What you can ask
The classes of question the assistant answers, the panel each one opens, and how to phrase a question so it lands where you want.
The assistant recognises a set of question classes and draws a different panel for most of them. You do not have to memorise them: the suggested pills and the rotating placeholder teach them a sip at a time, and the Shortcuts palette lists the whole set for your workspace. This is the same list, explained.
What you'll learn
- Which questions work, and the panel each one opens
- Why a question you were offered will never open an empty panel
- How to aim a question that landed somewhere unexpected
The questions and where they land
| Ask something like | Panel it opens |
|---|---|
| Which prompts am I losing, and who is taking them? | Prompts |
| How am I doing on ChatGPT? | Platforms |
| What is my citation rate on each engine? | Platforms |
| How do I compare to [a rival you track]? | Head to head |
| Where should I focus first, and what is it worth? | This week |
| Which pages are the engines citing me on? | Sources |
| Where do the answers come from? | Sources |
| Show me the exact sentences AI wrote about us. | Proof |
| What did you find overnight? | The last run |
| Is my website ready for AI answers? | Readiness |
| Hello | None, by design |
A greeting is answered and nothing opens, because opening a panel for "hi" would be theatre.
You will never be offered a question that cannot be answered
Every suggested question is gated on evidence your workspace already has. If no rival has been named in any answer, you are not offered a head-to-head question. If no answer has cited you yet, you are not offered Proof, because there would be no sentence to quote. If an engine returned nothing citable in the window, it is not offered as an example, because the only honest answer would be to report the exclusion.
That rule costs the product a few pills on a new account and buys something more valuable: a suggestion you accept always goes somewhere real.
How to aim a question
- Name your own rows with
@. Typing@offers your prompts, your competitors and the engines. The mention travels as an id, so the answer scopes exactly to the row you picked. Without it the assistant has to spot the name in your sentence, and a shortened rival name will be missed. - Say the engine's name to get a per-engine answer. "How am I doing on Perplexity?" focuses one surface. Ask without naming one and you get every surface, exclusions included.
- Ask for the sentences when you want the sentences. Questions about what AI said open Proof. Questions about where answers come from open Sources. They are different lenses on the same check.
- Ask about money and order when you want a plan. "Where should I focus first, and what is it worth?" opens This week, which ranks open work rather than listing everything.
If a question lands somewhere you did not expect, the phrasing carried vocabulary from a neighbouring
class. Re-ask it with the engine, prompt or rival named explicitly, or pick the question from ⌘K and
edit it.
Recall
Once a thread has history you can ask what you already discussed. That class is answered from the conversation store, so it is fast, deterministic and does not re-run the pipeline. On a brand-new thread there is nothing to recall, which is why it is never offered as a starting question.
Try this in Outercite
Press ⌘K in the assistant and read the Ask group. Every question listed there has been
verified against your own workspace, so anything in that list will open a panel with rows in it.
Related
How answers are grounded
The phases behind an answer, and the rule that no figure appears without evidence under it.
Panes and the ladder
What to do with the panel once it opens.
Shortcuts and navigation
The palette, the two menus, and how to get to a page instead of a panel.
Last updated on
How verification works
What happens between an AI engine's answer and a counted citation: the prefilter, the judge, the verifier that audits it, and the rules that decide a tie.
How answers are grounded
Every answer is built from counted rows in your own workspace, in phases you can see, with the evidence for each figure one click away.
