ZAMBO.

Direct answer

Two AIs, two answers. Only one can be the result of record, and the record should say which.

You ran the same job twice, once in each of two AIs, as a check, a second opinion, or a race. The results differ, in a figure, a recommendation, a whole section. The client has seen both, or will, and now asks the only question that matters: which one do we stand behind. If the answer lives in your head or in a chat message, it will be argued later, when a third version surfaces in someone's folder and nobody can prove which result the job was actually delivered on. Here is how to settle this pair honestly, how to mark the result that was not adopted so it stops circulating as a rival, and how to designate the standing result on a record the client can check, so the question stays answered.

See the instrument first: the live demo runs one real call and hands you its verifiable receipt, no signup.

Updated November 5, 2026. First published November 5, 2026. The two delivery records quoted below are real and public, with their client decision states read back from the live service on November 5, 2026.

Step 1. Expect two results to differ, and find where they actually differ

Two different AIs given the same brief will differ, in wording always and in substance often, because each works from its own reading of the brief and its own habits. Difference alone indicts neither result. Start by locating the real differences, not the surface ones: line the two results up section by section and mark each point as wording only, different method with the same conclusion, or different conclusion. Most pairs collapse at this stage into one substantive disagreement and a lot of cosmetic distance. That one disagreement is the decision the client is actually being asked to make, and naming it precisely, this figure, this recommendation, this section, is what turns an uneasy feeling about two versions into a question that can be settled. Do not skip this step and jump to picking a winner; a client shown the exact point of difference trusts the pick that follows.

Step 2. Judge both results against the brief and the sources, not against each other

With the difference pinned, test each result the same way. Against the brief: which one answers what was actually asked, covers every required part, and respects the limits the client set. Against the sources: for each claim that carries weight, can it be traced to material that was actually opened and read, and does the result say, anywhere, what was not checked. Against its own record, if one exists: a result whose steps can be reopened, inputs, tool calls, saved outputs, can be interrogated in a way a pasted answer cannot, and that interrogability is a legitimate reason to prefer it when the texts are otherwise close. Write the comparison down in a few lines per test. The point is not to crown the better AI in general; it is to choose, for this job, the result you can defend line by line to this client, and to have the reasons written before anyone asks for them.

Step 3. Read the two states a chosen result can sit in

A choice between results only holds if it is recorded as a decision, not announced in a message that scrolls away. The record format is AER-1, published as an IETF Internet-Draft (draft-zambo-aer1-26), and each step carries a verifiable receipt: a receipt proves the saved result was not changed since it was saved. The two public records below show the states a delivery can carry:

Accepted: the decision is on the record

Pending: checked, and still waiting on a decision

The accepted record shows a decision with a name and a date on it: status accepted, decided_by client, decided_at 2026-10-16T14:31:51.146Z, release_hint ok_to_release, over 4 steps, 4 checked. Translated to this problem, that is the shape of a standing result: one result, checked, accepted by the client on a date, with the acceptance attached to the exact steps that produced it. The pending record, 5 steps, 5 checked, release_hint hold, shows the exposed state your two results are in right now: both may be perfectly checkable as texts, but no decision has been recorded on either, so neither stands behind anything yet, and any later version can claim the job. The work of steps 4 and 5 is moving one result, and only one, from the second state to the first.

Step 4. Designate the standing result, in writing, with its reasons

Send the client one short designation message, and keep it with the job. It names the standing result exactly, by its record link or its delivery date and title, never just as the second one or the better one. It states the point of difference in one sentence and the reason the standing result was chosen on that point, traced to the brief or the sources, in the plain terms of step 2. It states what happens to the other result, covered in step 5. And it asks the client to accept the standing result on its record, so the designation carries their decision as well as yours. Suppliers worry this message invites an argument; in practice it ends one, because a client who can see the difference, the test, and the reason rarely reopens a choice they have accepted in that form. What reopens choices is silence, which leaves every version in the folder equally official.

Step 5. Mark the other result as a draft that was not adopted, and keep it

The losing result causes trouble only when it circulates unlabelled. Label it where it lives: not adopted, superseded by the standing result named in the designation, kept for the record. Do not delete it. Deleting it destroys the evidence that a comparison happened, makes the standing result look like the only thing ever produced, and leaves you unable to answer a later question about why the other path was not taken. Kept and labelled, the draft becomes an asset: it shows the client the work was checked against an alternative, and it shows any later reader exactly which text was never delivered as the answer. The pair, standing result accepted on its record and draft labelled as not adopted, is a complete and honest account of a two AI job. Anything less, one file with a hopeful filename, is how version disputes are born.

