This lesson is for mentors, coaches and coordinators reviewing session records. You will turn a short note into an evidence table and identify a clarification. The method applies to different local note formats; a shared structure for selected fields does not require every mentor to use an identical script.

1. Record the participant, session, author and relevant dates.
2. Separate statements, observations and interpretations.
3. Extract only what the note supports.
4. Record actions with their agreement and completion status.
5. Keep missing information explicit.
6. Review the record and its permitted audience.

What counts as evidence in a session note?

A direct quote records the contributor’s words when captured accurately. An attributed paraphrase records their account in the writer’s words. An observation describes what the writer witnessed. An interpretation explains what the writer thinks it means. These can all have a place, but they should not be presented as interchangeable.

A quote is not the only legitimate form of a note. Do not put quotation marks around a paraphrase or require a participant to provide a polished quotation. Attribution and honest source description matter more than making every field look certain.

Which fields should be shared?

Choose the fields needed for continuity: participant identifier, session date, author, relevant goal, next action and review date may be sufficient. Let local programs retain additional prompts that serve their work. A school mentor and a professional coach may use different notes while sharing a compatible action-status field.

Define the common fields in a data dictionary. For example, distinguish an action proposed by a mentor, an action agreed with the participant and an action later confirmed completed. A single “action” column can blur all three. Keep the note’s original wording beside any structured extraction.

How do you extract the note without adding meaning?

Work through one field at a time and retain the supporting passage. If the source says a draft was discussed, do not report that the draft was submitted. If a challenge is mentioned without a cause, record the challenge and leave the cause unconfirmed.

In a fictional note, “Sam brought an outline. We discussed a difficult opening. Sam will try another version by Friday” supports an outline reviewed and an agreed revision. It does not establish a completed application, lack of confidence or the reason the opening was difficult.

For AI-assisted extraction, specify the fields and ask for source passage, statement type and uncertainty. Require “not recorded” where the source provides no answer. A reviewer should be able to reject a suggested field value without changing the original note.

What should happen when a field is missing?

Distinguish “not recorded,” “not discussed,” “participant preferred not to answer,” and “no issue reported” where those states serve the work. Do not force a writer to invent a barrier just because a form requires text. An explicit status can be more useful than a mandatory narrative box.

If someone remembers additional information, record a dated addendum with its author and basis. Do not silently rewrite the original note. A model reading only the note cannot know that a blocker was raised in the room but omitted from the record.

How do actions stay connected over time?

Give each relevant action an owner and a review point. At the next session, check whether it remains relevant, was completed, changed or was deferred. Record the participant’s account and any supporting observation separately.

Changing a goal is not necessarily failure. Keep the earlier goal and the reason for the change where appropriate. The record should help understand the participant’s journey rather than force every session into a steadily improving line.

Where does Sopact help?

You can apply this method manually to a small set of notes. Sopact’s connected-record approach can retain session history and provide the shared definitions for structured analysis across notes, surveys and uploaded evidence. This makes the note available for review without requiring a person to reopen every file.

Test the extraction on a varied sample, including missing fields and ambiguous statements. Verify that original text, dates and reviewer corrections remain accessible to authorized staff. AI does not establish an emotion simply by labeling it, and a continuing identifier does not justify broader access to personal details.

Practice: correct an overconfident summary

Use the fictional Sam note above. Write an extraction table with topic, observed work, agreed action and unknowns. Then correct this summary: “Sam gained confidence and completed the application.” Explain which parts have no support and what the next session could usefully establish.

Frequently asked questions

Must every field have a direct quote?

No. An accurately attributed paraphrase or observation can be appropriate. Preserve how the information was obtained and do not represent a paraphrase as verbatim speech. Keep interpretation visibly separate from the participant’s account.

Should a barrier field be mandatory?

A status field may help ensure the topic is considered, but it should allow legitimate states such as not discussed or none reported. Requiring a narrative in every session can encourage invented or repetitive content.

Can AI detect an omitted part of a conversation?

Not from a note that does not contain it. It can identify a field missing from the record, but it cannot establish what was said outside that record. Any later clarification needs a source and date.

How should we correct an inaccurate note?

Use the organization’s correction process and preserve the original, the correction, its author and date as appropriate. Revisit any summary that used the inaccurate statement. Do not treat a new summary as correcting the underlying source automatically.