How to Feed an AI Tool Your Background Material Without Losing Control of the Output

The brief tells the tool what to write. Background material tells it what you know. These are different jobs, and the problems start when people mix them.

A brief is short and directive: here is the reader, here is the question, here is the shape. Background material is the pile of things you want the draft to draw from: a call transcript, your notes from a conference, an old blog post, three client emails, a chapter of a book you keep recommending. It is raw, messy, and full of things that should not appear in the final piece.

Most people handle background material in one of two ways. They skip it entirely, which leaves the tool writing from its general training data, and that is where the generic sound comes from. Or they dump everything in at once, which produces a draft that is long, disorganized, and weirdly confident about things they mentioned once in passing.

There is a middle path, and it is not complicated. You just have to treat loading material as its own step, separate from the brief, with its own small set of rules.

Why background material changes the output more than a better prompt

When you give the tool a brief and no material, it has to fill the space between your instructions and the finished draft with whatever it already knows. Whatever it already knows is the internet’s average take on the subject. That is why two people with different briefs on the same topic can get drafts that sound oddly similar. The shape is different, but the substance comes from the same pool.

When you give it your notes from three client projects, a transcript of a talk you gave, and the email where you explained your process to a skeptical prospect, the pool changes. Now the tool is pulling details, phrasing, and examples from material that is yours. The draft is not just structured differently. It is informed differently.

This is also why the quality of the material matters more than the quantity. Three paragraphs from an email where you explained something clearly will do more than a 5,000 word transcript where you said “um” forty times and circled the point for twenty minutes.

What counts as background material

Anything you have already written, said, or collected that contains the knowledge the draft needs. Some common types:

• Call transcripts and meeting notes. These are the richest source for most people, because you explain things more clearly and more specifically when you are talking to a real person than when you are writing for an imagined audience.
• Old blog posts or articles. Especially useful when you are updating or expanding something you wrote before.
• Client emails. The ones where you explained your process, answered a hard question, or talked someone through a decision.
• Slide decks and talk notes. These tend to be compressed, which is actually helpful. Compressed material gives the tool strong claims without a lot of filler.
• Internal documents. SOPs, onboarding docs, the thing you wrote for your assistant. These are often clearer than your published writing because you were not performing.
• Other people’s writing you want to respond to. An article you disagree with, a take you want to build on. Give the tool the source so it knows what you are reacting to.
• Data you collected. A spreadsheet of project timelines, a list of the questions clients asked in the last quarter, survey responses. Numbers from your own work are the hardest thing for a tool to invent and the easiest thing for you to supply.

How to prepare the material before you load it

Do not paste in a raw 8,000 word transcript and hope the tool finds the good parts. It will find parts. They may not be the ones you wanted.

Pull the useful sections. Read or skim the material and copy out the paragraphs, sentences, or data points that relate to the piece you are writing. If the transcript is an hour long and seven minutes of it are relevant, give it the seven minutes.

Label what you are giving it. Put a simple line at the top of each block, like one of these:

This is from a client call in which I explained why I recommend approach X over approach Y.

These are notes from a talk I gave about [topic]. The audience was [type of person].

The label tells the tool how to weigh the material. Without it, a casual aside and a core argument look the same.

Remove what should not appear. Client names, confidential details, pricing you do not publish, anything you would not want in a public draft. The tool does not know what is sensitive. You do, and this is the moment to handle it, not after the draft is written.

Mark what is fact and what is opinion. If your notes contain a number you are sure of and a guess you made on the fly, say so. “The project took six weeks (actual)” and “I think most projects like this take four to eight weeks (my estimate, not measured)” give the tool very different instructions about confidence.

How to load it: three approaches

Approach 1: Material first, brief second. Paste the prepared material into the conversation and say: “I’m going to give you some background material, then a brief. Read the material first. Do not write anything yet.” Then paste the brief as a separate message.

This works well when you have a moderate amount of material (under 3,000 words or so) and you want the tool to absorb it before it starts making structural decisions. The brief controls the shape. The material fills the shape with your knowledge instead of the tool’s.

Approach 2: Material embedded in the brief. Put the material directly into the “MY MATERIAL” section of your brief template. This works when the material is short: a few bullet points, a key paragraph, and a couple of data points. It keeps everything in one place, and the tool sees the relationship between the instruction and the content at the same time.

Approach 3: Material as a separate document. Some tools let you upload files or attach documents. If you are working with a long document (an old article, a full transcript you have already trimmed, a ten-page report), uploading it separately and referencing it in the brief is cleaner than pasting it inline. In the brief, say: “Use the attached document as your primary source. Do not add claims that are not in this document or in the brief.”

The third approach is also useful when you want the tool to pull from the material selectively rather than use all of it. You can say: “The attached transcript covers three topics. This article is only about the second one. Ignore the rest.”

The instruction that matters most

Whatever approach you use, include a version of this line in your brief:

Do not add facts, statistics, or claims that are not in the material I gave you or that I did not state in this brief. If a supporting fact would strengthen a point and you do not have one, write [NEEDS FACT] and continue.

This does not make the tool perfectly obedient. It will still drift. But it tilts the output toward using what you gave it rather than inventing what you did not, and it makes the gaps visible rather than hidden. Visible gaps are a two-minute fix. Hidden inventions are a reputation problem.

What happens when you give it too much

More material is not always better. Three problems show up when you overload:

The draft gets long and loses shape. The tool tries to use everything you gave it, because you gave it and it assumes everything is important. A 1,200 word article brief with 6,000 words of material will often produce a 2,500 word draft that wanders.

Details compete with each other. If your transcript contains three different explanations of the same concept (because you explained it three times to three different clients), the draft may blend them into something muddled. Pick the best one before you paste.

