Before you start an AI transcript analysis
Tell it what the meeting was for, who was there, and any relevant background. Ask it to flag anything that remains unknown as it works.
Before you start, what questions or context do you need from me so that you never have to make assumptions about the transcript I just gave you?
A project update, a sales call, and a brainstorm can contain the same words while giving those words different jobs. A tentative idea in a brainstorm shouldn’t automatically become a commitment in the analysis. Establish what people were trying to do before asking what their behavior means.
Include the participants’ names and roles, the purpose of the conversation, and any earlier decision needed to understand it. Say whether the transcript is complete or an excerpt. You can ask for unknowns to remain visible, but the opening prompt alone can’t guarantee that every assumption will be caught.
What to check in the answer
Look for a short list of missing facts. If the answer supplies a relationship, motive, or meeting objective you never provided, ask where it came from.
When speaker labels are missing
Listen to the recording if you have it. Checking 3 passages is a starting point. Check the speaker on every passage you later use as evidence, too. If the text doesn’t establish who spoke, keep that passage unassigned.
The two speakers are [name and role] and [name and role]. Please try to separate their contributions using clues in the transcript. Keep the original words intact. Mark uncertain passages as “speaker unknown.” Give me 3 passages from different parts of the conversation, with your proposed speaker labels and the clues you used. Wait for me to check them before analyzing anyone's patterns.
An unlabeled transcript can still help you locate a discussion worth revisiting. Attributing a habit or decision to a particular person requires another check. A plausible division of speakers can produce an account that credits the wrong person throughout.
Replace the bracketed names and roles before copying. If more than two people took part, name everyone you can identify and adjust the opening sentence. Clues such as a direct address or a reference to someone’s role can support a proposed label; they don’t make every nearby line certain.
What to check in the answer
Keep proposed labels separate from the original transcript. Record the clue for each passage you use. Leave ambiguous overlap or a disputed speaker unresolved until the recording or a participant can clarify it.
When you want to check an explanation
Use this when an analysis sounds convincing but you can’t see how it follows from the conversation.
Show me the words that support that explanation. What else could explain this decision?
A sentence such as “you were protecting the team’s focus” describes an interpretation. Ask to see the exchange that supports it. The same decision could have come from a deadline, a missing resource, or a restriction someone stated earlier.
Ask for alternatives that fit this particular exchange. A list of generic possibilities won’t help much unless the analysis explains which words support each one and which information would help distinguish them.
What to check in the answer
A useful answer keeps the quoted passage, the interpretation, and the alternatives separate. You should be able to disagree with the interpretation while preserving the original passage.
When you want to check who did what
Use this before turning a meeting summary into evidence about someone’s judgment or contribution.
Go back through your analysis and check every claim against the original transcript. For each claim, show the exact passage and speaker. Distinguish who proposed an idea from who agreed with it, questioned it, or developed it further. Flag uncertain attribution and revise anything you've credited to the wrong person.
Conversations develop through several people. Someone can propose an idea, another person can test it, and a third can turn it into a workable plan. A summary that attributes the finished plan to the person who repeated it last can erase that sequence.
Review one claim at a time. Keep the wording of the claim beside its supporting exchange and label the contribution the passage actually shows. When several speakers contributed, retain that distinction instead of forcing the result into one person’s column.
What to check in the answer
The revised analysis should identify the action supported by the quote. Agreement should stay agreement. A question should stay a question unless the exchange establishes something more.
When you want to find the assumptions
Use this when the explanation contains details that feel familiar but aren’t visible in the source.
Which parts of your explanation are stated in the transcript, and which parts have you inferred? Point out any missing context you filled in. Ask me about it before building on those assumptions, and leave it unknown if we can't establish it.
An analysis may connect two recorded facts by supplying an unrecorded reason. For example, a person postpones a launch and later mentions a staffing gap. That gives you a question to investigate. It doesn’t establish that staffing caused the postponement.
Ask the model to mark each added connection as an inference and name the missing information. If you provide that information from memory, label it as your clarification so it remains distinguishable from the transcript.
What to check in the answer
Watch for an unknown quietly becoming a fact in the next answer. Carry unresolved assumptions forward explicitly when you ask a follow-up question.
When a quote seems to be missing context
Use this when a short quote seems unusually decisive, complimentary, dismissive, or certain.
Show me the exchange before and after that quote. Does the surrounding conversation change your interpretation? Use any source locations already available so I can check it myself.
The next line can qualify a statement. The previous question can explain why the answer was short. Bring back enough of the exchange to understand what the speaker was responding to and whether they later revised it.
Use timestamps, paragraph numbers, or speaker turns if the source already contains them. Don’t ask the model to invent a timestamp. If your copy has no locations, keep a distinctive phrase you can search for in the original.
What to check in the answer
The answer should preserve the exchange’s order. If the source is an excerpt and the surrounding material is unavailable, the interpretation should keep that limitation.
