Stop ChatGPT Drifting With One Sentence

You start a chat with a clear goal. Twenty messages later ChatGPT is somewhere else entirely, confidently solving a problem you never asked about. (You looked away for one second and it wandered into traffic.)
Long conversations drift. One sentence keeps it on the leash.
State the goal and tell it to hold the line
Early in the chat, write one sentence that locks the target:
"Our goal for this whole conversation is X. Before each answer, check that it moves us toward X."
Now the model has a fixed point to measure against, and it stops chasing tangents.
Why drift happens
ChatGPT responds to the most recent messages most strongly. As a chat grows, the original goal slides toward the back of its attention and the latest detour takes over. It is not forgetting on purpose. The recent stuff just shouts louder.
Drift is the model following the conversation instead of the goal. Give it a goal to anchor to and it stops chasing the last shiny thing.
When you feel it wandering, paste the goal sentence again. It is a cheap reset that snaps everything back into focus.
One sentence up front, one reset when needed. Keep it on track and go finish the actual task.
Need a CTO, not another hire?
Fractional CTO. Technology leadership, part time. A weekly leadership call, architecture and code review, a written monthly plan, and on call when things break. $4,000 a month for 20 hours.
See the engagement