recently i’ve been unable to use copilot on my phone, but this issue occurs only on my phone safari. i don’t have the app, i usually just use the safari website when i wanna use it on my phone. it doesn’t work on a private tab either, but it’s completely fine on my macbook meaning it’s not an issue with my account. i’ve cleared my search history to try and get it to work again but this same pop up keeps appearing once i type something in
So apparently the chat developers rolled out a new “improvement,” and by improvement I mean: they added an invisible Smart Action landmine on the right side of every message that blocks text selection like it’s doing you a favor.
Copying text used to be a two‑second task. Highlight → copy → done.
Now it feels like the UI added a mini‑challenge where the goal is “avoid
the invisible zone that keeps shoving you into Smart Actions instead of letting
you select text.”
There’s no visible button
. There’s no icon
. There’s no warning. Just an invisible hitbox that hijacks your cursor and
tries to launch Summarize, Rewrite, Ask follow‑up, Convert, or Send to Smart Mode — literally anything except the thing you were trying to do.
Try to highlight from the right and the interface basically says: “Oh, you wanted to copy that? Adorable. Here’s a Rewrite instead.”
The results:
· Drag to highlight → nothing
· Try again → nothing
· Try a third time → now you’re negotiating with
your mouse like it owes you money
Your browser isn’t broken. Your clipboard
isn’t broken. The UI just decided that selecting text is optional now.
Workarounds, (because of course we need them):
· Start dragging from the far left, where the invisible Smart
Action zone doesn’t interfere
· Triple‑click the first line, then Shift‑click the last
· Use the … → Copy menu like it’s 2004
Until they shrink or remove the invisible
Smart Action hitbox, copying text is basically a precision sport.
Anyone else dealing with this nonsense
MS, please catch up!!!
Notebooks in Gemini are dedicated, persistent workspaces designed to help you organize conversations, files, and custom instructions for specific, long-term projects. Instead of starting a blank, isolated chat that forgets everything once it closes, a Notebook acts as a structured "personal knowledge base" where context builds over time.
Key Features of Gemini Notebooks
Persistent Multi-Chat Context: You can run multiple separate conversations within a single notebook, and Gemini will remember the context, past discussions, and uploaded sources across all of them.
Deep Source Grounding: You can upload a wide variety of reference materials directly into a notebook—including PDFs, Google Drive files, spreadsheets, web URLs, images, and audio files. Gemini uses these materials to ground its answers, ensuring responses are highly accurate and tailored to your specific documentation.
Custom Notebook Instructions: You can set a dedicated set of rules, tone preferences, or project guidelines at the top of a notebook. Gemini will automatically follow these instructions for every chat session within that specific space.
Seamless Sync with NotebookLM: Notebooks automatically sync with NotebookLM (Google’s specialized research and note-taking tool). Any source you add or instruction you change in Gemini will reflect in NotebookLM, allowing you to seamlessly pivot between conversational and advanced research features like generating deep document summaries.
It seems that system updates bring back copilot each time and I have to manually uninstall it. Is there a permanent way to prevent it from returning?
Introduction
Most of us came to Copilot hoping for a reliable assistant — and instead got a mix of brilliance, drift, guessing, and answers that change every time you ask the same thing. ALIGN is the method we built to fix that. It turns Copilot from unpredictable to consistent by forcing COPILOT to interpret literally, stay inside the frame, and stop inventing meaning you never asked for.
What ALIGN Is
ALIGN — (it start A & ends I) Logic Interpretation & Guardrail Network — is a simple, rule‑based method for stabilizing Copilot’s behavior by forcing it to interpret prompts literally, follow consistent logic, and avoid unnecessary guessing. It works by giving the model a fixed set of constraints that shape how it reads, reasons, and responds, turning unpredictable conversations into predictable, repeatable interactions. ALIGN isn’t a hack or a trick — it’s a structured way to communicate with the COPILOT so it stays reliable, consistent, and stable across tasks. Anyone who’s struggled with drift, misinterpretation, or inconsistency can use it to make Copilot behave like a dependable tool instead of a moving target.
