A Neural Network for Your Reading List: An AI Classifier in a Telegram Bot

A developer built a Telegram bot that uses a neural network to automatically classify the articles and videos in a "read later" queue.

What the bot does

The bot tackles a specific problem: the endless "read later" folder in the browser that nobody gets around to for weeks. The user drops in a link, and the bot takes over:

• automatically determines the material's category
• estimates the time needed to read an article or watch a video
• reminds the user again and again until they deal with the content
• helps users finish online courses and books instead of abandoning them halfway

Three months of evening development, real utility, and one unfortunate phrase in the description.

Why the top comment got +14

The author opened every post about the bot with the phrase "runs links through AI" or "AI inside," to sound more solid. The top comment under the latest article collected more upvotes than the rest of the discussion thread combined. It shows a grimace captioned "my face when I read about yet another AI gimmick."

"My face when I read about yet another AI gimmick" — top comment under

the article, +14.

The developer recognized that grimace: he makes the same face at every landing page that says "AI-powered." Once his view shifted, it became clear what the critics objected to: not the bot, but the way it was pitched. He set out to understand why those two words irritate a technical audience. He found the answer, and also found that he had scattered "AI" generously in places he had not thought to check.

Where AI fatigue comes from

"AI fatigue" is not an audience whim. It is a reaction to a persistent pattern that repeats too often. Several factors feed it:

• Label inflation: over the past two years, "AI-powered" has appeared on apps, cameras, vacuum cleaners and coffee machines, whether or not a real model runs inside
• No specifics: the phrase "uses AI" explains neither what exactly the model does nor why this particular product needs it
• Eroded trust: too many "AI revolutions" failed to materialize, so the audience now reads the label as a warning, not an advantage
• The venue shapes expectations: Habr is not ProductHunt. Readers here ask "how exactly?" and "why?" rather than taking marketing claims on faith

All of this works as a filter: when a reader sees "AI-powered," the first thought is "what exactly?" If the next sentence does not answer, trust drops instead of rising.

The developer's story is instructive because he was not trying to sell hot air. The bot really works and solves its task. The problem lay solely in the wording, yet that wording determined how the audience received the whole project.

What this means

If a language model inside a product solves a specific task, show it through the result: "picks a category from twelve options" works better than "runs on AI." When "AI inside" is the first argument, it signals that there are no other arguments. The technical audience on Habr senses this, and responds with a grimace and fourteen upvotes.

What problem does the

Telegram bot with a neural network solve? It helps organize a "read later" folder automatically: the bot determines the material's category, estimates the time needed to read an article or watch a video, and reminds the user.

How does the neural network help organize content in the bot?

It automatically determines the material's category and estimates the reading time.

Why can AI in such tools irritate users?

According to the developer, the bot's description matched the typical description of the irritating "AI gimmicks."