Gioia Methodology
Toolkit · Qualitative
A reusable AI toolkit for Gioia-style qualitative analysis.
What has been built
An agent skill, templates and quotation checker support concept development (Gioia et al. 2013). A synthetic memo preserves rival interpretations but also contains an unsupported negative-case inference that exact quotation matching cannot detect.
What comes next
Evaluate concept development with empirical materials and researcher judgments.
Open the toolkit, worked example and checks on GitHub
What it does
The toolkit helps a researcher move from supplied excerpts towards concepts and a data structure: an AI agent proposes, the researcher decides.
Throughout, what participants said is kept apart from what the analyst makes of it.
Use the skill and templates to develop a memo, then check its source expressions with the quotation checker. The example below shows where a correct match can still leave an interpretation unsupported.
The worked example, step by step
The folder’s example is a six-line interview passage written for it; no participant said any of it. An AI assistant, following the toolkit’s instructions, wrote an analytic memo on it; no researcher has yet decided on any of its proposals.
In a Gioia data structure, first-order terms stay close to what informants say, while second-order themes and aggregate dimensions are the researcher’s (Gioia et al. 2013; Gehman et al. 2018). The memo keeps the passage’s words apart from the assistant’s proposals.
Choose a part of the drawing, or step through it.
Keep the words with their place
The memo cites three expressions by their place in version episode-v1: “I waited for a reply” from Speaker A (line 2, characters 56 to 76) and from Speaker B (line 3, characters 134 to 154), and “The deadline mattered more than the sign-off.” (line 5, characters 260 to 305).
The checker finds the shared words twice and chooses neither; which speaker said which is the analyst’s reading of the surrounding lines.
Line 6, a third person’s view reported by Speaker B, is set aside with its reason recorded.
Propose a concept
The assistant proposes “Proceeding without sign-off” for Speaker A, who sent the draft when no reply came and gives the deadline as the reason.
The label is the assistant’s, not a speaker’s phrase. It is recorded with its reason and its origin, and its state is proposed.
Weigh a rival reading
As a rival, the assistant proposes “Using the wait”: Speaker B describes the same wait, spent checking the figures.
A reports sending after the wait; B reports checking figures during it. B’s eventual sending decision is unknown, so the shared words alone cannot establish a contrast in outcomes.
Record a negative case, without support
The memo then counts Speaker B as a negative case, because the same opening words are, in its words, “followed by not sending”.
The passage does not say that: Speaker B checked the figures while waiting, and whether B later sent the draft is not reported. The memo treats an unknown action as no action.
Keep the cause a hypothesis
The memo proposes that deadline pressure led Speaker A to send the draft, and keeps the link a hypothesis with its warrant missing, beside two rival explanations: Speaker A would have sent it anyway, or no sign-off was expected.
It also lists Speaker B as evidence against the link, which rests on the same unsupported reading.
With two speakers, no higher-order grouping is proposed.
Leave the decision to the researcher
Every proposal is still proposed: the fields for who adopted it, when and why are empty.
A researcher may adopt, revise, reject or defer each proposal; the agent cannot adopt on the researcher’s behalf.
The folder’s example is a six-line interview passage written for it; no participant said any of it. An AI assistant, following the toolkit’s instructions, wrote an analytic memo on it; no researcher has yet decided on any of its proposals.
In a Gioia data structure, first-order terms stay close to what informants say, while second-order themes and aggregate dimensions are the researcher’s (Gioia et al. 2013; Gehman et al. 2018). The memo keeps the passage’s words apart from the assistant’s proposals.
Keep the words with their place
The memo cites three expressions by their place in version episode-v1: “I waited for a reply” from Speaker A (line 2, characters 56 to 76) and from Speaker B (line 3, characters 134 to 154), and “The deadline mattered more than the sign-off.” (line 5, characters 260 to 305).
The checker finds the shared words twice and chooses neither; which speaker said which is the analyst’s reading of the surrounding lines.
