Working from your own documents
The difference between a tool that writes like the internet and one that writes like your department. This is the module that pays off over time.
There are two ways to use these tools, and the gap between them is enormous.
Asking it to write without source material makes it rely on general patterns learned during training, and possibly search if that feature is enabled. The result may sound plausible without reflecting your organization or evidence.
Asking it to work with your material uses your gap statement, your data, your approved objectives, your last three reports. The tool supplies possible structure and wording. You supply the substance. That version is usually more relevant and easier to verify, and it’s what every recipe on this site is built around.
The instruction that does the work
Work only from the material I have given you. Do not add facts, figures, citations, guideline references, or examples from any other source. If something is missing that you need, write [NEEDS: what's missing] instead of filling the gap.
Keep that paragraph somewhere you can get to it quickly. It reduces unsupported additions, but it cannot enforce accuracy. You still need to check the result against the source.
The [NEEDS: ] bit matters more than it looks. Without it, gaps get filled with plausible filler that reads exactly like fact and is nearly invisible when you proofread. With it, you get a to-do list.
Five files, made once
The real shift is going from pasting context every time to having it already there. Build these five and you stop repeating yourself.
1. Writing samples. Two or three things you’ve written and were happy with. One report, one email, one description aimed at participants. This does more for the output than any other single file.
2. House style notes. Your conventions, written down. How objectives get phrased. Words your organization doesn’t use. Whether you say participants, learners or attendees. Preferred date format. The thing your director always changes.
3. Approved examples. Objectives that got through review. A report that came back without comment. An evaluation summary someone praised. These teach by showing rather than telling.
4. Your standing rules. The paragraph above, plus things like “never state an outcome level higher than what was measured”, “no promotional language in accreditation documents”, and your list of banned adjectives.
5. Your name-removal checklist. What you strip out before pasting, so it’s a routine rather than a decision every time. See the guardrails.
Why this is the one that pays
The setup takes time, but repeated work can become faster because the template, examples and standing rules are already available. You spend less time rebuilding context and more time reviewing the result.
That’s the compounding, and it’s the whole reason a handbook is worth having. People who give up after two weeks give up because they were paying the setup cost over and over and never collecting the return.
One caution
Working from your own documents makes the output easier to ground and verify. It does not make it checked.
The tool can still invent, misread your table, swap two figures, attach a quote to the wrong theme, or turn a careful finding into something more definite than you wrote. Source material reduces some risks. It does not remove the need to audit the numbers.
Try this
Build number 1 today. Just the writing samples, nothing else. Then re-run the last thing you asked a tool to write, with your samples attached.
That single comparison is what convinces people, and it takes about four minutes.