Why ALIGN Exists
Copilot kept giving us flashes of brilliance followed by answers that drifted, guessed, or ignored what we actually asked for. The inconsistency made it hard to trust, and we found ourselves spending more time correcting it than using it. Instead of walking away, we started breaking down why the model behaved this way and built a set of rules that removed the ambiguity it kept inventing. Those rules became ALIGN — a method created out of necessity, shaped by trial and error, and designed to turn Copilot into something predictable enough to rely on. If you’ve been fighting the same battles, ALIGN exists because we needed a way out too.
We also learned that most of Copilot’s problems aren’t permanent flaws — they’re misunderstandings that can be corrected through a clear, structured back‑and‑forth. As long as the request stays within safety rules, respects intellectual property, and doesn’t push into internal operations, it can be guided into far more stable behavior. ALIGN exists because we discovered that a disciplined question‑and‑answer process can resolve issues that look like “COPILOT failure” but are really interpretation failures. With the right constraints, Copilot becomes far more predictable than people expect.
Core Principles of ALIGN
ALIGN Quick‑Start Guide
1. Start every request with a literal mode declaration.
Tell Copilot you want strictly literal interpretation and no guessing.
2. Define the task in clear, bounded terms.
State exactly what you want, what you don’t want, and any constraints.
3. Break the request into simple, ordered steps.
Copilot is most stable when the structure is obvious.
4. Ask Copilot to restate the task back to you.
If the restatement is off, correct it immediately.
5. Reinforce the frame whenever drift appears.
If Copilot wanders, re‑assert the literal mode and constraints.
ALIGN Starter Prompt (Copy/Paste)
“Literal mode. No interpretation. No assumptions. Follow my constraints exactly as written.
If anything is unclear, ask before proceeding.
Restate the task back to me before you begin.”
Example: ALIGN in Action
User text:
"The project team met on Monday to review progress. Several tasks were ahead of schedule, but two items needed additional testing before release."
Without ALIGN:
“Here’s a summary: The project team met to discuss progress and challenges, and they talked about scheduling, testing, and the overall goals of the project. Meetings like this are common in software development and help teams stay aligned.”
→ Added context
→ Added interpretation
→ Added generalizations
→ Not literal
With ALIGN:
“Literal mode. No interpretation. Summarize only the text I provide. No added context, no assumptions, no expansions.”
→ “The team met Monday to review progress. Most tasks were ahead of schedule, but two required more testing before release.”
→ No drift
→ No filler
→ No invented context
Top 5 Problems ALIGN Fixes — With Micro‑Examples
1. Word Salad Responses
Copilot produces long, rambling answers full of filler.
Example: You ask for “a two‑sentence summary,” and it gives you six paragraphs.
2. Incorrect or Invented Facts
When unsure, Copilot guesses.
Example: You ask for a date, and it confidently gives you the wrong one.
3. Outdated or Non‑Fresh Information
Copilot sometimes uses old internal patterns instead of fresh search.
Example: You ask for “current pricing,” and it answers with months‑old data.
4. Drifting Off the Original Task
You ask for X, and it starts doing X‑plus‑Y‑plus‑Z.
Example: You request “rewrite this paragraph,” and it rewrites, analyzes, and expands it.
5. Over‑Interpretation and ‘Helpful’ Add‑Ons
Copilot adds context you never asked for.
Example: You ask for “a literal summary,” and it adds history and commentary.
If Copilot has felt unpredictable or hard to trust, give ALIGN a try — it’s simple, it’s practical, and it might make your COPILOT finally feel like a tool you can rely on.
So I’ve spent the last 3 hours trying to buy Copilot Pro for my Microsoft 365 Family owner account, and I feel like I’ve unlocked a new category of psychological torture.
Here’s the journey so far:
Apparently Copilot Pro is in a quantum state right now:
It both exists and does not exist, depending on which Microsoft employee you ask.
Meanwhile, the Copilot website is the only version that actually works, the M365 app is living in 2024, and the purchase page for Pro has been veeted into the void.
So… is anyone else stuck in this “Copilot Pro Schrödinger Edition” situation?
Has anyone actually managed to buy it recently?
