It's mid-February. You've been writing a lot lately — three Substacks a week, daily threads. The tone has been working. People are sharing. By March you notice the voice has shifted: you're using a phrase a peer-founder coined on her launch post in January, and you're using it without quotes. By April, the last three months of writing laid side by side, you don't quite recognise the person who wrote it. The line between admiring an idea and quietly putting on a stranger's face has blurred, and the blur happened one sentence at a time.
This is narrative drift. It rarely arrives as one bad week. It arrives as ninety days of small drafts, each nudged a little toward the comparison post that landed yesterday. You are the last person to see it, because you are the one writing the drafts.
Maybe the right question is: what happens if something keeps a word-for-word copy of who you said you were three months ago, and reads it back when the new draft arrives.
A daily routine with an AI assistant — workflow with AI, in short — built for exactly that. This one is an inner routine, not a task manager: it holds the direction of your own story rather than the structure of your work.
Stage 6 — The energetic signal beneath the moment
The drift is not first an integrity problem. It is a phase. You're in a Purge stretch — the old self's curated story is being let go, and the next version hasn't finished arriving — and your vibrational set-point is broadcasting from Performance. The public voice runs ahead of the felt life, and the gap is where a stranger's phrasing slips in. The mirror keeps returning it: that flat I don't recognise the story I'm telling feeling is the drift coming back to you. Most people read it as a branding problem. It is closer to a frequency problem — the body is set to perform, so the language keeps reaching for whatever performed best yesterday.
Underneath the energetic read, the chart-level confirmation, for those who want it: this pattern sits on an Open G center (the identity-and-direction center, the diamond at the chest on the BodyGraph (Human Design body chart) with no color — sense of self comes from the field around you) running its specific Not-Self loop (the version of you shaped by what your open centers picked up from others) — fear of being directionless — with an Open Head center (the inspiration center at the top with no color — pressure to find meaning flows in unfiltered) taking each new inspiration in as an identity-prompt.
The Open G's identity is geographically and socially permeable. It updates when the room changes; that's the design. The healthy version brings the genuine signal home. The Not-Self version brings home whoever it has just consumed, with the seams hidden. The Open Head, sitting above, makes the import easier: each new inspiration lands as this is who I should be writing as. Without a quarterly anchor, twelve weeks later you're wearing a costume made of the loudest posts you read.
A defined G (the identity-and-direction center with color — fixed sense of who you are) holds direction from inside. An Open G needs an anchor that says here is what you wrote three months ago, in your own words, before the costume started. The whole job of the assistant is to be that anchor — and to quote it back without softening.
Stage 7 — The question the AI asks first
Quarterly, the partner asks one question:
"Who are you right now? Three to six sentences. Not a CV — what kind of person, what kind of work, what you're building, what you're learning."
Then it follows up with one scene from the past four weeks that was you, and one that surprised you. These two scenes are the data. The self-story is the interpretation. Both go on the record.
Same DNA as the reflection bot — surfacing the hidden assumption before doing anything with it. Here the hidden assumption is a quiet one: the story I'm telling myself this week is the same story I was telling three months ago. It almost never is.
Stage 8 — The workflow with AI itself
A long-context surface (Claude Project, Custom GPT, or — most reliably — a single dated text file the assistant reads at the start of each session) holds the previous three or more entries.
Trigger: quarterly (March, June, September, December). After-event triggers also fire: post-launch, post-major-conflict, post-identity-relevant decision (a job offer, a partner, a move).
The four-step run:
What the assistant does: holds the long context, quotes verbatim (never paraphrases), refuses to editorialise, surfaces the drift without judging it.
What you still own: deciding what to keep. The assistant does not say that's not really you. It says two months ago you wrote X. Today you wrote Y. The recent scene matches X. Then waits.
A concrete quarterly session, in four lines:
8:30pm, end of March. Bot opens. Reads back the December entry: "I'm a consultant who teaches by example, slowly, on long arcs. I write to think, not to perform. The work I'm building is a small practice with a few real clients." Founder writes today's: "I'm a content founder running a multi-platform brand. The work is scaling reach, building distribution, claiming a category." Bot, slowly: "December said small practice, few real clients. March says scaling reach, claiming a category. The recent scene you described — the L proposal, refused because it would have been too loud — matches December. Where did the March story come from?" Founder, after a long pause: "M's launch in January. I started writing in her register the week after." Bot logs the drift note. Closes.
Stage 9 — The 7-day experiment that grows the capacity underneath
The workflow surfaces drift once a quarter. The experiment is the daily-scale practice that makes the drift visible faster.
Update Life Story (7-day) — for seven days, retell one stuck chapter of your story in past tense, then write a flying-pig redemptive future, then name one real next action. The hypothesis: identity is more flexible than it usually feels, but only when the chapter is being deliberately re-told. Passive drift puts on a stranger's face. Deliberate retelling brings the signal home.
Daily signal: one sentence on what changed when the chapter was retold. Week-end review: which old chapter would you actually keep?
(Companion: recontact-bot — milestone-day sibling. Recontact tracks values; narrative-keeper tracks identity. A wide gap between who I said I am and what I just chose to chase is exactly the substitution pattern recontact-bot was built to surface. Run both. They produce different evidence.)
Stage 0 — Return / Become
After four quarters, the assistant has a real story arc. Not a CV, not a brand, not a positioning statement — a long record of who you said you were, four times over twelve months, with the scenes attached.
Some quarters show a steady self. Some show a quiet becoming — you are turning into someone, slowly, in a direction you chose. Some show a stretch where the loudest outside voice moved into the inside one.
The assistant doesn't fix that. It makes it legible. Once you can read it, you get to decide whether to keep the costume or take it off. That decision is the whole repair.
It's not branding. It's the slow practice of staying in your own voice across a year. The voice is the thing. The launches, the milestones, the platform — all of it sits downstream of whether the voice is yours, or whether it has quietly been someone else's for the last twelve weeks.
The story is the moment. The assistant keeps the older drafts in the room, so the new draft can't be written as if the older one had never existed.
This walks with anyone in a Purge phase whose set-point is broadcasting Performance — anywhere the public voice runs ahead of the felt life and identity stays permeable. The chart-level confirmation underneath, for those who want it: an Open G center (no-color identity-and-direction center), Open Head center (no-color inspiration center), Open Ajna center (no-color conceptual mind), or a Reflector (the rare lunar mirroring Type, ~1% of people). The milestone sibling is recontact-bot; the daily-scale companion is the reflection bot; the witness layer is witness-bot when a stored entry needs to leave the assistant and meet a real peer. See your chart →