Summary
Ask ChatGPT reply feature, where selecting text from a response creates a reference chip above the input box, displays raw markdown syntax instead of rendered text. Bold markers, headers, table pipes, and dividers all show up as literal characters (*, #, |, -, =) rather than the clean formatting visible in the conversation itself. This is confirmed across two distinct content types: divider-separated text and full markdown tables.
Problem:
When you select part of a response and use the Ask ChatGPT feature, the chip that appears above the input box shows the raw markdown source of that selection such as asterisks, hash marks, table pipes, dividers, etc. (*, -, =, |, #), instead of the clean text you just read in the conversation.
-
In any response with markdown formatting (a heading, bold text, a table, or a — divider), select a span that includes that formatting.
-
Trigger the Ask ChatGPT action.
-
The chip above the input box renders the literal markdown characters instead of formatted text.
-
Confirmed with two distinct cases: a divider-separated text block and a full markdown table with headers and bold values.
Why this matters:
Users need to see what they’re about to reply to or ask about. That only works if the chip looks like the thing it was copied from. Right now it doesn’t - a bolded table cell becomes **text**, a section header becomes ### text, and other symbols like ===, ***, | , and — show up unrendered too. For anything table-shaped, the reply chip turns into a wall of pipes and dashes that’s genuinely hard to parse. The feature is supposed to reduce ambiguity about what’s being referenced; instead it introduces a confusion moment.
Suggested fix:
Have the chip render through the same markdown pass as the conversation view, or fall back to a stripped plain-text version if a lightweight render isn’t feasible in that component.
Thanks for taking the time to look into this, happy to share more repro steps or screenshots if useful. Flagging it here in case it’s helpful for the team.