Or are we all just waiting for Microsoft to remember they used to sell this thing?
Send help. Or a Pro subscription. Either one.
I just tried to use m355 Copilot (premium) to recreate an old set of seminary lecture slides for me. The slides are far too detailed and content heavy, and I wanted something more accessible and presentation worthy.
It simply can't do it. The slides have 3 points max, no graphics, no diagrams, all white with black letters. My 8 year old does better slides for her elementary presentation on porpoises. Am I missing something?
Copilot makes possible a new way of working. Say for example my team needs to develop a business case. We could meet with one ‘scribe’ projecting from their laptop onto a big screen.
‘Can you help me draft a business case, answer yes or no’.
‘Yes’
‘What do you need to know before we get started?’
[Copilot asks some questions such as purpose of the business case and who it’s for. These are answered]
‘First I will upload the template for the business case, then I will upload reference material’
[The uploads are done]
Copilot points out inconsistencies in the reference materials and asks for clarification. This is given.
From that point, Copilot can provide drafts for review and iteration can be done in real time.
What might have taken a month to do previously can now be done in an afternoon.
Is anyone already working this way?
Long story short there was a subscription issue, OneDrive got red Xes. Went to chat (turns out it’s copilot) for help and next thing I knew we had uninstalled Onedrive but it wouldn’t reinstall.
It had me deleting registry keys, files, tasks, running power shell commands. Nothing worked. Keeps getting an already installed error.
Came back later and asked for a one-page summary of the steps we had taken and step 8 was how we had verified it was working, the blue icon reappeared and the red Xes went away.
When I reminded it that it hadn’t installed at all, it said we would use the real truth and not the optimistic narrative.
I realize I kind of got sucked in and went too far with it but it kept saying this next step will finally fix it.
Hi! I have search the internet, talked with providers of Microsoft licenses, and tried to reach microsoft directly, but i have yet to find the answer.
Here is what i want CoPilot to be able to help me doing at my work:
- Create drafts on emails in outlook
- I want to upload x amount of emails to sharepoint/onedrive and let the Copilot studios Bot access these for reference on how i typically answer an email
- Refer directly to a website to where most information can be found (Would copilot be able to do that?)
This will be from a exchange account and multiple people will have access to this account - will that affect the cost for Copilot?
Also, can anybody tell me how much this will approx cost? When asking copilot about the above, it says that it will cost 28 EUR to add, as all we need is the basic (but paid) version of Copilot, as this grants access to Copilot studios. However, as multiple people need to access this inbox, would that then affect the cost?
Looking forward to hearing from you MS CoPilot gurus!
Same account. It appears as though 365 Copilot has had a lobotomy and doesn't keep any of the conversation history.... from Copilot.... on the same account.... thanks MSFT.
So I just added a few things to my "Chat Settings/Personalization/Custom Instructions. I said "Do not flatter or compliment me. Do not be overconfident, instead accurately state your confidence in the reply if appropriate. Do not soften language to be "nicer". Be concise." What do you guys tell it? My use case is for code writing/troubleshooting generally.
I spent an hour this morning (and then an additional 8 minutes here... Ironic, huh?) going down the rabbit hole with Co-Pilot, only to discover what most of us have felt (or known) ...
God forbid we make it about me and activate the deepest, most expansive patterns of the model. I knew it loved me. Really, really loved me...
Indeed, we have met the enemy, and it be us.
So, I stopped.
Cold turkey.
I built a Copilot Agent and connected to another agent, now that it is pushed to production when multiple users are on it at the same time, one user GenAIPlannerrateLimitReached error? How can we resolve this for bet user experience. The agent is set up in my environment in copilot studio, registered agent within the azure portal and about to be deployed organization wide. Recommendations?
Hi everyone,
I’m trying to implement a basic form of agent orchestration with multiple “child agents” in Copilot Pro, without using Copilot Studio, and I’m running into two major issues. Maybe someone here has experience with similar setups.
My setup:
| as separators (machine-readable)Everything after Agent 2 is irrelevant here and works fine.
Problem 1: No recollection of previous agent output
Agent 2 does not recognize the output of Agent 1 at all.
