The Flipped Library Classroom: Designing a First-Time User Workshop That Actually Works

If you're about to teach a 50-minute library workshop to students who have never typed a word into PowerSearch, you've already felt the squeeze. The students don't know what a discovery layer is. They don't know the difference between a Library Catalog and an article database. And they almost certainly don't know that the AI tool they're going to use first can confidently invent a citation that doesn't exist. One session, one shot, twenty-five strangers. This tutorial walks you through how to design that session as a flipped workshop — the pre-class micro-lesson, the live demo, the hands-on retrieval loop, the AI verification step, and the wrap-up — drawing on the strongest evidence and the cleanest practitioner cases available as of September 2026.

The voice here is for you, the librarian or first-year-experience instructor who has to run the session. The students are your audience, but you are the reader. Every example is something you can adapt into a sentence you read aloud, a screen you project, or a card you hand out.

1. Why Make This Flipped — And What the 2026 Evidence Actually Says

Flipped classroom is the design where students watch or read the core explanation at home, and class time is spent doing and discussing. The term traces back to two Colorado high-school chemistry teachers, Jonathan Bergmann and Aaron Sams, who in 2007 started screen-recording their lectures for students who missed class. To their surprise, the students who hadn't missed anything started using the recordings too. That flipped the question from "how do we re-teach the absent ones?" to "what is the best use of face-to-face class time if students can watch the lecture on their own?" The ISTE/ASCD book in 2012 fixed the label in the fieldBergmann & Sams 2012.

For ten years after that, librarians borrowed the idea and adapted it. Sara Arnold-Garza's January 2014 C&RL News article — the first peer-reviewed flipped-library assessment article in ACRL's flagship publication — described a Towson University pilot across 14 sessions, 7 librarians, and 14 courses (10 freshman-level and 4 upper-level across education, political science, sociology, computer science, and English). More than 90% of the 148 student respondents agreed the flipped format helped; 86% said they learned from the in-class activities; and 45% still said they'd rather have class time devoted to explanation. Arnold-Garza read that last number not as a failure but as an honest signal: students can find a model helpful and still not prefer it. The activities themselves have to be worth giving up the lectureArnold-Garza 2014.

The cleanest current evidence on whether flipping actually works is the August 2026 meta-analysis by Jeongwon Lee and Hae-Deok Song in Educational Technology Research and Development. They pooled 121 empirical studies. The headline numbers: teaching-related variables on learning outcomes r = .352, learning-related variables r = .303. Both sit just above the medium threshold by Cohen's cut-points. The four highest-effect moderators tell you exactly which levers to pull:

The biggest design finding, though, is that teaching-related effects are significantly larger when pre-class materials and in-class activities are diverse. One video plus one walkthrough plus one paired exercise beats three videosLee & Song 2026.

The conceptual model behind that finding is the one to keep in mind while you plan. Pre-class materials and teaching-related variables (course design, instructor characteristics) and learning-related variables (task factors, learner characteristics) all feed into cognitive and affective outcomes through in-class activity. The model has no arrow weights and no significance markers; it is qualitative. That's deliberate — Lee and Song publish the moderation numbers separately in their tables. What the model gives you is the map: don't just swap "lecture at home for homework in class," change what students do with each other while you are in the room.

Caption: Conceptual model of factors influencing flipped-learning outcomes (Lee & Song 2026, Fig 1). Pre-class materials, teaching-side variables, and learning-side variables converge on in-class activity, which feeds both cognitive and affective outcomes. The diagram carries no numerical coefficients — it is a factor map, not a regression result.

For your workshop, the model is essentially permission. Pre-class video alone won't do it. The in-class hands-on is the engine. And the strongest single move you can make is to give students some autonomy in what they search for — that's not a nicety, it's the moderator with the largest effect size in the entire meta-analysis.

2. The Pre-Class Micro-Lesson: Just Enough to Make Live Time Productive

The job of the home video is not to teach. The job of the home video is to load the one piece of context students need so the live session can be about doing. If the video tries to teach everything, the live session becomes Q&A on the video, and you've just moved the lecture to a different room.

The conceptual scaffold for what the home video should carry comes from the ACRL Framework for Information Literacy for Higher Education. Six threshold concepts, alphabetical: Authority Is Constructed and Contextual, Information Creation as a Process, Information Has Value, Research as Inquiry, Scholarship as Conversation, and Searching as Strategic Exploration. For this workshop, you'll lean on three of themACRL Framework 2016.