Line 6, a third person’s view reported by Speaker B, is set aside with its reason recorded.
The assistant’s proposals
Propose a concept
The assistant proposes “Proceeding without sign-off” for Speaker A, who sent the draft when no reply came and gives the deadline as the reason.
The label is the assistant’s, not a speaker’s phrase. It is recorded with its reason and its origin, and its state is proposed.
Weigh a rival reading
As a rival, the assistant proposes “Using the wait”: Speaker B describes the same wait, spent checking the figures.
A reports sending after the wait; B reports checking figures during it. B’s eventual sending decision is unknown, so the shared words alone cannot establish a contrast in outcomes.
Record a negative case, without support
The memo then counts Speaker B as a negative case, because the same opening words are, in its words, “followed by not sending”.
The passage does not say that: Speaker B checked the figures while waiting, and whether B later sent the draft is not reported. The memo treats an unknown action as no action.
Keep the cause a hypothesis
The memo proposes that deadline pressure led Speaker A to send the draft, and keeps the link a hypothesis with its warrant missing, beside two rival explanations: Speaker A would have sent it anyway, or no sign-off was expected.
It also lists Speaker B as evidence against the link, which rests on the same unsupported reading.
With two speakers, no higher-order grouping is proposed.
Leave the decision to the researcher
Every proposal is still proposed: the fields for who adopted it, when and why are empty.
A researcher may adopt, revise, reject or defer each proposal; the agent cannot adopt on the researcher’s behalf.
Where the memo goes wrong
The memo contains an unsupported negative-case inference: it treats Speaker B as not sending the draft, although the passage only reports checking figures. B’s eventual decision is unknown.
Every cited expression matches the passage. The checker locates words; it cannot judge the inference drawn from them. The original memo is retained so you can compare the two.
Try the quotation checker
The checker confirms that the text matches its recorded fingerprint, then reports every exact occurrence of a quotation. It leaves interpretation to the researcher.
Here the same rule runs in the browser on the example’s passage. Try a quotation, change the text, or withdraw a declaration.
Declared text
episode-v1 · self-authored synthetic text · version v1
Interviewer: What happened after the review? Speaker A: I waited for a reply. Nobody answered, so I sent the draft anyway. Speaker B: I waited for a reply too, but I used the time to check the figures. Interviewer: Did anyone object afterwards? Speaker A: No. The deadline mattered more than the sign-off. Speaker B: My lead said the check was the right call.
- Recorded fingerprint
- 3491496b4e08aee2eb6bb40cebfc039dd7a79dd8916fc548cf2ea14753d3ea76
- Fingerprint now
- 3491496b4e08aee2eb6bb40cebfc039dd7a79dd8916fc548cf2ea14753d3ea76 (same as recorded)
Recorded basis: Synthetic passage written for this example; no participant data
Candidate quotation
Try
Report
More than one match
Reported as AMBIGUOUS_MATCHES · exit 1
Every occurrence is reported, overlapping ones included; the checker does not choose between them, so attribution is left to the analyst.
- Matches
- 2
- Matching version
- episode-v1
- Characters 56–76 · bytes 56–76 · line 2
- Characters 134–154 · bytes 134–154 · line 3
Character and byte offsets count from zero and exclude the end; line numbers start at one.
Not assessed by the checker
- What the words mean
- Whether the context supports the reading
- Permission to reuse the quotation
- Whether the source is authentic
This browser does not compute fingerprints on this page; the report shows the saved two-speaker request (request 3).
What remains proposed
- The skill, templates and checker work on the supplied synthetic example. Empirical evaluation with research materials and researcher judgments remains proposed.
- The AI-written memo contains an unsupported negative-case inference, and no researcher has adopted its proposals. Neither exact matches nor software tests establish interpretive validity.
- The checker does not extract PDFs, authenticate sources, verify declared permissions or assess meaning. The toolkit is an independent implementation, not endorsed by the method’s authors.