How it works
From a scattered topic to a send-ready issue
Quick Letter runs the research loop a newsletter writer would run by hand: search, score, curate, draft. This page walks through what happens at each step, and the mechanics underneath it.
Start with a topic, not a brief
A title is enough to get going. Add a sentence of context if you want to point the agent at a specific angle, industry, or audience — the query generator uses whatever you give it to plan its first searches. You can also set how fresh results need to be for each content type, so a weekly roundup and a slower-moving beat don't use the same filter.
Search runs in the background, curation stays yours
The agent turns your topic into three or four targeted search queries, pulls articles, news, images, and video, scores each result for relevance, and drops duplicates automatically. You star what's worth keeping, dim what isn't, and drag to reorder — the review pass is fast because the queue already leans toward what fits.
Starring does more than tidy the queue: the next research run reads your starred titles and sources back into query generation, so search keeps drifting toward what you've already told it is good.
One click turns curation into a draft
Pick a format — a numbered roundup, a clean curated-links list, a thread-style draft — plus a tone and a length, and the draft generator writes from your starred items only. Every link in the draft is checked against the sources it was given; anything that doesn't match gets flagged before you publish. From there it's a normal rich-text editor: reorder, rewrite, add images, export.
Under the hood
The mechanics behind each step
Multi-source search
One topic produces three or four AI-generated queries that run against web articles, news, images, and video in parallel, so a single research pass covers every content type at once.
Relevance scoring
Every result gets a base relevance score from the model, then a calibration pass that rewards known-trustworthy domains and recent publication signals. An uncertainty band keeps promising niche results instead of cutting anything below a hard threshold.
Relevance feedback
Starred items aren't only a filter for you, they're context for the agent. This is a lightweight, explicit form of relevance feedback: the next research run reads your starred titles and sources back into query generation, so curation compounds instead of resetting every time.
Grounded drafts
Every markdown link in a generated draft is checked against the URLs of the items it was given. A separate pass flags numbers and names that appear in the draft but not in the source material, so unsupported claims don't slip through quietly.
Background refresh
A scheduled job re-researches active topics automatically, so coming back to a topic means fresh material is already waiting instead of a blank queue.
Flexible drafts
Format, tone, and length are independent choices. A numbered roundup in an enthusiastic voice reads nothing like a curated-links list in an analytical one, and the generator adapts structure and prose to match.
Curious how it holds up on a real topic?
Quick Letter is invite only for now. Sign in if you already have an account.