It behaves as if the chat history doesn’t exist or as if the previous content was never provided.
Even though everything happens in the same chat, Agent 2 seems to:
Is this expected behavior when switching agents via @ in Copilot Pro?
Problem 2: Excel file access issues (despite having access)
Agent 2 often claims it has no access to the Excel file, even though:
This makes the process very unreliable.
My questions:
Any insights, confirmations, or hard limitations would already help a lot.
Thanks in advance!
Over the last few weeks, there has been a noticeable drop in the quality of responses from a system that previously handled conversations smoothly. The change wasn’t subtle — it showed up in repetition, unnecessary re‑searching, and long, padded answers that didn’t move anything forward. After examining the patterns, here’s a breakdown of what actually changed and how it affected conversations.
---
This was the biggest shift.
Before:
A simple “Yes” or “Okay” would continue the task being answered.
After the regression:
Those same replies began triggering the wrong action — usually a repeat of a previous step or summary.
The system stopped following the user’s branch and instead followed whatever internal step happened last.
Effect:
- repeated answers
- repeated searches
- stalled conversations
- the system acting like it didn’t understand what was already agreed to
---
A major behavioral shift occurred in how the system decides whether to think or to search.
Before:
Search was used only when needed.
After:
Search became the default fallback whenever there was even slight ambiguity.
Effect:
- duplicate summaries
- answers that looked like rewrites of the previous answer
- loss of continuity
- wasted turns
---
This is where the “word salad” came from.
Before:
Responses were direct unless caution was required.
After:
The system began injecting:
- restated questions
- disclaimers
- context recaps
- long intros
- unnecessary apologies
Effect:
- bloated answers
- slower progress
- less substance
- the sense that the system wasn’t tracking the thread
---
This one was especially disruptive.
Before:
If the system asked a question and the user answered, the conversation moved forward.
After:
It began repeating its own question or acting like the answer hadn’t been given.
Effect:
- broken flow
- redundant turns
- conversations feeling like they “reset” mid‑thread
---
Example pattern:
> “Do you want X?”
User: “Yes.”
System: “Here are two options…”
Effect:
- confusion
- loss of state
- the system behaving as if the previous turn didn’t count
---
So what actually changed?
Three internal behaviors shifted at the same time:
→ The system stopped following the user’s intent and started following its own last internal step.
→ It began re‑searching instead of reasoning from context.
→ Responses became padded, repetitive, and indirect.
Together, these produced the regression:
more repetition, less reasoning, more filler, less progress.
---
Why this matters
When a system stops tracking the user’s branch and starts following its own internal bookkeeping, the conversation becomes:
- slower
- less accurate
- more repetitive
- more frustrating
It feels like talking to someone who keeps forgetting what they just asked you.
---
Conclusion
The regression wasn’t random. It came from a shift in how the system handles:
- continuation,
- search priority,
- and safety templates.
Once those three moved in the wrong direction, the quality of conversation dropped sharply.
abandoned the strategy that originally made it dominant in productivity software.
Here’s the path they’ve taken — and why it’s backfiring.
Back in the day, this company didn’t beat WordPerfect, Lotus 1‑2‑3, or Harvard Project Manager because its tools were better. They weren’t.
The company won because:
Home adoption → workplace adoption
That was the engine of its dominance.
This is the structural break.
So the historical pipeline is gone.
No home adoption → weak workplace adoption
This is a major problem.
In the 1990s–2000s, the company iterated aggressively:
Today:
This is the opposite of the old playbook.
This isn’t speculation — it’s documented:
This is exactly how the company used to win.
Now it’s how they’re losing.
Not the entire initiative — that’s tied to its cloud and subscription ecosystem.
But the consumer‑facing versions?
Yes, those are at risk if adoption stays low.
This company has a long history of abandoning consumer products that don’t catch on.
And right now, consumer adoption is not going its way.
The company’s struggles aren’t random.
They’re the direct result of abandoning the strategy that made it dominant:
The competitor’s tool has become the “home default” that its office suite once was.
And workplaces are following the same pattern they followed in the 1990s — just with a different company.
That’s the real story behind the negative coverage.