◉GEMINI LABJP
●ARGON — gemini-4-argon is generally available as of 10/13, aimed at complex coding, agents, math and science●DEEP RES — deep-research-pro-preview-12-2025 shuts down 10/23, 8 days left. Move to deep-research-preview-04-2026●NANO — gemini-nano-banana-2.1 is GA (10/06). gemini-3.1-flash-image is deprecated with no shutdown date yet●MCP — One write-up cut an 85,000-token tool result to 310 tokens by wrapping it in an ADK FunctionTool●FLASH — gemini-3.7-flash and 3.5-flash now route automatically to newer Flash models (10/08)●NEW — Stop Re-Reading Every Crash Report: Gre…●ARGON — gemini-4-argon is generally available as of 10/13, aimed at complex coding, agents, math and science●DEEP RES — deep-research-pro-preview-12-2025 shuts down 10/23, 8 days left. Move to deep-research-preview-04-2026●NANO — gemini-nano-banana-2.1 is GA (10/06). gemini-3.1-flash-image is deprecated with no shutdown date yet●MCP — One write-up cut an 85,000-token tool result to 310 tokens by wrapping it in an ADK FunctionTool●FLASH — gemini-3.7-flash and 3.5-flash now route automatically to newer Flash models (10/08)●NEW — Stop Re-Reading Every Crash Report: Gre…
Articles/Gemini Basics
◉ Gemini Basics/2026-09-13Beginner

I edited the Drive doc, asked Gemini again, and it quoted the old sentence back

When Gemini answers from an outdated version of a Drive file, the file usually isn't the problem — the conversation is. Here's how sources stay attached for the whole conversation, why you can't remove one mid-thread, and what to do about partial reads.

Gemini91Google Drive5sourcesWorkspace5troubleshooting86

I keep client drafts in a Google Doc on Drive, passing revisions back and forth. The other day I folded in a round of edits, rewrote a whole paragraph, and then asked Gemini to summarize the current draft in three lines.

One of those three lines contained the sentence I had just rewritten away.

I reopened the document. The edit was saved. Sharing hadn't changed. Gemini still handed me the old wording. My first assumption was lag, so I waited a few minutes and asked again, twice. That got me nowhere.

It wasn't lag. The file wasn't stale — the conversation was.

A source stays attached for the whole conversation

There's a short line in Gemini's help about how sources behave. When you add a source, it remains active for your entire conversation, giving Gemini consistent context for its responses.

Consistency is usually a gift. When you're asking ten questions about the same document, you don't want to re-attach it every time.

But "remains active" also means that the content as it stood when you attached it settles into the conversation and stays there. I had been thinking of a source as a pointer to a file. It behaves less like a pointer and more like context the conversation is already holding.

So when you edit the document and ask again in the same thread, the pre-edit content is still sitting there as raw material. The file moved forward; the conversation didn't. That gap was my three lines.

For reference, here's where the attach controls live. In Docs or Slides, open Ask Gemini in the top right, click Add sources at the bottom of the panel, then Add from Drive. From Drive itself, click Ask Gemini in the top right and add under Your sources in the left panel. Either way, what you add is bound to the open conversation.

You can't detach a source mid-thread

The help continues. Because Gemini can reference sources from a previous turn, the guidance for excluding a source you already included is to start a new conversation and pick what you want in your source settings.

In other words, there is no "remove this from the thread" operation. You can untick a box, but content handed over in an earlier turn can still be referenced.

That stopped me for a moment the first time I read it. The interface shows a source list you can add to and delete from, so it looks like a live roster. What you're actually looking at is a list of what to prioritize next, not a way to retract what already entered the conversation.

Consistency and freshness rarely come together. Pin a source to the thread and you stop re-attaching it, but the thread stops tracking your edits. Re-read the file on every turn and you stay current, but the ground shifts under a discussion in progress. Neither choice is wrong, and Gemini has picked the first one — once I read it that way, I adjusted my own routine instead of arguing with the product.

When I ask about a document I just edited, I start the conversation over. Deciding that one thing removed all the guessing.

It also helps to know what a fresh conversation turns on by default.

Search locationDefault in a new conversation
DriveOn
GmailOn
ChatOn
CalendarOn
Web searchOff

Drive, Gmail, Chat, and Calendar are all live unless you say otherwise. Web search is the one that's off. If you don't know that and you ask Gemini to "use only this document," a neighboring email thread can quietly join the answer. Open your source settings once, before the first question, and narrow the locations down.

