How we make our guides

The “what to say” guides are written by NeverBlank, the company that makes the app, with help from AI. Each one starts from a conversation people practice in the app, is checked by automated tests and an AI editor, and, since November 7, 2026, is read by a person before it goes live.

Updated November 7, 2026 · NeverBlank

Who writes them

The guides are published by NeverBlank as an organization. We don’t put a person’s name on them, because no single person writes them: the text is drafted with an AI model (Claude, made by Anthropic) and edited and checked as described below.

How a guide is made

  1. Pick a question people ask. We look for evidence that people want the answer: searches that already reach our guides, and questions people ask in public forums. A topic with no sign of demand waits.
  2. Start from a practice scenario. Every guide is built from one of the conversations in NeverBlank: the situation, what the other person typically says, how they push back, and the mistakes people make.
  3. Draft with AI. An AI model writes the first draft: the line to say, other ways to say it, weak, OK and strong answers, the pushback to expect, steps and common questions.
  4. Edit and score. A second AI pass reads the draft as the person searching and asks for fixes; the draft is rewritten once if needed. The weak, OK and strong answers are then scored by the same judge that grades live rounds in the app, and a guide whose scores don’t rise from weak to strong is dropped.
  5. Automated checks. Every guide must pass tests for format, wording rules, claims and duplication (below) before it can be proposed.
  6. A person reads it. Since November 7, 2026, a person at NeverBlank reads every new or changed guide, including its title, description and image, before it is published. Approval covers that exact version; any later change needs a new review. Guides last updated before that date passed the automated checks, and each guide says which applies to it.

What the checks look for

  • No statistics, studies or surveys in the text. The only numbers about people come from NeverBlank’s own practice data, shown as totals across at least 20 people and 30 live rounds, never individuals.
  • No health, therapy or outcome promises. The guides suggest words to say; they are not legal, HR, medical or financial advice.
  • No near-copies: a guide that mostly repeats an existing one is dropped, and the existing guide is improved instead.
  • Header images are illustrations made with an AI image tool and checked to show no faces and no readable text.
  • We change a guide’s “updated” date only when its content changes.

We publish a small number of guides each week and add more only when people find the existing ones useful. Comparisons of other apps, such as this feature comparison, describe what each product says about itself, with a source for every fact; we don’t test or rate other products.

Report an error

If something in a guide is wrong, unclear or could cause harm, email neverblank.app@gmail.com with the page link and what you noticed. Every guide also has a “Report an error” link at the bottom. We read every report; when a guide is wrong we correct it or take it down, and a corrected guide shows its new “updated” date.