I am a consultant. How do I prove the AI research behind my recommendation was actually done?
Run the research as a recorded job in its own right: the questions asked as steps, the sources consulted with their resolution states saved, and the synthesis that produced the recommendation as the final steps. Review the receipt and actual deliverable before accepting a handover.
Consultants sell judgement, and judgement increasingly sits on research an assistant performed. The deliverable shows the conclusion; the client, or the client's own reviewer, reasonably asks what the conclusion rests on. "Extensive research was conducted" is the AI-era version of "trust me," and sophisticated clients have stopped accepting it.
The recorded research job answers with structure. The scoping step saves the questions the research had to answer. Source steps carry what was actually consulted, with the source check's state on each: resolved, not resolved, not checked. That state line matters commercially: a recommendation resting on sources that were consulted and resolved reads differently from one resting on sources nobody opened, and the record shows which without adjectives. The synthesis steps then connect evidence to conclusion, so a challenger can attack the reasoning at a specific joint instead of doubting the whole.
Deliver the research record as an annex to the recommendation, or as its own handover the client accepts. Consultants who do this discover a side benefit: their own quality control improves, because research that will be read is research that gets scoped properly in the first place.
Template: the research record annex
1. Scoping step: the questions the research must answer, as agreed with the client. 2. Source steps: each source consulted, its resolution state, and the claim it supports or contradicts. 3. Gaps step: questions the research could not settle, named before the recommendation relies on them. 4. Synthesis steps: how the evidence combines into each recommendation point. 5. Delivery: the annex link in the report, and the client's acceptance recorded on the handover. Rule: a recommendation point with no synthesis step behind it is labelled opinion, proudly, because consultants are allowed opinions when they are labelled.
Worked example
The source-state discipline in a live record: https://zambo.dev/handover/48479a33-d996-4419-a9c8-d1b9a55e62b9 carries sources in three states, {1 resolved, 1 not_resolved, 1 not_checked}, pinned exactly as checked. A research annex built to that standard lets a client weigh the recommendation by the state of its evidence, line by line.
Frequently asked questions
Will clients actually open the annex?
Some will not, and it still changes the engagement: the annex's existence signals the work survives inspection. The clients who do open it, procurement reviewers, boards, technical staff, are exactly the ones whose trust decides renewals.
The research used client-confidential sources. What then?
The record carries the sources' identities and states as agreed in the engagement; where a source cannot be named, the step records its class and the agreement under which it was consulted. Confidentiality governs disclosure, not whether the work is recorded.
Does this slow the research down?
The research runs through the connection as it happens; the record is the byproduct. The only added minutes are the scoping step, which is the discipline consultants already claim to practise.
By Brennan Zambo, founder of Zambo.
Last verified: 2026-10-17.
Try a handover
A handover gives your client one proof link with verifiable receipts to review, then Accept or Flag.
try it: say 'use zambo to hand this over to <client>' in your ai