#Settings
What each setting changes, and which formats it changes. All of them are in the extension's options page.
Settings are applied to the conversation before any formatter runs, so they affect every format identically unless noted. Statistics are recomputed afterwards: exclude reasoning and the export's own message count drops to match.
#Content
#Reasoning (scrapeReasoning): on
Include the model's thinking, not just its answers.
Off removes those turns entirely. On, each format handles them differently:
| Format | How reasoning appears |
|---|---|
| Markdown, plain text | A labelled turn, inline |
| HTML | A <details> block, collapsed by default |
| Obsidian | A > [!note]- callout, collapsed by default |
| PDF, Word | An amber-tinted block with recoloured body text |
| JSON, CSV | A turn with isThought: true / is_reasoning: true |
| JSONL | Dropped, unless jsonlIncludeReasoning is on; see format guide |
#Images (scrapeImages): on
#Other attachments (scrapeAttachments): on
Whether to download image attachments, and whether to download everything else. Split because "keep the screenshots, skip the 40 MB of CSVs" is a common want.
The two are decided by MIME type, not extension. Anything excluded is counted,
not forgotten: the message still says 2 attachments not included in export,
and the JSON carries excludedAttachments: 2. Without that, a message whose
only content was an excluded file renders as a bare heading, which reads as a
bug.
Turning both off also skips the downloads, which is the fastest way to export a large conversation.
#Condense attachments (condenseAttachments): on
AI Studio files each attachment as its own message, so a question sent with three screenshots is stored as four messages: three carrying only a file, then one carrying the text.
On, each run of attachment-only messages is folded into the message it was sent with, matching what you actually did. Token counts are summed and the earliest timestamp wins, so no numbers change.
Off, AI Studio's own structure is preserved. Useful if you are reconciling an export against the API, and misleading otherwise: it produces headed blocks with nothing under them.
A trailing attachment with no message after it is never folded; there is nothing to fold it into, and dropping it would lose the file.
#Metadata
These control fields AI Studio's backend exposes but its page does not render. All default on; each is individually switchable because some people want a clean transcript and nothing else.
| Setting | Adds | Removed from |
|---|---|---|
includeModelConfig |
Model name, temperature, top P, top K, max output tokens | Header of every format; config in JSON; frontmatter in Obsidian |
includeSystemInstruction |
The system instruction | Its own section in every format; systemInstruction in JSON; the system message in JSONL |
includeTimestamps |
Per-message and last-modified times | Message subtitles; timestamp in JSON; the timestamp column in CSV |
includeTokenCounts |
Per-message and total token counts | Message subtitles; tokenCount in JSON; the token_count column in CSV |
includeSourceUrl |
Link back to the AI Studio conversation | Header; url in JSON; source in Obsidian frontmatter |
In JSON, a field switched off is absent, not null; null is reserved for
"AI Studio had no value". See
data reference.
CSV always emits all nine columns; a disabled field leaves its cell empty rather than dropping the column, so the file stays parseable.
#Format-specific
#jsonlIncludeReasoning: off
JSONL only. Off, reasoning turns are dropped from the dataset; training on a
model's private thinking as if it were the reply teaches it to emit thinking as
output. On, each reasoning turn becomes an assistant message and merges into
the answer beside it.
#Other
#defaultFormat: markdown
Which format the popup preselects.
#Theme (theme): system
system, dark or light. Changes nothing about an export; it is the only
setting here that does not.
It applies to every surface the extension owns: the popup, this settings page, its dialogs, and the progress overlay drawn over AI Studio during an export. The overlay follows this setting rather than AI Studio's own, so a dark extension on a light AI Studio stays dark.
system follows the operating system and updates live if the OS switches
while a page is open. Unlike everything else here, choosing a theme applies
immediately rather than on Save: it is the one setting whose result is
already on screen, so making it wait would mean picking a theme you cannot see.
#DOM fallback only
These four apply solely when the API path is unavailable and the page has to be scraped instead. A normal export ignores them, because the API returns the real file rather than what the page happened to render.
| Setting | Default | Effect |
|---|---|---|
scrapeAttachmentTitle |
on | Record the file name shown on the page |
scrapeAttachmentSize |
on | Record the file size shown on the page |
scrapeAttachmentPreview |
on | Save the page's thumbnail image |
loadDelay |
700 ms | How long to wait for content to render while scrolling |
#What settings cannot fix
Some things are limits of the format, not of a toggle:
Non-Latin text becomes ? in the PDF |
The standard PDF fonts are Latin-only. Export HTML and print to PDF from your browser. |
| An attachment shows a name but no file | The bytes could not be fetched from Drive. path is null in the JSON. |
| Everything metadata-ish is missing at once | The export used the DOM fallback. Check source in the JSON; see data reference. |