafts never consume the allowance.
"Drafting timed out — please try again."
Cause: generation has a 2-minute watchdog so a hung model or dropped stream can't leave the page stuck on "Generating draft…".
- Retry — a full contract normally streams in well under a minute.
- If it times out repeatedly, shorten the prompt and check your network connection.
- The 2-minute watchdog applies to cloud drafts. Counsel Mode drafts run against your local Ollama and aren't cut off at 2 minutes — on slow hardware a local model can simply take a long time, so try a smaller or faster model if drafting crawls.
"Model returned an empty draft."
Cause: occasionally the model produces no usable text for a vague or contradictory prompt.
- Rephrase with more concrete details — parties, dates, amounts, governing law.
- Make sure the prompt matches the selected template (e.g. don't describe a lease change while the Freelance Contract template is selected).
"AI drafting is unavailable right now." or "AI is temporarily unavailable"
Cause: server-side AI is temporarily off. "AI is temporarily unavailable (usage ceiling reached)" means the service's global AI budget has been hit (the daily window resets at midnight UTC); "AI drafting is unavailable right now" or "temporarily disabled by the operator" means cloud drafting isn't configured or has been switched off.
- Try again later.
- If you have Counsel Mode available, turn it on to keep drafting locally through your own Ollama in the meantime.
Counsel Mode: "Ollama at http://localhost:11434 returned …" or not reachable
Cause: Counsel Mode sends the prompt from your browser to your local Ollama, and that request is failing.
- Check that Ollama is installed and running on this machine.
- Pull the configured model:
ollama pull llama3.1(or whichever model the error names). - Check the base URL and model the Counsel Mode panel reports (defaults:
http://localhost:11434,llama3.1) — they must match your running Ollama. - If you enabled Counsel Mode for the first time on this device, reload the page (the panel shows a Reload to activate button) — the browser security policy that permits the localhost connection only takes effect on the next page load.
- "Ollama returned an empty draft." means the connection worked but the model produced nothing — retry, add detail to the prompt, or try a larger model.
Characters come out as "?" in the PDF
Cause: drafts are rendered with standard Latin (Times) fonts, which can't encode Cyrillic, CJK, emoji, and some accented characters; those are substituted with ?.
- Keep names and terms in Latin script where you can.
- Need the original characters? The generated text is intact in the Draft page's source panel — visible as Streaming draft… while the draft generates, and under Show source markdown afterwards if the PDF downloaded directly instead of opening in Sign. Copy it from there into your own editor to produce the final document.
The draft downloaded instead of opening in Sign
Cause: the automatic handoff to the Sign tool uses browser storage (IndexedDB), which some private-browsing modes block. AttachKit falls back to a direct download (e.g. nda-draft.pdf) and shows "Draft downloaded."
- Nothing is lost — open the Sign tool and drop the downloaded file in.
- To get the automatic handoff back, use a regular (non-private) window.
Still stuck?
Contact support and mention which template you used and the exact error text — you don't need to share your prompt or the generated document.
Related
Still stuck? Contact support →