How to train your AI Twin without dumping low-value copied content

The strongest Twin training set is material you created in the course of real work: workshop outlines, client FAQs (anonymized), research memos, design critiques, code architecture notes, teaching curricula, and post-mortems. These artifacts carry voice, judgment, and structure that generic web pages lack.

Importing a public webpage or a résumé dump can help bootstrap, but it is rarely enough on its own. Advertising networks and human visitors both look for added value — commentary, organization, examples, and clear authorship. If a page is mostly content copied from elsewhere with no transformation, do not make it the centerpiece of a public hub.

Before you mark a source public, ask whether a stranger would learn something specific from it that they could not get from a generic search result. If the answer is no, keep it private or rewrite it into an original guide.

• You wrote most of the words, or you substantially curated and explained third-party material
• Examples are concrete (projects, constraints, trade-offs) rather than vague slogans
• Sensitive client data is removed or generalized
• The document has a clear topic so retrieval can find it later
• You would be comfortable putting your name on the page

Imports are accelerators, not finished public pages. After connecting LinkedIn, Notion, Docs, GitHub, or a website, open the extracted text and edit. Add a short “how I use this” intro, trim boilerplate, and split giant dumps into focused notes. That curation step is what turns raw material into a Twin people trust.

For third-party articles you did not write, prefer summarizing the idea in your own words and linking out, rather than republishing the full text. If you must keep a long excerpt for private retrieval, leave it private so it never appears as a public content screen next to ads.

For Discover and review quality, aim for: a unique display name and headline, an original 2–3 paragraph bio, a profile photo, three or more public sources with distinct topics, and Twin chat tested with five real questions from your audience. Then enable public chat.

Resist the urge to publish an empty shell “to claim the username.” Incomplete hubs hurt network trust and can look like low-value pages to crawlers. Finish the content bar first; growth mechanics second.

After launch, review Twin inbox transcripts weekly. When the Twin misses, add a source or clarify an existing note — do not paste random web articles to “fill gaps.” Quality compounds when each new upload answers a real failure mode you observed.

Quarterly, archive outdated advice and update pricing or consultation boundaries. A Twin that cites obsolete guidance is worse than a Twin that admits uncertainty.