Path 07 · 5 min read
Using the AI chat without wasting it
The pop-up chat — a bottom sheet on mobile, a right drawer on desktop — is the same three-model pipeline that produces the verdict, exposed as a free-form text box. It reads as a chatbot; it isn't one. It has a specific shape, and the shape is what makes the answers worth reading.
Four entries cover what the chat sees, how spatial context works, how to pin a thread, and why the chat is intentionally blind to your portfolio.
Start with what the surface actually is — and what's already loaded into context before you type a word.
01AI chat sheet
The pop-up chat — a bottom sheet on mobile, a right drawer on desktop — that you can ask anything about the stock you're looking at.
The chat is the same model pipeline as the verdict, exposed as an ask-anything text box. Replies stream a sentence at a time and arrive with the same citations a verdict carries — every claim points back to a filing, a print, or a data point.
On any stock profile, the chat already knows which ticker you're on and has pre-loaded that ticker's full data bundle, recent transcripts, feed history, and latest verdict into context. You can ask follow-up questions without re-naming the company — "why is the bear case thin?", "compare margins to AMD", "what changed in the last 10-Q?" — and the model has the inputs to answer specifically rather than generically.
The chat is intentionally blind to your portfolio. It knows what stock you're viewing, not what you own — that's the publisher-posture line, and the chat composer enforces it.
The reason a terse follow-up like "why is the bear case thin?" actually works.
02Spatial context
The fact that the AI chat tracks which page you're on. Open NVDA, ask "what's the bear case?", and it knows you mean NVIDIA.
Every stock profile registers its ticker with the chat so the model always knows what page you're reading. Walk from NVDA to AMD and the chat's context follows you — the next question lands against AMD's bundle, not NVDA's.
This is what lets the chat handle terse questions like "why now?" or "how does this compare to the last quarter?" without you having to restate the ticker every time. It is also why the chat composer never sees your portfolio: spatial context is about the page, not about you.
When you want a thread on NVDA to keep going while you walk around the rest of the app.
03Pin a ticker
Lock the chat to a specific company so the context stops following the page. Useful when you want to keep digging into NVDA while you browse elsewhere.
The chat's default behaviour is to follow the stock you're currently viewing. Pinning overrides that: the chat will keep answering against the pinned ticker even after you navigate to other pages or the feed.
Unpinning returns the chat to following the current page. The pin is the one-tap escape hatch for the times you want to keep a research thread open while you compare against other tickers.
Every claim the chat makes points back to something. If it can't cite, it doesn't ship — same rule as the verdict.
04Citation
A traceable claim → source mapping. Every verdict must point at the specific filings, prints, or data points behind it.
Verdicts are required to cite. Each claim the model makes is tied to a source (Edgar, FRED, OpenInsider, Quiver, transcript), a one-line claim, and the value pulled.
If a verdict can't cite, the pipeline doesn't ship it. This is what separates the verdict from a generic chatbot summary.
And the line the chat won't cross. Knowing why "should I sell my NVDA?" gets the same answer as "what's the read on NVDA?" is the whole product in one sentence.
05Publisher posture
The publisher publishes opinion on securities, not advice tailored to you. The same verdict reaches every reader; no one's situation, holdings, or goals are inputs.
This product runs under the publisher exclusion in §202(a)(11)(D) of the Investment Advisers Act — the same posture courts upheld for Seeking Alpha in Lingley v. Seeking Alpha. The exclusion fits when three things are true: the content is impersonal, the same content goes to everyone who reads it, and the publisher doesn't render personalised advice to any subscriber.
Three rules drop out of that for the surfaces a reader sees. Verdicts are opinions on the security, never instructions to a person — no "you should buy." The AI chat is blind to your portfolio and your situation. Marketing language never claims to "predict" markets or describe itself as "AI-driven" investment advice.
The non-AI compliance guard at the end of every verdict and every chat reply enforces those rules in code, on every output, live or mock. Crossing them isn't a code change — it's a multi-month regulatory move into RIA registration.
Ask the chat about the security, not about yourself. The model has the bundle, the transcripts, the feed history, and the verdict — it does not have your cost basis, your tax situation, or your goals. The questions that get the best answers are the ones that respect that split.
Quick check
Did it stick?
3 questions · pass at 3/3.
1. When you open a stock profile, what does the AI chat already know?
2. If you ask the chat "should I sell my NVDA?", what happens?
3. Pinning the chat to a ticker means…