Free enrolment | Fully online | Self-paced | ~10 to 14 hours across six modules plus a bonus seminar
Every qualitative researcher now works within reach of tools that will happily code a transcript, propose six polished themes, and supply supporting quotations, some of which do not exist. The debate about whether that help is legitimate has largely produced position-taking. This course produces something else: procedure.
You will run a language model against your own unassisted analysis, measure where it holds and where it collapses, catch it fabricating, audit its themes back to the source, use its very conventionality as a mirror for your own assumptions, and leave with the two documents current academic integrity standards actually ask of you: a disclosure statement and an evidence-based conditions note, both built from checks you performed yourself.
The course takes no side in the AI debate. It teaches you to hold whichever side you choose with evidence.
Its two mottos: fluency is not analysis. Demand receipts.
Every module follows the same rhythm: watch a short video in which the instructor demonstrates the workflow live on a purpose-built interview dataset, model included, mistakes included; practice the same workflow yourself, offline, on practice transcripts reserved for you, with structured workbooks that compute your figures as you go; pass a short quiz to unlock the next module. You never need to share your work; the quizzes gate progress, the practice builds your evidence.
You will need: a browser, a spreadsheet application, and access to any major AI chat tool. No participant data is ever used or required; the course runs entirely on a synthetic teaching corpus of founder interviews, so you practice the risky moves where nothing is at risk.
A short brochure instead of a welcome lecture: where the debate stands, the three rules that protect your results (synthetic data only; your reading sealed and dated before any model contact; no model claim trusted unsearched), what to download, and the course’s honestly labeled position.
You will know: what the course will and will not claim, and what evidence you are about to build.
The engine-room module. What a language model actually produces and why it is non-deterministic; the three-part data gate (consent, institutional approval, provider terms) that decides whether real data may ever touch a model; and the fabrication hunt, in which you make a model invent quotations and catch it doing so.
You will be able to: run the consistency and fabrication checks on any model, keep a decision log, and state exactly when participant data may not be processed.
The most important module contains no AI at all. You code a transcript by hand, write a dated memo of what struck you and why, mark the passages that resist your categories, and set it all aside. That dated record becomes the instrument every later module measures the model against; without it, you can never again know what you would have seen on your own.
You will be able to: produce a defensible unassisted coding and a dated salience memo, and explain why it must exist before any model contact.
The most defensible delegation: you fix a codebook, with definitions, exclusions, and stated provenance, and the model only classifies. Then you measure it: agreement per code (the headline number hides exactly the code that matters), stability across runs, and an error analysis that catches the model force-fitting codes against your own written exclusions. You end with a verdict memo: what you would delegate, under what conditions, and what never.
You will be able to: build and apply a machine-usable codebook, evaluate model coding per code rather than by headline, and write an evidence-based delegation verdict.
You ask the model for themes twice, once cold, once through a human-gated staged pipeline, and then audit everything: every quotation searched in the source, every theme binned as grounded, plausible-but-generic, or unsupported. You meet the oracle effect, the pull to read fluent output as insight, and you compare both passes against your Module 2 memo with two honesty columns: what the model found that you missed, and what you found that it never will.
You will be able to: run a full traceability audit, distinguish grounded findings from statistical mimicry, and report both honesty columns without flinching.
The course’s most distinctive move, built on a method proposal currently under peer review and presented with exactly that status. Models tend toward the canonical reading, the interpretation a field has made most available. This module turns that limitation into an instrument: elicit the model’s reading of a puzzling passage while revealing nothing of yours, locate the divergence, and interrogate the gap until you can name, in one sentence, an assumption of your own you could not see before. Then decide on the record: defend, revise, or flag.
You will be able to: run the estrangement protocol on any passage, name the assumption a divergence exposes, and stress-test the method itself.
The current state of academic integrity on AI use reduces to five requirements: disclose, own it, verify, protect the data, keep records. You will meet all five already holding the evidence, because the course made you build it. You draft the two documents your thesis actually needs, a disclosure statement generated from your own decision log, and a one-page conditions note in which every delegation cites your own figures, then check both against an integrity checklist, including the sharpest question in the course: what evidence would show your safeguards had failed?
You will be able to: write a disclosure statement a reader can verify, and a conditions note you could defend in a viva.
The debate’s landmark exchange, a mass rejection letter and its sharpest response, presented at full strength on both sides; a five-position map of how much power a technology has, with the levels insight that explains why the camps argue past each other; the standard objection (“it was only designed to predict text”) dissected into its four failing parts; two fatalisms refused; and a deliverable: a half-page position statement, conditional use, principled refusal, or open trial, that you can place directly in a methodology chapter or ethics application.
You will be able to: place any claim in this debate before answering it, and state your own position knowing exactly what it costs.
Doctoral researchers and supervisors in the social sciences and management; qualitative researchers weighing these tools for the first time; methods instructors seeking teachable, evidence-first workflows; and skeptics, who are warmly expected: the course’s first rule is that the AI stays switched off until your own reading exists.
Your own measured evidence (agreement figures, audit tables, an estrangement record); a portable template library covering every workflow for any coding setting; your disclosure statement and conditions note, drafted; and one habit above all: your reading first, dated, before the model’s, every time it matters.
Enrol free at BlendedIQ.ai. Start with the Course Overview today.
