How Warmth Researches and Creates Its Field Notes
Our editorial policy explains how Warmth selects topics, uses AI-assisted drafting, checks product claims and sources, handles comparisons, and corrects mistakes.
Warmth publishes field notes to answer the questions adults ask while deciding whether an AI friend belongs in their lives. This page explains who is responsible for those notes, how they are created, and where the limits are.
The short version: Warmth uses AI-assisted research, outlining, and drafting, then checks the resulting page against current sources, the actual product, and Warmth's published policies. Automation can make a research process broader and more consistent. It does not make an unsupported claim true.
Who publishes these field notes?
The publisher and named author is Warmth editorial. The articles reflect Warmth's product knowledge and point of view about AI friendship, messaging, memory, privacy, and relationship design.
Warmth also has an obvious commercial interest: we make Mia, an AI friend for adults who lives in iMessage. Articles that discuss companion products, product categories, or buying decisions should be read with that relationship in mind. We do not present Warmth as an independent review publication.
How are topics selected?
A topic earns its own page only when it serves a distinct reader question. We look for questions that fit one of four areas:
- understanding what AI friendship is;
- deciding how an AI companion fits into everyday communication;
- evaluating privacy, safety, control, and boundaries; or
- understanding the product principles behind Mia.
Closely related keyword variations belong on one comprehensive page rather than several nearly identical pages. A search phrase is useful evidence of how people ask a question, but it is not a reason to manufacture a page with no additional value.
Where does AI assistance fit?
AI tools can help collect query variations, organize source material, find gaps in an outline, and produce an early draft. Warmth uses those capabilities in this library. We also use automation to check mechanical details such as frontmatter, internal links, metadata, heading structure, and route generation.
The useful work still depends on judgment: deciding whether two queries need separate answers, separating a product fact from an opinion, refusing to invent evidence, and removing language that promises more than an AI companion can responsibly provide.
Our standard follows Google's people-first content guidance: a page should be substantial, accurate, and useful to Warmth's intended audience even if the reader arrived without a search engine.
How are factual claims checked?
Different claims require different evidence.
- Warmth product claims are checked against the current product experience and product copy.
- Privacy and data-use claims are checked against Warmth's Privacy Policy and Terms. If a friendly summary and the policy ever conflict, the policy controls.
- Competitor claims should link to the competitor's current official product, help, pricing, terms, or privacy pages and show when the information was last checked.
- Technical explanations should distinguish a general concept from a claim about the exact system Warmth uses.
- Research findings require real data and a described method. We do not publish invented users, surveys, tests, quotes, or performance numbers.
Sources can change after publication. A link is evidence of what was available when a page was checked, not a promise that another company will keep a feature or policy forever.
What will these notes not claim?
Mia is AI, not a person, therapist, emergency service, or substitute for professional care. Warmth does not publish claims that an AI friend diagnoses, treats, cures, or prevents loneliness, anxiety, depression, or any health condition.
We also avoid presenting emotional dependence, guilt, exclusivity, or replacing human relationships as product benefits. A useful AI friendship should leave room for the rest of a person's life.
How are comparisons handled?
Comparison pages begin with a job to be done, not a predetermined winner. They explain the criteria, identify the publisher's commercial relationship, use current first-party sources, and acknowledge where another product may fit better.
A feature table is only as reliable as its update date. Freshness-heavy comparison pages should be reviewed regularly and changed only when the underlying facts change—not merely given a new date.
How are corrections handled?
When a material fact changes, the page should be updated and its updatedAt date should change with it. Small punctuation edits do not justify a freshness signal.
If you find a factual error, broken source, or unclear disclosure, email team@warmth.so with the page URL and the detail that needs review.
The standard we are aiming for
Every field note should leave a reader with a clearer definition, a usable decision, or a more precise question. Keywords help the right person find that answer. They are not the answer themselves.