• Searching as Strategic Exploration — covers the whole workshop. Its core idea is that searching is non-linear and iterative; you determine initial scope, use divergent and convergent thinking, design and refine as results come in, and persist when the first attempt fails.
• Authority Is Constructed and Contextual — covers your source-evaluation block. Authority varies by community and context; indicators of credibility depend on what you need the source to do.
• Information Has Value — covers the "when the book isn't there" escalation. Access isn't free, but the library reduces the friction.

The pre-class video, then, has to teach one thing under the first frame: that the discovery layer in your library is doing two different jobs and uses two different result types depending on which one students point it at. Put two screenshots side by side. The left one shows a Library Catalog search for a book title, with the View it button (ebook link) and the Map it button (physical shelf location). The right one shows an Article+ search for a journal article title, with the Get PDF button and a Find similar articles link. That's the entire video: "These are two different search boxes, doing two different things, returning two different buttons. Press the right button for the thing you actually need."

The screen labels in your library's discovery layer will look different — PowerSearch, Summon, EDS, and Primo all label this split slightly differently. The split itself is universal. Use your own library's screenshots; the lesson is "two boxes, two jobs," not "find these exact words."

A second anchor for the pre-class design comes from Jennifer Ferguson's 2023 portal: Libraries and the Academy article on the Tufts Tisch Library redesign. Ferguson's User Experience and Student Success Department rebuilt their entire first-year information literacy program during summer 2020, moving the foundational content into asynchronous Canvas modules and reserving synchronous class time for active work. The article explicitly frames the redesign as "a sustainable program for the future," not a pandemic stopgap. The lesson for you is that the async module's job is to load "just enough context for live time to be productive" — and the synchronous session's job is to supply the "personal touch" of immediate help and active learningFerguson 2023.

The Arnold-Garza pilot, you may remember, capped pre-class videos at twenty minutes total. You don't have twenty minutes. You have six. The goal is to make sure every student arrives knowing what a call number looks like, what View it does, what Map it does, and that these are two different buttons on two different kinds of searches. That's the whole pre-class load.

If you keep the home video to that scope, students will arrive with the vocabulary they need, and you can spend the live session on the moves.

3. Live, Topic 1: From "I Want That Book" to "I Have It"

Open the live session with the question they actually came in with. Don't restart from the textbook table of contents. Instead, put a shared board up — physical whiteboard, Padlet, Mentimeter, whatever you use — and ask each student to write one question they have about the library. The questions you'll see in the first five minutes are usually some version of: how do I find a specific book the professor put on the syllabus, how do I get the PDF of an article the professor mentioned, and where do I go if I can't find either. The shared board is doing two things at once: it tells you what to teach next, and it tells students that their question is the one being addressedStellwagen, Rowley & Otto 2022.

Topic 1 takes the first book question and walks through it live on the projector. Pick a book the entire class has heard of — for a first-year cohort, this might be the common-reading text, a book mentioned in the syllabus, or Catcher in the Rye if you teach at a school where it still appears. Type the title into your discovery layer in front of them. Three things should happen on screen, one after another:

1. The result list comes back, and most records have two buttons visible: View it and Map it. Read the buttons out loud. Ask the room what they think each one does. Someone will guess right; let them say it.
2. Click View it. A new tab opens to the ebook. Show the URL — it's the library's proxy prefix, then the publisher's domain. Tell them: "If you're off campus, you'll be asked to log in with your student account. Same login as your email."
3. Click Map it. A new tab opens to the library's map view — the floor, the wing, the shelf. The screen also shows a call number. Stop here. Define a call number: it's the item's address on the shelf, the same way a dorm room number is the building's address. The Library of Congress system is alphabetical by subject; the Dewey system is numerical by subject; in either case the line of letters and numbers on the spine is how you find the book.

The conceptual anchor under this exercise is two ACRL frames: Searching as Strategic Exploration (you determine scope, you match strategy to tool, you refine as results come in) and Information Has Value (the library paid for access, the call number is the way you claim it). Both are now doing their job, not as words on a slide but as moves the student is about to make.

Now the hands-on. Pre-stage two or three exercise books with call numbers you can confirm in advance. Each pair of students gets one title and one call number. Their job is to walk to the shelf, find the book, and report back. This is the shelf-hunt race. Two rules: first pair to find their book shows the librarian and gets a small win (a sticker, a "you found it" — whatever your teaching culture supports); second pair that finds it gets the same. If your library doesn't allow students to run between the stacks, run it as a "report in" version — first pair to return to the desk with the book wins, and the others just have to find it before the timer ends.