"It read everything" is an assumption, not a guarantee

If a fresh conversation still gives you an answer that doesn't line up, there's a second thing to check: how much actually got read.

The help states that each request has a limited processing capacity, and that if you have too many sources or include sources with too much text, Gemini might base its response on some, but not all, of the content.

This failure is silent. No error, no warning. I once handed over a long meeting record and asked for the decisions made in the back half, and got a summary of the front half. At the time I assumed my instructions were sloppy. Looking back, the more likely story is that the back half never arrived.

The remedies are unglamorous.

  • Send less. Detach the sources that aren't relevant and rebuild the conversation
  • Split the file. Break a long document into sections and hand over only the part you're asking about
  • Scope a spreadsheet by range. Highlight the cells first, then ask — and note that ranges only cover the sheet tab you're currently on

All three are the same move phrased three ways. More context does not reliably mean a better answer. I believed the opposite for a long time, adding every file that looked adjacent and expecting the answer to firm up.

Open the "Sources" list every time

Under each response, Gemini lists the files that contributed. Making a habit of opening that list is what turned "this answer is wrong somehow" into "I can see roughly why."

The help attaches an honest caveat to it. Gemini might miss a source it actually used, it might cite a document it didn't draw on, and — this is the heavy one — it can make up a source and present it as factual even when it wasn't part of the request.

So the list is evidence to check against, not the answer itself.

My compromise is to verify selectively. When a summary is going to become text I send to a client, or the basis for a decision, I open the original document and read the passage with my own eyes. Verifying everything costs more time than I have, so I decided in advance which situations are worth it.

Reading the symptom

Here's the triage table I actually use.

SymptomFirst suspectWhat to do
An edited-out sentence keeps appearingStale source held by the conversationStart a new conversation and re-select sources
Unrelated material bleeds into the answerDrive and Gmail on by defaultNarrow search locations in source settings
The back half of a document is missingPer-request capacity limitDetach sources and split the file
"You need access to some files"View permission on the fileGet view access, then add it again
A PDF can't be selected as a sourceThe PDF isn't on Drive yetUpload it to Drive first

The file types you can attach as sources are Google Docs, Sheets, Slides, and PDFs — and a PDF has to live on Drive before you can pick it. You also need view permission on anything shared with you, which is worth remembering before you spend twenty minutes hunting for a subtler cause.

Take the detour once

If you try one thing from all of this, make it this one.

Don't continue the thread you have open. Start a new conversation, re-attach the document you edited, and ask the identical question again. If the answer changes, the conversation was the problem. If it doesn't, you've narrowed things down to capacity or permissions, which is real progress.

It feels like a detour the first few times. Measured against the minutes I used to lose squinting at an answer after every round of edits, it has been much cheaper.

I spent a while convinced it was just slow to sync. If this saves someone else that particular stretch of waiting, I'll be glad.

Share

Thank You for Reading

Gemini Lab is ad-free, supported entirely by members like you. We publish practical guides daily with implementation code, benchmarks, and production-ready patterns. If you've found it useful, we'd love to have you on board.

  • ✦Copy-paste ready implementation code
  • ✦New advanced guides published daily
  • ✦$5/mo or $10 for lifetime access
View Membership →

If you found this article helpful, a small tip ($1.50) would mean a lot to us. Your support helps keep this site ad-free and covers server and hosting costs.

Related Articles

◉ Gemini Basics2026-10-08
Free Tier Shows Up, but You Get a 429 With `limit: 0` — What to Check in the First Hour of a New Gemini API Project
A new Gemini API project says Free tier, yet the very first call returns 429 with limit: 0. Here is the order I check things in, how to read the error body, and a 40-line script that tests models one by one.
◉ Gemini Basics2026-08-16
The Strongest Gemini Is in Preview and the Cheap One Is GA. That Ordering Should Drive Your Model Choice
Sort the Gemini lineup by release stage instead of capability and the order inverts: the strongest reasoning model sits in preview while the fast, inexpensive Flash models are GA. Here is how a solo developer handles that, plus a 30-line script that counts how many preview models your code already depends on.
◉ Gemini Basics2026-04-27
Mastering Custom Instructions in Gemini Gems — What Actually Works After Two Months of Testing
Custom instructions in Gemini Gems can swing response quality wildly. After running half a dozen Gems for two months, here's what actually moves the needle — and what wastes characters.