Skip to content

Your AI is working from an old brief

Your AI is working from an old brief, with the cover of the video: Anna and Patrick under the words Your AI's old brief

Most of us set our AI up once, wrote it some instructions, and moved on. The business kept changing. The instructions did not. In this week’s video Anna asks Patrick what happened when he finally went back and checked ours.

Plays from YouTube, which may set its own cookies once you press play. Watch on YouTube (opens in a new tab)

What we found

All our Built Not Boosted content runs through one Claude project. It holds the brief: who we are, where we post, how we sound and what we will and will not talk about. Every caption, script and article starts from it.

That brief was written when it was just Patrick, on three platforms. It is the two of us now, eight channels and a website, and nobody had told the AI.

So Patrick asked it to check the old brief against where we are now. It found four clashes:

  • The wrong scheduling tool. The brief named a tool we were not using for these channels.
  • The wrong channels. It still thought in terms of one person’s three profiles.
  • No rule on “I” or “we”. With two of us, every post needs to know who is speaking.
  • A content boundary that had completely changed. Something it was told to steer clear of had moved, and the brief had not moved with it.

Why it matters

None of those clashes shows up as an error. The AI does exactly what the brief says, so everything looks fine until a post goes out wrong: the wrong voice, the wrong platform, or a topic you had decided to leave alone.

Your AI only knows the version of you that you last told it about. If your business has changed since you wrote its instructions, and most have, it is working from an old brief.

One thing to try today

Find the instructions you gave your AI. That might be a custom GPT, a Claude project, a saved prompt, or the “about me” notes in your settings. Paste them back into a new chat with a list of what has changed, and ask what no longer fits. Something like this:

Here are the instructions I gave you. Since I wrote them, this has changed in my business: [your list]. What in these instructions no longer fits?

Then fix what it finds, and put a reminder in the diary to do it again whenever something big changes. How many clashes did yours find? Tell us in the comments under the video.

Where to find us

A new video goes out every week on YouTube, TikTok, Instagram and Facebook, all as @built.not.boosted. Each one gets a post here too, with a bit more detail than a short video has room for.

If you want to see how we keep our AI’s knowledge up to date behind the scenes, read how we run our AI brains.

Transcript

Patrick: So your AI is probably still working from a brief you wrote months ago, and it’s quietly getting things wrong.

Anna: Speaking from experience, are we?

Patrick: Well, guilty, yes. I’m Patrick.

Anna: And I’m Anna, and together we run Built Not Boosted. Drop us a follow, and let’s get into it.

Patrick: Now the other day I opened the Claude project behind all our content. The brief in there was written when it was just me, on three platforms.

Anna: And it’s not just you any more.

Patrick: No, it’s two of us now. That’s eight channels and a website, and nobody’s told the AI.

Anna: And so what did you do?

Patrick: Well, I asked it to check the old brief against where we are now, and it found four clashes. The wrong scheduling tool, the wrong channels, no rule on whether we write as an I or a we, and a content boundary that had changed completely.

Anna: And none of that shows up until a post goes wrong.

Patrick: Your AI only knows the version of you that you last told it about.

Anna: So, one thing to try today. Paste your AI’s instructions back into it, tell it what’s changed, and ask it what no longer fits.

Patrick: How many clashes did yours find? Let us know in the comments.

Anna: Speak to you soon.