Most AI workflows are easy at the first draft and messy after that. The hard part is not getting something. The hard part is improving one part without breaking the rest.
That is why annotations matter. They turn revision into a more local, controlled workflow instead of forcing you to restart or reprompt the whole artifact.
AI Search Snapshot
Codex annotations let you point to a specific part of a site, document, chart, slide, or other output and ask Codex to refine only that selection. OpenAI’s June 2, 2026 product article positions annotations as a way to improve work after the first draft, when teams need judgment, iteration, and human review instead of a full reset.
Direct Answer
Use Codex annotations when most of the artifact is already useful and you only want to improve one part of it. OpenAI’s examples include updating a site navigation font, asking where a claim in an investment thesis came from, and making a chart label clearer on a slide.
The practical value is not just speed. It is preserving the parts you already like while focusing revision on the exact place that needs human judgment and improvement.
Annotations Workflow Table
| Focus | What it means | Best fit | Review gate |
|---|---|---|---|
| Best use | Refine one selected part | Ideal when most of the work is already acceptable. | A human review step still checks whether the local change creates a broader problem. |
| Artifact types | Sites, documents, spreadsheets, slides, and code outputs | Good fit for knowledge-work revision after the first draft. | Review the surrounding context before accepting the update. |
| Main benefit | Faster revision without restarting | Useful for teams iterating on claims, labels, layouts, or wording. | Human review should confirm that the new version is still accurate and coherent. |
| Bad use | Assuming a targeted edit validates the whole artifact | Annotations improve revision, not full verification. | Keep a human review step on the whole output when accuracy matters. |
Evaluation Criteria
- Use annotations when the artifact is mostly right but locally weak.
- Treat annotations as a revision tool, not a substitute for verification.
- Preserve useful parts of the artifact instead of restarting the whole workflow.
- Keep human review visible when the selected change affects claims, numbers, or decisions.
What OpenAI Means by Annotations
OpenAI’s June 2, 2026 product article says annotations let you point to the exact part you want to refine and tell Codex what needs to change. The article explicitly extends that beyond code into documents, spreadsheets, and slides. That is the key shift: annotations are not just a developer convenience. They are a knowledge-work editing pattern.
Why This Is Better Than Starting Over
Restarting a whole artifact often destroys useful work along with the broken part. Annotations create a narrower loop. You keep the draft, mark the weak point, and revise with more precision. That is especially helpful when the main problem is a claim, a label, a section heading, a chart, or one design element instead of the whole document or app.
When Human Review Matters Most
Annotations are strongest when paired with human review. If the selected item is a claim, a metric, a legal sentence, or a shared site element, the team still needs to verify the broader context. A local fix can still create a global problem. The win is more focused iteration, not zero-review automation.
Where to Go Next
If annotations are the workflow you need after the first draft, this article is the right starting point. If the real need is a shared internal app, continue to the Sites guide. If the challenge is getting the right tools and admin controls behind the workflow, continue to the plugins guide. For the broader June 2 story, return to the Codex knowledge-work hub.
Review Checklist
- Use annotations when the artifact is mostly useful already.
- Select the exact weak point instead of restarting the whole draft.
- Keep human review visible when claims, numbers, or design standards matter.
- Do not confuse targeted revision with full verification.
- Use annotations to speed iteration without losing the parts you already trust.
Bottom Line
Codex annotations are best understood as a revision workflow for the post-draft stage.
They create more focused iteration, but they still work best when human review owns the final judgment.
FAQ
Are annotations only for code?
No. OpenAI’s June 2, 2026 article explicitly extends annotations to documents, spreadsheets, slides, and sites.
Do annotations verify the whole document?
No. They help refine one selected part, but a human review step should still assess the broader artifact.
When should I use annotations instead of a new prompt?
Use annotations when most of the output is already good and one section, chart, claim, or element needs revision.
Why do annotations matter for teams?
Because teams often spend more time revising and approving work than generating the first version.
Verified External Sources
- OpenAI: Codex for every role, tool, and workflow
- OpenAI: Codex is becoming a productivity tool for everyone
Related 3RK Guides
- Codex for Knowledge Work: What OpenAI’s June 2, 2026 Shift Means for Teams
- What Are Codex Sites? When to Use Them for Internal Tools and Shared Work
- Codex Plugins Explained: How Role-Specific Workflows, App Access, and Admin Controls Fit Together
- How Non-Developers Use Codex: Research, Dashboards, Briefs, and Internal Apps
- AI Governance Operating Model