The tool emphasizes the wrong thing. If 80 percent of the material is about one subtopic, the draft will be heavy on that subtopic even if it is not the main point of the article. The tool follows volume. If you want balance, balance the material you give it, or tell it explicitly which parts matter most.

A useful rule: if you would not read all of the material yourself before writing the piece, the tool probably does not need all of it either.

What happens when you give it too little

The opposite problem is quieter and more common. You write a detailed brief with a specific reader and a clear structure, but you give the tool no material because you planned to “add your own stuff in editing.”

The draft comes back structurally sound and substantively empty. Every claim is general. Every example is hypothetical. The voice might even be right, if you have a style sheet. But there is nothing in it that could not have been written about anyone in your field.

The editing pass where you “add your own stuff” turns into a rewrite, because adding specific material to generic scaffolding means rewriting every paragraph that touches it. You would have been faster giving the tool three bullet points of real material and getting a draft that was 70 percent there.

The minimum viable background material for most articles is: one specific example from your own work, one opinion you hold that not everyone shares, and one concrete detail (a number, a timeframe, a thing that happened). Three things. They can be three sentences. They change the draft from something that sounds like a search result to something that sounds like a person who does this work.

A worked example

Here is someone writing an article about onboarding new clients.

Without background material:

READER: a freelance consultant who just started getting enough clients to need a process.
QUESTION: how do I onboard a new client without forgetting something?
OUTCOME: she has a basic onboarding checklist she can use this week.
SHAPE: about 1,000 words. Sections: why onboarding matters, what to include, mistakes to avoid.
VOICE RULES: first person, short sentences, no jargon.

The draft she gets will describe onboarding in terms anyone could have written. It will probably include “set clear expectations,” “establish communication channels,” and “define project scope.” These are not wrong. They are also not useful, because they are the generic version of advice that only becomes valuable when it is specific.

With background material:

READER: a freelance consultant who just started getting enough clients to need a process.
QUESTION: how do I onboard a new client without forgetting something?
OUTCOME: she has a basic onboarding checklist she can use this week.

BACKGROUND MATERIAL:

[Label: my current onboarding process, developed after losing a client over a miscommunication]
- I send a welcome email within one hour of signing. It has the project timeline, my working hours, and how to reach me. Before I did this, at least one client a quarter would email me on a weekend and get annoyed when I did not reply.
- I ask three questions before the kickoff call: what does done look like, who else needs to approve the work, and what went wrong with the last person who did this for you. The third question has saved me twice.
- I stopped sending a ten-page welcome packet because nobody read it. Now it is one page with five bullet points. Completion rate went from maybe 30 percent to almost everyone.
- The biggest mistake I made was not confirming the budget in writing before starting. The client and I remembered different numbers. Now it goes in the welcome email.

[Label: things I have tried that did not work]
- Project management tool access on day one. Clients were overwhelmed. Now I add them in week two.
- Asking clients to fill out a detailed questionnaire. They either skipped it or spent so long on it they resented the process. Replaced with the three questions above.

SHAPE: about 1,000 words. Sections: the problem with onboarding by memory, the five things I send or ask in the first 48 hours, what I stopped doing and why.
VOICE RULES: first person, short sentences, no jargon.
DO NOT: invent statistics. If a claim needs support and you do not have it, write [NEEDS FACT].

The second brief takes about five extra minutes. The draft it produces has real detail, a real point of view, and real examples. The editing pass is tightening and checking, not inventing.

Mistakes that cost the most time

Dumping a full transcript without reading it. The tool will use whatever is in front of it, including the part where you went on a tangent about something unrelated. Read first, pull the relevant sections, paste those.

Giving material without labels. The tool does not know that one block is from a published article and another is from a private email unless you say so. Labels take ten seconds and prevent the draft from treating a casual aside with the same weight as a core argument.

Loading material that contradicts itself. If your notes from last year say one thing and your current thinking says another, the draft will try to reconcile them and produce something that hedges in a confused way. Remove the outdated material, or label it explicitly: “This is what I used to think. I now believe the opposite, for these reasons.”

Forgetting to remove sensitive information. Client names, internal pricing, details covered by an NDA. Check before you paste. It is much easier to strip this from the input than to hunt for it in the output.

Using material as a substitute for a brief. Pasting in three old blog posts and asking the tool to “write something like this but updated” gives it material but no direction. It does not know who is reading, what question the piece answers, or what shape you want. Material without a brief produces a remix. You need both.

Common questions

How much material is the right amount? Enough to cover the main points you want the draft to make, with specific examples. For most articles, that is 300 to 800 words of prepared material. If you are going past 2,000 words of material for a 1,200 word article, you are probably giving it too much and should trim.

Can I use the same background material for multiple articles? Yes, if the articles cover different angles of the same subject. Label the material the same way each time, and adjust your brief to point the tool at the relevant sections. The material is your knowledge base. The brief is the lens.

What if my material is mostly in audio or video? Transcribe it first. Most phones and many free tools do a reasonable job of transcription now. A rough transcript with errors is still far more useful to the tool than no material at all. Clean up the parts you plan to use, and do not worry about the rest.

Should I give it material from other people? Yes, if you are responding to it or building on it, and if you label it clearly: “This is an article by someone else that I want to respond to. Here is where I agree and here is where I disagree.” Do not give it someone else’s work and ask it to rewrite it as yours. That is not using background material. That is copying.

Does the tool remember material from a previous conversation? In most tools, no. Each conversation starts fresh unless you are using a feature specifically designed to retain context (like a saved project or a system prompt). If you are starting a new conversation, paste the material again. If you are continuing the same conversation, it will still have it, but check, because long conversations can lose early material in practice.