News & reading

What news and reading apps are worth building?

Reading tools earn their place when they reduce a decision: what to read next, what to skip, what to keep. A feed that shows everything is already available in many forms. A tool that takes a pile of saved articles, ranks them against a question the reader sets, and says why each one ranked where it did is a different product. The same logic applies to a long government report, a stack of research papers, or a book assigned in a course. Each of these is a pile, and a pile is a problem a small tool can address.

Four shapes recur. Triage tools sort material against a question the reader supplies. Summary tools shorten a source and keep a link or quotation back to the passage they came from. Reading-plan tools break a book or a long series into dated portions that fit a real schedule. Digest tools collect newsletters or alerts into one page, with the items that match the reader’s stated interests first and a count of what was left out. Triage and summary carry the most risk, because a wrong summary is a claim the reader may repeat to someone else.

A reading idea should also say whose material it works on. A tool that works on text the reader supplies is a different kind of product from one that chooses the articles itself. The second kind needs editorial standards, source policies and moderation, which is a much larger job than a first version should take on.

Who reads with these apps, and how do you reach them?

Students with a reading list they have not finished, researchers working through papers in a narrow field, professionals who are expected to know a sector’s news without the time to read it all, and readers who keep a long saved-for-later list are the clearest groups. Each has a specific pile. Ask someone in one of these groups to show you their pile, and build for that pile before you think about the category as a whole. The pile tells you what the first version has to handle: how many items, what format, and what decision is stuck.

Reach them where the pile lives: the communities for a particular field, course forums and the places where people share reading lists. A tool that handles a real list, shown publicly with the reader’s permission, is more persuasive than any feature page. Test with the same list twice. Does the tool save time on the second pass, when the reader knows what to expect? A tool that only works the first time is a demonstration, not a habit.

What makes a news or reading idea good?

• The material already exists in the reader’s hands: saved links, a PDF, or an export from a read-later service. Asking the reader to find new material first adds a step most will not take.
• The question is set by the reader. “Show me what matters” is vague. “Which of these mention the budget vote in the spring” is a question the tool can answer or fail to answer, and the reader can check.
• Every output points back to its source. A summary without a link to the sentence it came from cannot be verified, and an unverifiable summary is worse than none.
• It works on the reader’s own material. That keeps the app out of ranking and recommending what other people read, which is where most of the difficulty in this category sits.
• It shows what it left out. A digest that drops items without telling the reader is hard to trust. Show the count and let the reader open the rest.

Write the question the reader would type into the tool, and check whether a machine could answer it honestly from the text alone. If the honest answer is “it depends on what the author meant”, narrow the question.

What does a first version look like? A worked example

Hypothetical triage page for a backlog of saved articles. The reader pastes the text of up to fifteen articles, each separated by a line, and one question. The app returns the articles in order of how directly they answer the question, each with the single sentence that matched and a note on anything it could not judge. It works on text the reader already has, so it never has to fetch or republish an article.

Build a single-page web app. Input: one line for the question, and up to fifteen pasted article texts, each separated by a line containing three dashes. For each article, find the one sentence most relevant to the question and quote it exactly. Return the articles in order of relevance, each with the quote and a score from 1 to 3. If an article has no sentence that answers the question, mark it “no match” and leave it unranked. Show a count of the unranked articles at the top. Store nothing on a server.

Test it with a reading list you already know well. If the top three matches are the ones you would have picked, and every quote appears word for word in the source, the first version is working. Then ask a question you have not yet answered yourself, and check that the “no match” items really do not answer it. A tool that always finds something has no way of saying no, and that is the failure to design out first.

What goes wrong with news and reading apps?

• Summaries that sound authoritative and are wrong. A summary can drop a qualifier, a date or a figure. Show the quoted sentence beside every summary so the reader can check it in seconds.
• Copying articles. Reproducing the full text of a paywalled or copyrighted article on a public page is a different act from letting a reader work with their own copy. Keep processing on text the reader supplies, and do not republish it.
• Turning into a news feed. Once an app decides what a reader sees next, it takes on editorial responsibility for those choices. A reader-supplied triage tool avoids most of that.
• Engagement mechanics. Streaks, unread counters and “you are falling behind” messages make a reading list feel like a debt. Let the reader close the list without penalty.
• Ignoring how people read. Many people read on paper or in a PDF reader. A tool that accepts only links shuts them out. Accept pasted text and common document formats.

Where to start: post the first idea

No idea in this category was on the board when this page was written, so there is nothing to link from here. Name the reader, the pile they have, and the question the app answers about that pile. A post such as “triage a saved-articles backlog against one question” gives a builder enough to start.

If you have a pile of your own that no tool handles, write it up on the board: the reader, the pile, the output. To see how other categories phrase their users, browse the whole board. A builder who takes your idea can use how to validate an app idea to decide how many readers to ask before spending a weekend on it.