Five Changes to Reflacto's Prompt, Six Models: What Really Helps
In the model test, the biggest jump came from my own prompt. So I tested five more changes, on their own and together, with six local models and 9,348 answers.
Five targeted changes to Reflacto's AI instructions.Afterwards five of six models answer better, all but the smallest.
Average score per local AI model, 1 to 10, rated blind. The number next to each model is the change.
With a local AI model, Reflacto helps you at every step of a reflection, for example with “Help me say it” or “Give me ideas”. Behind each of these eight helps is a prompt, an instruction to the model. In the answers from the model test, I found five mistakes the models kept making. I wrote one change against each, three for a single help and two for all eight, and tested them with six models.
An example: a teammate has missed several handover deadlines. The step asks “How can you phrase feedback constructively and solution-focused?”, the field is still empty, and the person picks “Help me say it”.
Previous prompt
Since you haven't written your phrasing yet, I cannot provide a draft for you to use. Please describe how you would like to open the conversation …
With the change
“Felix, the handover for the search indexing work was delivered on [Tuesday] instead of [Friday], and I didn't receive a message beforehand. Because of this, Priya could not start the indexing work as planned and had to move two of her own tasks to the next sprint. …”
The short version
- From now on, Reflacto uses all five changes. That is exactly the version that ran in the test, and it meets every rule I set beforehand.
- Three changes measurably work. “Help me say it” no longer refuses, “Give me ideas” delivers the kind of ideas the step asks for, and small models stop copying templates.
- What works names a specific mistake. A general rule like “don't ask for something that is already there” changed almost nothing.
- The smallest model gets no better with all five. Below a certain size, a better prompt barely helps.
Three changes measurably help, and so do all five together.
- meets every rule set beforehand
- all five together, also proven
- not proven
Large dot: average of the six models. Bar: margin of error. If it reaches zero, a better score is not proven. Small dots: the individual models. “Shape in words” still counts as proven, because it was meant to cut template leftovers, not to raise the score.
- Qwen3.8-27B
- Gemma 4 26B-A4B
- Bonsai 2 27B
- gpt-oss-20b
- Gemma 4 E4B
- LFM3-1.7B
- Facts or interpretation?own answer form for “no judgment”+0.23-0.21 to +0.65
- Help me say itnever refuse, invent nothing+0.66-0.17 to +2.28
- Give me ideasideas of the kind asked for+0.41-0.02 to +1.22
- Don't ask againdon't ask for what's written+0.04-0.07 to +0.25
- Shape in wordsGoal: fewer template leftovers (3.3% → 0.7%)+0.04-0.12 to +0.23
- All five togetherall changes at once+0.35-0.03 to +0.59
The five changes
What it looks like in Reflacto
A step of the SBI method
What behavior did you observe?
My manager reviewed my pull request three times and asked for status updates in chat twice a day.
Thinking partner
What would help you most right now?
- Facts or interpretation?Check what a camera would have seen
- What am I missing?Blind spots and open points
- Ask me a questionOne question that takes you further
All eight helps
- Help me get startedThree questions to get you going
- Ask me a questionOne question that takes you further
- What am I missing?Blind spots and open points
- Facts or interpretation?Check what a camera would have seenchanged only here
- See it from their sideHow the other side might see it
- Give me ideasOptions you may not have consideredchanged only here
- Help me say itWording you could say or sendchanged only here
- How might it land?Read it through their eyes
“Don't ask again” and “Shape in words” change the prompts of all eight helps.
- Facts or interpretation? Observable actions such as “interrupted me” no longer count as judgments.
- Help me say it. With an empty field, the model drafts from earlier answers instead of refusing. It makes up no numbers, dates or names.
- Give me ideas. When the step asks for downsides, it gets downsides, not ways to avoid them.
- Don't ask again. No questions about what is already written, no yes/no questions.
- Shape in words. The answer form is described instead of shown as a template with
<placeholders>.
What got better
- before
- less often with the change
- more often with the change
- Judgment found where there is noneFacts or interpretation? · fixed check75.0 → 45.2%
- No wording with an empty fieldHelp me say it · fixed check13.0 → 0.0%
- Made-up contentHelp me say it · per judge13.4 → 7.6%
- Ideas of the wrong kindGive me ideas · per judge19.4 → 11.4%
- Yes/no questionDon't ask again · fixed check12.0 → 5.9%
- Asks for what is writtenDon't ask again · per judge4.0 → 4.3%
- Template leftoverShape in words · fixed check3.3 → 0.7%
With an empty field, “Help me say it” refused in one answer in eight before, now never. Ideas of the wrong kind drop from 19% to 11%, template leftovers from 3.3% to 0.7%. For “Facts or interpretation?”, false alarms drop from 75% to 45%, but by my rules that falls just short of proof. The models still ask about what is already written as often as before.
Every model reacts differently
- the previous prompt
- all five changes
- Qwen3.8-27BUD-IQ3_S5.85 → 6.38
- Gemma 4 26B-A4BUD-IQ4_XS5.72 → 6.31
- Bonsai 2 27BQ2_05.59 → 5.98
- gpt-oss-20bMXFP45.31 → 5.53
- Gemma 4 E4BUD-Q4_K_XL5.12 → 5.49
- LFM3-1.7BQ4_K_M3.38 → 3.35
Gemma 4 26B gains 2.3 points with “Help me say it”. This is the model that refused with an empty field in the model test, so the prompt was the cause. Gemma 4 E4B, my recommendation for laptops, gains 1.2 points with “Give me ideas”. For LFM3-1.7B, all five together bring nothing.
If you use a local model, you get the new prompts automatically. The local AI guide shows how to set one up.
How it was measured
- 147situations, each in German and English
- 2356prompts, as the app sends them
- 31,558requests per model: previous prompt, each change, all together
- 46models from 1.7 to 27 billion parameters
- 59,348answers, no request failed
- 66,420of them rated blind by Claude
I ran 47 situations in German and English through six models on my PC (GeForce RTX 5070 Ti, 16 GB), each change on its own and all five together against the previous prompt. Claude rated the answers blind from 1 to 10. A second Claude instance gave almost the same scores for 645 answers (correlation 0.93). I decided before the first run when a change counts as better and did not change that afterwards.
An AI did the rating, not a person. The situations come from the same AI that proposed the changes, so I froze them before the run. Every request ran only once.
Try it yourself
Reflacto is free, needs no account and runs in your browser. Your reflection stays on your device.
Start a reflection
