Build a Facts File Before You Open an AI Tool

Before you ask any AI tool to write a line of your resume, write down what actually happened. A facts file is a private document — never sent anywhere — that lists your employers, exact dates, official job titles, what you were responsible for, and what you can prove. Everything an AI produces afterwards gets checked against it. Without one, you are asking a language model to fill gaps it has no information about, and it will fill them plausibly rather than accurately.

What a facts file is for

An AI drafting tool works from whatever you put in the prompt. If you paste in “I was a marketing coordinator for a while and helped with campaigns,” you have given it three vague facts and a large empty space. The output will be fluent, confident, and partly invented, because fluency is what the model optimises for. The facts file closes that space in advance.

It also serves a second purpose that matters later. When someone asks you in an interview to explain a bullet point, you want to be reading from your own memory of a real event, not reconstructing what a tool wrote about you six weeks earlier. The file is the thing you go back to.

Keep it in whatever tool you already use. A plain text file or a single document is fine. It should be ugly and complete rather than tidy and partial.

What to put in it

Work through it employer by employer, oldest to newest, and record:

The legal employer name. Not the brand you worked under, if they differ. If you were placed by an agency, note both the agency and the client site.
Start and end dates, to the month. Take these from a payslip, offer letter, or the final payslip rather than memory. Memory drifts, and dates are the one part of a resume most likely to be checked against a third party’s records.
Your official title as it appeared in the HR system. Also note the title you actually used day to day if it was different, and keep them separate.
Who you reported to, and who reported to you. A number of direct reports is a factual claim; write down the real one, including if it was zero.
What you were accountable for versus what you contributed to. These are different, and conflating them is the most common way a resume becomes indefensible.
Concrete outcomes you personally witnessed, with a note of where the evidence lived — a dashboard, a report, a shared drive, an email thread.
Tools and systems you used, with an honest note on depth: built with it, maintained it, used it occasionally, watched someone else use it.
Education and credentials, with awarding body, completion status, and dates. Include anything unfinished or expired, marked as such.

Write down where the proof was

For each outcome, add one line: how would you show this to a sceptic? Sometimes the answer is a document you still have. Sometimes it is a person who would remember. Sometimes it is nothing at all, because the system was decommissioned and you left the company.

That last case is not a reason to leave the achievement out. It is a reason to phrase it in a way you can still stand behind. Knowing in advance which of your claims have evidence and which rest on your own recollection changes how you word them, and it stops you from being surprised. If you have already lost access to the underlying numbers, there are honest ways to describe the work anyway.

Do this before, not after

The order matters. If you draft first and gather facts second, you will find yourself trying to justify sentences that already exist, and the pull towards accepting a slightly-too-generous phrasing is strong because the sentence reads well. If you gather facts first, the draft is constrained from the start and there is nothing to argue yourself into.

This is also the point at which you decide scope. A facts file covering your entire history lets you decide separately, for each application, what to include and what to leave out — a selection decision rather than a memory exercise.

What not to put in the prompt

Your facts file will contain things that should not be pasted into a third-party tool: former colleagues’ names, internal project code names, client identities, figures that were never public. Treat the file as the source and the prompt as an extract. Strip identifying details and anything covered by a confidentiality obligation before it goes anywhere. If a large part of your work falls into that category, describing it without disclosing it is a skill of its own.

Keeping it current

The best time to add to a facts file is while you are still employed and still have access. Note the dates, the outcome, and where the record lives, on the day it happens or the week it finishes. People who do this find the next job search dramatically easier, not because the writing is faster but because they are not trying to reconstruct three years from an old calendar.

If you are only starting now, do the most recent role in detail and the older ones in less. Depth of recall drops off, and a resume usually gives older roles less space anyway.

What comes next

With the file in place, an AI tool becomes genuinely useful: you are asking it to compress, order, and phrase material you have supplied, which is what language models are good at. You are not asking it to know things about your life, which it cannot. The next step is a checking habit — reading each generated line back against the file and confirming it says something you can defend.

The short version: collect the facts first, in private, in more detail than the resume will need. Then draft. Every claim on the page should trace back to a line in that file, and if it does not, that is the signal to change the claim rather than the file.