When something looks like a pattern
Use this after you’ve checked the speaker and context of an example you want to investigate further.
Look for other examples of this in the material I've supplied. Also look for occasions where I did something different. What conditions seem to change my response? Say when you can't find enough evidence to tell.
One exchange can show a decision on one occasion. To investigate a recurring response, compare it with other supplied occasions and include the ones that differ. Repeated wording alone may tell you less than what someone noticed and chose in each situation.
Compare the circumstances as well as the action. A response to a new client may differ from a response to a colleague with years of shared background. Keep each example tied to its source, and ask whether the material is broad enough to support the claim being made.
What to check in the answer
An answer that reports too little evidence can be useful. Save the candidate pattern with the missing examples you would need, rather than treating repetition as proof of a permanent trait.
When you want to check the claimed result
Use this whenever the analysis says that a recommendation worked, solved a problem, or changed a result.
Does the source show what happened afterward, or have you inferred an outcome from the recommendation? Quote any recorded result and leave the outcome unknown where it's missing.
A meeting often records a plan before anybody has tried it. Agreement to take an action establishes an agreement. The outcome may be in a later meeting, a follow-up message, or a record you haven’t supplied.
Keep the recommendation, the decision to act, and any recorded result distinct. If somebody reports a result later, identify it as a report and retain the passage. A positive result by itself may still leave open which action caused it.
What to check in the answer
Where the recording ends with “let’s try it,” the outcome should remain unknown. Add later evidence when you have it, with its own date or source location.
When you’re unsure where to begin
Use this to choose a manageable starting point from a long meeting.
Read this team meeting or brainstorm and suggest one moment where I made a choice, asked a question that changed the discussion, or explained something to another person. Show the exact exchange and explain why it's worth a closer look. If there's too little context, ask me for what's missing.
Start with one exchange you can check against the original. You don’t need an account of your entire working style before looking closely at a single question or decision.
Once you’ve checked the passage, ask what you noticed before speaking and what options you remember considering. If you’re exploring expertise, the Six Expert Tells guide can help you notice words such as “just,” “obviously,” or “it depends.” Keep each interpretation connected to the occasion.
What to check in the answer
A useful starting point gives you the exact exchange and a reason to examine it. You can choose a different passage if the suggested one lacks context or doesn’t match what you remember.
A short example of checking attribution
This exchange is invented to demonstrate the check.
Alex: “Could we try the smaller pilot first?”
Sam: “Yes, provided we include the support team.”
Alex: “Agreed. Let’s invite them before we choose a date.”
If the analysis says Sam proposed the pilot, the first line lets you correct that attribution. Alex proposed the smaller pilot. Sam agreed with a condition. Alex then agreed to involve the support team before choosing a date.
The exchange doesn’t show whether an invitation was sent, whether the pilot ran, or whether it succeeded. Those outcomes remain open. Keeping that boundary gives you a more precise account of what each person contributed.
Keep an evidence note beside each claim
A small note makes later review easier. Keep the exact passage, the speaker and confidence in that label, any source location, and the claim you’re considering. Add the context that supports or limits the interpretation. Record what remains unknown.
If a later answer changes the claim, revisit the note. This gives you a way to see whether the explanation still rests on the same exchange or has picked up assumptions along the way.
You can do this in an ordinary document. Keep your annotations separate from the transcript so the source wording stays intact. You don’t need to send a transcript to this page; its buttons only copy the questions.
Common questions about checking a transcript
Can I use these with a summary instead of a full transcript?
You can use them to identify what needs checking, but a summary may omit the exchange needed to resolve the question. Ask for the original passage. If it isn’t available, keep the claim limited to what the summary actually records and mark the missing source.
How many examples establish a pattern?
These questions don’t supply a universal threshold. Examine how similar the situations are, whether the behavior is attributed correctly, and whether contrary examples change the claim. Repetition in a narrow set of meetings may support a narrower observation than a claim about how someone works everywhere.
Should AI assign speakers from writing style?
Treat any proposed speaker label as something to verify. A role reference, a direct address, or the recording may help you check it. Where the available material doesn’t establish a speaker, retain “speaker unknown” and keep that passage out of claims about a named person’s patterns.
What if the AI gives a quote I can’t find?
Search the original transcript for the quoted words and check the surrounding passage. Ask for an existing source location. If you still can’t locate it, leave that quotation out of the evidence and ask the analysis to revise any claim that depended on it.
From checking the analysis to examining your judgment
Once you’ve checked an exchange, you can look more closely at the decision it contains. The Six Expert Tells explores phrases such as “just,” “obviously,” and “it depends” that can mark a place to investigate. A tell helps you locate a question; the transcript and your recollection help you examine it.
Keep the original transcript available. These questions help you inspect the analysis; you still need to check the passages it returns.