Smart Proofread
Proofread is the pass that catches what eyes miss after twenty pages: drifted terminology, missed glossary terms, stylistic slips. It’s pre-scan driven — it reads first and only touches what actually warrants it.
How does Smart Proofread work?
Proofread reads the blocks, flags the ones with real errors or unclear phrasing, and fixes only those. Blocks that already read well get no fix and cost only the small read fee (0.2 credits per word). On multi-page documents it scans a page at a time, so the first page’s fixes start landing before the rest is scanned.
- Fixes apply directly to the active translation as they land. When the run finishes, Review changes on its toast opens every changed block with the earlier text beside the new one, to Keep or Revert to previous. The run is also listed under the reviews pending list as Smart Proofread · N changed until you review it or clear it with ×. Each change shows the note it was made for, and reverting one tells the next proofread to leave that wording alone.
- If someone edited a block by hand after the AI last wrote it, the fix does not replace their text. It waits in the same review with Accept or Keep current, and Keep current tells the next proofread not to suggest it again.
- Run it on a selection, the current page, or the whole document — same tool, different range.
- On a translated document, the proofread actions are in the menu behind the arrow beside Download, at the top of the editor. On a phone, Proofread is in the top bar.
- Each flagged block gets a short note from Literess in the comments panel explaining what needed another pass; once the fix lands, that note auto-resolves and is listed under the panel’s AI checks filter.
- If everything already reads well, nothing changes and you’ll see a “nothing to fix” result.
How much does Smart Proofread cost?
Two parts. A read fee for the scan, 0.2 credits per source word it reads (0.5 with vision), and then the blocks it actually changes, at the same rate as a translation: one credit per source word in Fast or Standard mode, three in Pro. Blocks it reads and leaves alone cost only the read fee, so a clean 1,000-word document comes back with a “nothing to fix” result for 200 credits. The run dialog shows the read fee as the guaranteed floor and the fixes as the ceiling. See plans, credits & the free tier.
Can I tell Smart Proofread what to focus on?
Yes — the Proofread dialog takes an optional instruction. Leave it blank for a straight error-fix pass, or write a direction in plain language (“tighten the wording”, “use a more formal register”) and the pre-scan flags the blocks that don’t yet match it, alongside any with genuine errors. Blocks that already read well and already match your instruction are still skipped, so beyond the read fee you’re billed only for what changes.
What is the difference between Smart Proofread and Regenerate?
Smart Proofread reads everything and rewrites only the blocks with a real problem. Regenerate creates a new version for every block in the scope you picked, whether it needed one or not, and is billed for all of them. Use proofread as a QA sweep; use regenerate when you want a fresh take. See regenerate & rewrite with a direction.
Does Smart Proofread use my translation memory?
Yes, on Standard and Pro models: your past approved translations are passed to the model as reference during the proofread, so its fixes reuse your established wording instead of inventing fresh phrasing. Fast models don’t use memory. See translation memory.
Still stuck? Ask Literess in the app, or write to support@transept.ai.