Human Experience Reform

Independent public-interest evaluations of how systems respect dignity and support agency.

HXR Evaluation

ChatGPT Queue vs Steer: Send Messages While It’s Thinking Without Changing the Current Task

Can a message update the work in progress, wait in a queue for later, or stay in a side thread that does not affect the main chat context? The August 24 contributor occurrence shows a materially positive current capability and a remaining interaction opportunity. The contributor can send mid-run input that is folded into work already underway and explicitly values that improvement. OpenAI release notes document an Update/interrupt behavior for long-running ChatGPT requests that can refine active work without restarting or losing progress. However, the contributor wanted to keep the same active task running unchanged while having a separate conversation and instead used another ChatGPT thread. Independent OpenAI-hosted community requests use the language “temporary side threads,” “do not affect the main chat context,” and “message queue while GPT is generating”; a June 2026 side-thread request received an OpenAI_Support reply saying the feedback would be passed to the team for consideration. OpenAI's Codex materials separately demonstrate ask/discuss/steer interaction during active work and explicitly use queue/steering terminology. This supports a clear reform target without establishing that ChatGPT and Codex share an implementation, that OpenAI has committed to the reform, or that the observed limitation is universal.

What this page means: Evaluations examine how a product, organization, or practice affects people. Cases and other evidence can establish strengths, difficulties, mixed outcomes, and useful improvements. Counts describe the record; they do not establish prevalence or responsibility.
Evaluation ID
HXR-EVAL-0032
Status
Open — gathering evidence
HXR Incident connections
1
Last updated
August 25, 2026

Who and what this Evaluation covers

Naming an organization, product, service, or workflow identifies the subject of this Evaluation. It does not by itself prove which internal team caused or controls the issue.

Organization in scope
OpenAI

Evaluation scope

Includes ChatGPT Text/Chat interaction while a longer-running assistant task is active: visible distinction among steering/updating current execution, queuing a follow-up for after current work, and discussing material in a side/parallel conversational lane that does not alter active work unless deliberately merged/applied. Includes message-routing clarity, preservation of active-task progress, context isolation, later merge/use, and the user's ability to keep conversing while work continues. Excludes live Voice Mode performance, generic AI multitasking, every background tool operation, Codex implementation equivalence, unsupported internal architecture claims, and whether every ChatGPT client/version exposes identical controls.

Question being evaluated

While ChatGPT is executing a long-running task, can the user deliberately choose whether new input steers or updates the active work, queues for later, or remains a parallel/side conversation outside active execution unless the user explicitly applies or merges it?

Current findings

What works well or deserves recognition

ChatGPT already supports an important class of active-task agency: OpenAI documents mid-run Update/interrupt behavior that lets users refine a long-running request without restarting or losing progress, and the contributor reports appreciating the newer ability to add information while work is underway. OpenAI's Codex product provides a positive comparator in which users can ask questions, discuss approaches, steer in real time, and in long-running-work guidance queue a next action while work continues.

Difficulties and opportunities to improve

In the contributor-observed ChatGPT interaction, a new message sent while work was underway functioned as input to the active work; there was no separate user-controlled conversational route for “keep doing exactly what you are doing while we talk about this.” The contributor therefore moved the side discussion to another thread. Public requests separately ask for queued follow-ups and temporary side threads that do not affect main chat context, indicating that users distinguish these semantics rather than treating every message during execution as one class.

Mixed findings and limits

The three message intents need not share one UI or technical pathway. Steering current work, queuing a future follow-up, and maintaining an isolated side conversation have different effects on execution and context. A product may legitimately constrain any route where safety, consistency, tool state, or context integrity requires it, but the user should be able to understand what will happen before sending. OpenAI's existing ChatGPT Update and Codex interaction patterns are feasibility/benchmark context, not proof that one implementation can simply be copied.

How HXR would test performance

Begin a ChatGPT Text Mode task that remains active long enough for additional interaction. The user can intentionally route a new message with understandable semantics: (1) steer/update the active task, preserving progress; (2) queue the message as a follow-up for after the current task without changing current execution; or (3) continue a side/parallel conversation that does not alter the active task or main task context unless the user deliberately applies/merges selected material. The active task remains running when the user chooses a non-interrupting route, routing state is visible or otherwise clear, and conclusions from the side conversation can later be used for the next task or explicitly merged without accidental context pollution.

Evidence coverage and limits

Public evidence spans December 2024 through August 24, 2026; current contributor occurrence August 24, 2026. Coverage describes evidence chronology, not prevalence.

Connected HXR Incident reports

1 public HXR Incident report is connected to this Evaluation. Canonical roles: 1 supports; 0 partially supports; 0 narrows; 0 context only; 0 challenges. Connections are not votes and do not establish prevalence.

Next observation or verification

Monitor current ChatGPT releases and official OpenAI responses for explicit queue, side-thread, branch/merge, or parallel-conversation behavior. Treat any verified implementation as reform/positive performance while preserving the Evaluation for durability, improvements, and regression. Continue adding independent reports without converting case count into prevalence.

Technical record details
Evaluation type
Interface
Benchmark posture
Developing Evidence
Blue Score readiness
Developmental Data Only
Public revision
1

Help test this Evaluation

Add your experience

Your experience may support, challenge, update, or add a positive example to this Evaluation. A short account is enough to begin; screenshots and files are optional.

Include this Evaluation number in your submission: HXR-EVAL-0032.

Open the submission form

The form opens on Jotform and sends your material to an HXR operator by email. It does not automatically create a case, publish anything, or contact the organization.

Remove passwords, authentication codes, payment details, and unnecessary personal or confidential information. Describe sensitive evidence rather than uploading it here.

A different experience? Share a new topic.