Ask Your Own Evidence a Question — Then Write the Answer Back

You have forty documents in a client's corpus. Three sales call transcripts, two win/loss interviews, a positioning deck from the previous agency, six months of campaign reports, the CEO's board narrative. Somewhere in there is the answer to "why do deals stall at security review?"

You know it's in there. You just can't get at it without reading all forty again.

That re-reading is the hidden tax on evidence-based practice. Every serious question you ask about a client requires reloading context you already loaded last week — and across four concurrent engagements, that reload happens multiple times a day. It is the single most expensive habit in the fractional model, roughly [add per-practice estimate] in unbillable hours a year.

The utility of this post: a repeatable way to interrogate your own evidence base — pick the slice, ask a question that can produce a wrong answer, get back a sourced answer plus structured insights, and write the good ones back into the corpus so you never have to re-derive them.

Why "ask the AI" and "ask your evidence" are not the same thing

Ask a general assistant to draft your client's positioning and it answers instantly, fluently, and from nowhere. Upload forty documents and you have the opposite problem: it now answers from everywhere. The stale deck from the prior agency gets the same weight as the win/loss interview you ran last Tuesday.

Signal Inquiry works differently in two ways that matter:

• You choose the slice. The question runs over an explicitly selected set of documents, not the whole pile. "Only the Q3 sales calls." "Only the two win/loss interviews and the ICP doc." What you exclude is as much a decision as what you ask.
• Every claim carries provenance. The answer comes back with structured insights, each tagged with the module it applies to (ICP, personas, value prop, messaging) and the source it came from. You can check the receipt.

That is the difference between an answer you can put in a board deck and an answer you have to go verify yourself.

Four question patterns that surface what a summary hides

Most people ask their corpus to summarize. Summaries confirm what you already believe. These four patterns are built to produce a wrong answer if your evidence is wrong — which is the only kind of question worth asking.

1. The contradiction hunt. "Across these five sales calls, where do prospects describe the buying trigger differently than our ICP doc does?" Select the calls and the ICP doc together. You are asking the system to find where your locked strategy and your fresh evidence disagree.

2. The gap probe. "Which of the four persona pains has no supporting quote in this document set?" This is the question that tells you what to go collect. One of our own users hit this in the product: "there's only three documents really that apply to the Persona, so you may want to have more before you start."

3. The syntropic reframe. Vague questions produce vague answers. Compare "Are events working?" with: "From July 1 to Aug 31, what pipeline and win rate are attributed to Event A vs Event B for ACV ≥ $25k, and what change would raise Event A's SQO-to-win conversion?" Time-bound, segmented, and ending in a decision.

4. The synthesis brief. Instead of a question, a construction: "Turn all the sales calls into one objection-handling doc, grouped by objection, with the strongest verbatim under each." This is what you use before a QBR — a new document assembled from things already in your signal.

The loop that makes the corpus compound

Here is the part that changes the economics. When an insight comes back and it's right, one click writes it back as new signal.

That answer is now a first-class source. The next question you ask reads it. The next asset you generate reads it. You derived it once, and it stays derived.

The thumbs feedback does the other half. Thumbs-down a stale deck or confirm a strong transcript, and you aren't tuning one output — you're re-weighting the evidence for every future generation, because everything reads through the same digest. One person's five-second correction improves everyone's next draft.

Run this in twenty minutes on your next engagement

1. Pick one live decision. Not "tell me about the client." Something like: which persona do we lead with in Q1 outbound?
2. Select the slice deliberately. The 4–6 documents that should contain the answer. Note what you excluded and why.
3. Ask one contradiction question and one gap question. Both, in that order.
4. Check three citations. Open the sources behind the three insights you'd actually act on. If a citation doesn't hold, thumbs it down.
5. Write back the survivors. Every insight that passes verification goes into the corpus as signal.
6. Re-run the same question next month. The delta is your evidence base maturing.

Six months in, your corpus isn't a folder anymore. It's an interrogable record of every conclusion you reached and the evidence you reached it on — which is exactly what you hand a nervous CEO in week eight when they ask how you know.

The retainer no longer rests on the deliverable. It rests on the reasoning behind it, and now the reasoning is on the record.