what edgeful AI is built for
edgeful AI draws on edgeful's full report library. that means you can ask it to research any report, across any ticker, across any session — and it'll pull the data and give you a structured answer.
the 3 things it does best:
1. report deep dives
going deep on a single report for a single ticker. instead of scanning the report page yourself, you ask the AI to walk you through what the data shows — the key numbers, the recent trend, what conditions look like right now.
this is useful when you want more than a quick glance at the screener. you're not just checking a colour — you're understanding what's behind it.
2. multi-report confluence
asking the AI to look across multiple reports for the same ticker and tell you what they're collectively showing. this is where edgeful AI really earns its place — pulling together data from reports that you'd normally have to check one by one and synthesising a directional read.
if you're building a bias and want to know whether the IB, gap, and previous day's range reports are all pointing the same way for NQ today, this is the question to ask.
3. historical pattern research
digging into how price has behaved historically under specific conditions. not "what will happen" — but "when these conditions have shown up before, what happened?" that's where the edge is, and it's what edgeful's data is built on.
example prompts by use case
copy these and adapt the ticker, session, and lookback to your setup. these map to the 3 question types above — start with whichever fits what you're trying to learn.
deep dive on a single report
walk me through the IB by rejection report for NQ in the NY session over the last 6 months. give me the key numbers, the recent trend, and what conditions look like right now.
what does the gap fill report show for ES, NY session, last 3 months? fill rate, what's typical when gaps don't fill, and what the setup is today.
break down outside day frequency for GC across all weekdays. which weekdays produce the most outside days, and how does price typically resolve?
multi-report confluence (building a bias)
I'm building a bias for NQ today. look at the IB, gap, and previous day's range reports for the NY session. what direction are they collectively pointing?
for ES in the NY session, cross-check the engulfing candle report and the IB report. when these have agreed historically, how often did the setup play out?
compare the previous day's range, gap fill, and ORB reports for CL in the NY session. what's the most common day-of-the-week pattern when all 3 line up?
historical pattern research
look at every NQ double break day in the NY session over the last 6 months. find commonalities — I want to be able to predict double break days before they happen.
find every day the gap didn't fill on ES, NY session. what features did those days share — opening candle direction, prior day's close, IB size?
show me every IB single break day on GC over the last year. find commonalities and rank by frequency.
when you're stuck
what questions can I ask you about these reports? — use this after attaching reports. returns a menu of productive angles.
summarise what this data table is telling me about [ticker] in 3 bullets — fastest way to get a read on a fresh analysis.
how to frame a good question
the single biggest factor in the quality of the AI's response is how specific your question is.
a vague question produces a vague answer. a specific question — one that includes the ticker, the session, and the report you care about — produces a focused, data-driven response you can actually use.
include these 3 things whenever you can:
ticker — which instrument are you asking about? ES, NQ, GC, AAPL, etc.
session — NY, London, or Asian? the data is session-specific, so this matters
report or condition — which report or market condition is the focus of your question?
the more of these you include, the more precise the response. asking about "the IB report for NQ in the NY session" will produce a materially better answer than asking about "the IB report" in isolation.
date range is also useful when you have a specific lookback in mind — for example, asking about how a pattern has behaved over the last 6 months vs the last 3 months can surface different reads.
match the column labels in your data table
when you've run an Analyze Reports flow and have a data table attached to the chat, the single biggest precision boost is using the AI's own vocabulary back at it. the column names in the data table are exactly what the AI is looking for.
a concrete example. both of these ask the same thing about the IB:
break down the IB break type by weekday — works, but "break type" isn't a column name. the AI has to translate it into the actual columns (breakout / breakdown / double break / no break) before it can answer
how often does price break out, break down, double break, and no break by weekday — uses the actual labels in the table. zero ambiguity
both produce usable answers. the second is more reliable, especially on edge cases. keep the data table panel open while you type — being able to see the column names is the fastest way to ask precise questions.
use "find commonalities" to surface predictive setups
the most powerful prompt pattern in the Analyze Reports flow is asking the AI to find commonalities across a set of days that share an outcome:
look at all double break days and find commonalities. I want to be able to predict double break days before they happen.
tell me every day where the gaps didn't fill and find commonalities
show me every day the IB single broke and find commonalities
the AI scans every row in the data table that meets the condition, then looks across the rest of the columns (size, weekday, high/low formed first, opening candle direction, etc.) to surface which features show up most often on those days. the answer is usually a ranked predictive checklist — a short list of features that historically appear when the condition is true.
it's the closest thing edgeful AI offers to letting you act like a quant without being one. you give it the outcome you care about, and it finds the conditions that lead to it.
if you're stuck, ask the AI what to ask
once you've analyzed a set of reports and the data table is attached, a great opening prompt is simply what questions can I ask you about these reports?. the AI knows what's in your table and will suggest the most productive angles — use the list as a menu for the rest of the conversation.
it's especially useful when you've stacked 2 or 3 reports and aren't sure which cross-report angles are worth digging into.
let the follow-up suggestions guide the next step
most responses end with a suggested follow-up question — something like "want me to break down those Tuesday double breaks by how they resolve?" or "want me to check the opening candle direction on those 12 days?". the AI is reading what's still unanswered in the data and pointing at the next logical question.
reply with yes, run that or copy the suggested question verbatim. either works. follow-ups are how multi-step research happens — one analysis seeds the next, and you end up two or three layers deeper than you would have asked on your own.
use the TL;DR if the response is long
when an answer runs long, edgeful AI appends a TL;DR section at the end with the short version. if you're skimming, jump straight to it — the full breakdown is there above it when you want to dig in.
what it can't do
edgeful AI is a research tool — it doesn't provide real-time data, predict future price movement, or access your personal account or trade history. for the full breakdown, see what edgeful AI can and can't answer.
the right mindset for using it
think of edgeful AI as a research analyst who's read every report on the platform and can pull any data point on demand — but who works from historical records, not a live screen.
the questions that get the most out of it are research questions. "what does the data show for X ticker under Y conditions" is the core pattern. the more you lean into that framing, the more useful every conversation becomes.
if you're not sure where to start, the suggested prompts on the AI's home screen are a good first step — they're designed to show you the kinds of questions the AI is optimised for.