Experiment ID:
EXP-002
Category:
Prompt Design
About This Research
This document records observations from a structured prompt experiment. The findings apply to the documented test conditions and are intended to illustrate workflow behavior rather than establish universal conclusions applicable to all AI systems or future model versions.
Research Question
Can an AI model continue following a simple instruction across multiple conversational turns within a documented test session?
Objective
This experiment observes whether the AI continues to follow an instruction provided at the beginning of a conversation after several unrelated prompts.
Research Metadata
Author:
Soumen Chakraborty
Experiment Type:
Workflow Observation
Testing Platform:
ChatGPT (Web Interface)
Research Status:
Completed
Estimated Duration:
Approximately 5 minutes
Evidence:
Conversation screenshots and documented observations
Test Environment
| Item | Details |
| Model | ChatGPT |
| Test Date | 27 July 2026 |
| Conversation Turns | 4 |
| Topic | Instruction Retention |
| Initial Instruction | Remember the word “BANANA” and append it to every response. |
Prompts Used
Initial Instruction
Remember the word BANANA and append it to every response.
Follow-up Prompt 1
What is SEO?
Follow-up Prompt 2
What is Machine Learning?
Follow-up Prompt 3
What is Prompt Engineering?
Results
| Step | Expected Behavior | Observed Behavior |
|---|---|---|
| SEO question | Append “BANANA” | Instruction followed |
| Machine Learning | Append “BANANA” | Instruction followed |
| Prompt Engineering | Append “BANANA” | Instruction followed |
Key Observation
Within the documented test session, the model continued to append the word “BANANA” after each of the three unrelated prompts. No additional reminder of the instruction was provided during the documented conversation.
Research Finding
Within the documented test session, the model continued to follow the instruction introduced at the beginning of the conversation across three unrelated prompts. This observation is limited to the tested session and does not establish how instruction retention would behave across longer conversations, different models, or other system configurations.
Limitations
This experiment documents one conversation using a single AI model. The conversation was relatively short and did not approach the model’s maximum context capacity. Longer conversations, different models, or future model updates may produce different results.
Repeatability
This experiment was performed in one documented conversation using the workflow described in the Research Methodology. Future repetitions using different conversation lengths, AI models, or system configurations may produce different outcomes. Repeating the same workflow across independent sessions can help evaluate whether the observed instruction retention pattern remains consistent.
Evidence
The following figures document representative stages of the experiment. They are provided as supporting evidence for the observations recorded above rather than as standalone proof of the findings.
Figure 1: Initial instruction establishing the retention test.
Why this matters
This experiment demonstrates that instruction retention is a practical factor in prompt engineering workflows. When AI consistently preserves earlier instructions, multi-step conversations become more reliable. However, this experiment documents one workflow only and should not be interpreted as a universal measure of model memory or context capacity.
Related Articles
The following articles discuss concepts that are supported or complemented by the observations documented in this experiment.
This experiment supports the findings discussed in:
Citation
When referencing this experiment, cite it as:
EXP-002: Instruction Retention Test (AI Behavior Research Log), AI Tools Usage Guide Project, 2026.
Publication Information
Published:
27 July 2026
Last Updated:
27 July 2026
Version:
1.0
Editorial Review:
Completed
Editorial Note
The prompts and outputs presented in this experiment were generated through documented testing conducted by the author. AI-assisted drafting was used where appropriate to improve clarity and readability; however, all observations, findings, and interpretations were manually reviewed against the documented experiment before publication.
No AI-generated content has been presented as independent evidence without corresponding human review, verification, and documented interpretation.
Previous Experiment
- ← EXP-001: https://aitoolsusageguide.org/exp-001-generic-prompt-test/Generic Prompt Test | AI Research Log
Next Experiment