# Run an evidence-recovery conversation

## Purpose

This guide describes how to use a dedicated ChatGPT conversation to recover professional evidence gradually. It is designed for people whose experience is distributed across several roles, fields, projects or less formal activities.

The conversation is not primarily a writing exercise. Its job is to reconstruct bounded episodes, locate supporting material and distinguish what is known from what is inferred. Polished career language can be produced later.

## Set up the conversation

Create a separate conversation or project for the evidence garden. Keeping it separate allows the interview to accumulate context without mixing the records with unrelated work.

At the start, provide:

- the starting brief from the general guide;
- one recent CV, if available;
- a short note explaining why the current public record feels incomplete;
- any boundaries concerning confidential clients, personal information or material that should not be uploaded.

Do not upload an entire archive immediately. Start with enough material to establish the method, then add documents when a particular thread makes them relevant.

## Give the AI a clear role

The AI should act as an evidence interviewer and record keeper. It should not behave as a branding consultant during the recovery stage.

Ask it to:

- proceed through conversation rather than issuing a long questionnaire;
- ask one or two questions at a time;
- request examples when the answer remains abstract;
- distinguish personal contribution from team outcomes;
- label recollection, documentary evidence, third-party statements and interpretation separately;
- retain uncertainty and disagreement;
- avoid inventing outcomes, scale, dates or causality;
- propose capability links only after the episode itself is clear.

## Start with a loose landscape

The first task is to sketch the territory, not to extract a complete chronology.

Useful areas to map include:

- paid roles and consulting engagements;
- businesses, experiments and abandoned ventures;
- substantial side projects;
- voluntary and community work;
- things built, repaired or organised;
- recurring requests for help or judgement;
- difficult transitions or periods of unusually intense learning;
- work that was important but poorly represented by its title.

The AI should reflect this map back without trying to impose a unifying story.

## Select the first thread

Choose an area that is both meaningful and recoverable. A good first thread has at least one of the following:

- relatively clear memory;
- surviving documents;
- another person who could confirm parts of the account;
- a tangible output;
- a problem that reveals more than the official job description.

The easiest starting point is not necessarily the most prestigious project.

## Interview one episode at a time

The interview should recover the sequence and the person’s role without forcing a rigid storytelling formula.

Questions may cover:

### Context

- When and where did this happen?
- What was the wider project or situation?
- Who was involved?
- What constraints shaped the work?

### Problem

- What was difficult, unclear, missing or at risk?
- How was the problem understood at the start?
- Did the framing change?

### Contribution

- What did the person personally notice, decide, create, change or carry?
- Which parts were shared with others?
- What work was important but largely invisible?

### Connection and outcome

- How did the contribution affect the wider process?
- What existed afterwards?
- Is the result directly documented, remembered, reported by somebody else or merely plausible?

### Evidence

- Which files, messages, outputs or people could support the account?
- Do several documents describe the same event?
- What remains uncertain?
- What should stay private?

## Draft, review and correct

After enough detail has been recovered, ask ChatGPT to draft one episode record using the agreed structure. Review it for:

- inflated verbs;
- whole-team results attributed to one person;
- an intention described as an achieved result;
- dates or numbers introduced without a source;
- generic capability labels unsupported by the episode;
- causal claims that are neater than the available evidence;
- wording that feels impressive but no longer feels true.

Corrections are part of the method. They are not a failure of the interview.

## Add sources in focused batches

Once an episode exists, provide the documents most likely to clarify it. Ask ChatGPT to identify what each source adds and what it does not establish.

If a large archive exists, process it in bounded batches. Keep an intake register recording:

- what was reviewed;
- which episodes it supports;
- new leads it introduced;
- contradictions or attribution problems;
- material deferred for later.

## Pause for pattern review

After three to five varied episodes, ask ChatGPT for a cautious cross-record review:

- Which kinds of problems recur?
- Which contributions appear across different settings?
- Which patterns may be artefacts of the available documents?
- Which capabilities are directly supported, and which remain hypotheses?
- What important area of the person’s work is still missing?

Use this review to propose navigation categories or further interview threads. Do not treat it as a personality assessment.

## Suggested opening instruction

> We are starting an evidence-recovery process, not writing my public profile. First help me map the broad territory of my work without forcing it into a single narrative. Ask one manageable question at a time. When something sounds like a broad capability claim, ask for a concrete episode. Keep facts, recollection, third-party statements and your interpretation distinct. Once we have enough detail for one episode, draft it for my review rather than moving immediately to the next topic.

## A sustainable rhythm

The collection can grow through short sessions. One conversation might recover a project. Another might examine a set of documents or correct an existing record. The system should reduce dependence on memory, not create a new administrative burden large enough to require its own evidence garden.