Step 6. Next time, run the comparison inside one job record

The clean version of this workflow keeps both runs where the decision will live. Run the job as steps on one record; where a second AI is used as a check, its run is a step or a linked draft on the same job, not a separate thread in a separate account where it acquires a life of its own. The chosen result is the step that gets wrapped and delivered; the client accepts that delivery on the record; the alternative stays attached, labelled. Run this way, the question which one do we stand behind has a one link answer from the day the job closes, and the comparison itself, the difference found, the test applied, the reason chosen, is part of what the client can see. That visibility is worth having commercially as well as defensively: clients keep suppliers whose choices they can inspect.

What a record does not settle

A verifiable receipt proves the saved result was not changed since it was saved; it does not prove the standing result is the correct one, that the other AI was wrong in general, or that a client who accepted a result cannot ask for a change. The comparison against the brief and the sources, the written designation, the client's acceptance, and the labelled draft are what this answer stands on. This page is general information, not legal advice.

Frequently asked questions

I ran the same job in two different AIs and got two different results. How do I show my client which one is the result we stand behind?

Show them a designation, not a preference. First line the two results up and find the substantive point where they differ, separating wording differences from different conclusions. Test both against the brief and against the sources each one used, including what each says was not checked, and write the comparison down. Then send one designation message that names the standing result exactly, by its record link or delivery date and title, states the difference and the reason for the choice in one or two sentences, and labels the other result as a draft that was not adopted, kept for the record. Ask the client to accept the standing result on its record, so the choice carries their decision and its date. From then on the answer to which one stands is a link and a date, not your recollection of a conversation.

Does getting two different results mean one of the AIs made a mistake?

Not necessarily, and treating difference as error is the fastest way to pick the wrong winner. Two AIs read the same brief with different emphases, choose different methods, and phrase differently as a matter of course; many differences are cosmetic, and some substantive differences are two defensible answers to an underspecified brief. A mistake is a specific thing: a figure that does not trace to a source, a requirement from the brief that was missed, a claim presented as checked when it was not. Hunt for those, in both results, with the same test. If one result carries a real mistake on the point that matters, the choice makes itself. If neither does, say so, and choose on fit with the brief and on which result you can defend line by line, recording that the choice was a judgement between two sound answers.

The client prefers the result I do not recommend. What should I do?

Record their choice as the decision it is, with your recommendation kept beside it in writing. The client owns the call on their own job; your professional duty is to make sure the call is informed and documented, not to win it. Put the comparison in front of them, the exact point of difference, what each result rests on, what was not checked on either side, and your recommendation with its reason. If they choose the other result, designate it as the standing result at their direction, note your recommendation in the job record in neutral words, and deliver on their choice properly. If the point of difference is one where you believe their choice creates a real risk for them, say that once, plainly, in the same record. A supplier who documents a disagreement this way is protected either way the result turns out, and usually gains trust rather than losing it.

Should I delete the result we did not choose, to avoid confusion?

No. Label it, keep it, and let the label do the work. Deleting the unchosen result removes the evidence that a comparison was ever made, leaves the standing result with no visible alternative it was tested against, and guarantees that if a copy survives anywhere, in the client's folder, a forwarded thread, a backup, it resurfaces later as an unexplained rival version with no label on it at all. A kept draft marked not adopted, superseded by the named standing result on its date, can never be mistaken for the delivery. Confusion comes from unlabelled copies, not from kept ones, and the cheapest version control in a professional practice is a clear label attached to a record that cannot quietly change.

How do I designate one result as the one of record, so it stays designated?

Put the designation on a record instead of in a message. When the work runs through Zambo, the chosen result is delivered as a wrapped job whose every step carries a verifiable receipt under AER-1, published as IETF Internet-Draft draft-zambo-aer1-26, and the client accepts that delivery on the record, with the decider and the date kept, while the other result stays attached and labelled as not adopted. New users get 20 calls and 1 wrap-up free every day, and picking up work in another AI is always free. For $0.99, deliver the next two version job that way and make the standing result a matter of record, not memory: get the Day Pass. A Day Pass is unlimited for 24 hours, no subscription. Ongoing use is the Zambo Pass at $29 a month or $290 a year, and $19 a month for $ZAMBO holders, which adds private handover links, your branding, and the approval record as PDF and JSON files.

Related