First, find out whether it has the post
ChatGPT has web search. A source returned through search may be the original page, a snippet, or another page discussing it. For a specific X post, ask ChatGPT to identify which text it actually retrieved.
Use a short request: “Read this post and summarize its main claim. If you cannot retrieve the full text, tell me what you could see.” Then add the X URL.
Check whether the answer refers to the post's actual wording. If it only describes the topic or identifies the account, provide the source text before continuing.
When the link fails, supply the text
If you can open the post in X, copy its text and original URL directly. For a supported public link you cannot read there, try CarryFeed's Twitter Viewer. Check the returned author, date, text, and original URL. Copy full text gives you the wording; Export → Copy as Markdown includes source details.
Paste the material into ChatGPT with one task. Text plus the original URL works for a simple summary; keep the publication date when the wording refers to today, next week, or a rollout. A screenshot is useful for layout or visual context, while copied text is easier to compare word by word.
- Read the returned text through to the end, especially if the original preview had Show more.
- For a thread, include the posts your summary needs in order and retain their source links.
- For a chart or photo, attach the image when your question depends on it. A media URL or filename supplies location, not the visual contents.
- For a video, provide a transcript or the relevant quoted passage for a text task. CarryFeed's X media metadata does not include a complete video interpretation.
One captured post, one precise task
The example uses OpenAI's May 13, 2024 announcement. A direct request to the public-post endpoint used by CarryFeed's reader returned HTTP 200 and 162 characters of post text on October 15, 2026. The returned author was @OpenAI, with a publication time of 21:23:14 UTC on May 13, 2024.
A separate browser check that day submitted the same URL in CarryFeed's viewer with JavaScript enabled. It displayed the post text, author, publication date, and source link. That confirms the reader handoff for this example; the model run used the separately captured text.
The first sentence reads “All users will start to get access to GPT-4o today.” The remaining sentence places new voice and vision capabilities in the following weeks and names ChatGPT Plus as their audience.
For the recorded run, Codex received the complete returned text, author, publication date, and original URL, followed by the task below. This is a Codex-assisted source-note example. The retrieval was a separate public HTTP request, so the example evaluates the supplied-text summary rather than ChatGPT's ability to open X.
Summarize this historical announcement in three bullets for a research note. Distinguish access starting on the publication date from voice/vision planned for the following weeks, preserve which users the second statement concerns, and include the original URL. Use only the supplied text. Do not describe current availability.
The first output
The run produced the three bullets below, reproduced as generated.
- On May 13, 2024 (UTC), OpenAI announced that all users would start getting access to GPT-4o that day.
- The new voice and vision capabilities demonstrated that day were planned to begin rolling out to ChatGPT Plus in the following weeks.
- Original announcement: https://x.com/OpenAI/status/1790130708612088054
Check the conditions that change the meaning
The useful check is whether the note preserves the announcement's scope. Smooth prose can still turn a staged rollout into a promise that everyone already has access.
The first bullet retains the historical date and the start of access. The second preserves both the later timing and Plus audience. The third keeps the original post attached. No wording change was needed for those points in this output.
If your question is whether your own account has received a feature, this post supplies no account-specific answer. Use current product guidance or your account's available controls for that next question; keep it out of the historical summary.
| Source condition | What the output preserved | An error to catch |
|---|---|---|
| The post was published May 13, 2024. | The date replaces the relative word today. | Reading today as the day you copied the post. |
| Access was beginning for all users. | Would start getting access. | Saying every account already had access. |
| Voice and vision were planned for the following weeks. | Planned to begin rolling out later. | Merging those capabilities into the same-day access statement. |
| The later rollout named ChatGPT Plus. | ChatGPT Plus remains the audience of the second bullet. | Extending the later promise to all users. |
| One original status URL identifies the source. | The same URL appears after the claims. | Keeping a reader's homepage or a similar announcement as the citation. |
Use a connection for repeated source lookups
If you regularly discuss X sources, CarryFeed MCP brings the lookup tools into your AI client. It uses CarryFeed OAuth at `https://api.carryfeed.com/mcp`. Follow the official ChatGPT connection guide for the client setup; account and workspace policies can affect custom connections.
After connecting, ask the assistant to fetch the public post by its ID or resolve its URL, then summarize the returned text with the source link. For the announcement above, `get_public_x_post` takes ID `1790130708612088054` and optional handle `OpenAI`.
For a known X link, use `resolve_public_x_input` with the original URL. When you only remember an account, phrase or date, `search_public_x_posts` can find candidate posts. Open the relevant result before asking for a summary; the search result is a bounded set, not a complete account history.
For the returned source fields, use the agent metadata guide. Public X lookup tools are free. Some media tools use your CarryFeed Tools balance or prepaid Instagram lookups; check current media prices before asking the assistant to run one. Your AI client may charge separately.