The race is the autonomy-support lever in compressed form: each pair picks their own path through the shelf, makes their own micro-decisions about whether they're in the right wing, and gets an immediate success signal when they find the spine. The cognitive effect is the cognitive engagement one — students are physically executing the call-number-to-shelf mapping, not reading about it.

When the book isn't on the shelf, walk the escalation path out loud. This is the Information Has Value frame in operation: access isn't free, but the library has built a system to reduce the friction.

Three sentences, in order:

1. Place a hold. In the catalog record, click Place Hold. You'll go on the wait list and get an email when it's your turn. Don't walk to the shelf — there's nothing there to take.
2. Check Course Reserves. Many course-required books live here. The ebook is a direct link; the print copy is at the Ask Us Desk, and you check it out for a few hours at a time.
3. Ask Us Desk. A librarian will find a copy somewhere in the system — a different campus, a partner library, an interlibrary loan. Don't leave empty-handed.

That third step is also the moment you tell them that asking is free and that the desk staff have been doing this all day, every day, for years.

The whole Topic 1 arc — the question board, the live demo, the shelf-hunt, the escalation path — runs roughly twenty-five minutes. That leaves you time for one short transition and then Topic 2.

4. Live, Topic 2: From a Research Question to a Working Search

The user fixed the running example for this workshop in the intake chat: "What is the impact of social media on body image among Generation Z?" Use it. Every cohort has students who already know what body image is; every cohort has students who use social media daily. The question is broad enough that students will disagree about what counts as an answer, which is exactly what makes it a good teaching example.

The exercise structure is the one Lee and Song's meta-analysis named as the biggest cognitive teaching-side moderator: learner-learner interaction, with the affective side covered by autonomy support. In plain English: students work in pairs, and each pair picks a sub-question under the main one.

Walk the class through the move from natural-language question to a working search. The move has four parts, and you say each one out loud before students do it on their own laptops.

1. Pull out the core concepts. Three of them, in this example: social media, body image, Generation Z. Write them on the board.
2. Brainstorm related terms for each concept. For social media: TikTok, Instagram, Snapchat, online platforms. For body image: self-esteem, appearance, eating disorders, body dissatisfaction. For Generation Z: Gen Z, adolescents, young adults, teens. Write these under each core concept. The point of this step is to show that no single word covers a concept; the search has to combine words to be useful.
3. Combine with AND across concepts, OR within concepts. The Boolean phrasing is: (social media OR TikTok OR Instagram) AND (body image OR self-esteem OR appearance) AND (Generation Z OR adolescents OR young adults). Show this on screen. Read it out loud. Then show the same query inside the multi-box search in your discovery layer, with one box per concept — this is what most students will use first because the interface breaks the query into chunks they can read.
4. Run it. Look at the result list. Decide what to fix. If you get thousands of results, the query is too broad — add a date range, add a population filter, narrow one of the OR lists. If you get zero results, the query is too narrow — drop one of the AND concepts, swap a controlled-vocabulary term for a natural-language one, or run a single-concept search to see whether the database has any coverage of this topic at all.

The frame under all of this is still Searching as Strategic Exploration. The dispositions for that frame are the ones to read out loud at least once during Topic 2: mental flexibility, understanding that first attempts often fail, valuing browsing and serendipity, persisting in the face of search challenges, and knowing when enough is enough. These are the habits of mind the workshop is trying to install; the Boolean mechanics are the surface skillACRL Framework 2016.

The visual structure of the loop is the part students will remember. Searching is not a one-shot action; it is a cycle. You run, you look, you adjust, you run again. The Boolean mechanics are one way to describe the adjustment, but the deeper lesson is that the loop is the skill.

A useful structural note: have students run their own sub-question, not the class example. Bring their own question to the work — the autonomy-support lever. Each pair picks a sub-question (e.g., "How does TikTok use affect body image in 18–24 year olds?" or "What's the relationship between Instagram use and self-esteem in teens?"), runs the full loop, and ends with two anchor sources — one they trust, one they don't. That last step bridges you directly into the source-evaluation block.

5. Telling Good Sources from Bad: From CRAAP to SIFT, with an AI Layer

A decade ago, the source-evaluation tool you'd hand students was the CRAAP checklist — 27 questions, ranging from "What is the purpose of the source?" to "Does the source have a bibliography?" The checklist was designed in the mid-1990s when librarians adapted a 1981 collection-development tool from the Medical Library Association for use on early web pages. Mike Caulfield's 2018 critique in A Short History of CRAAP documents the problem: even when students have the checklist, they tend to focus on a single criterion (currency, authority, one of the 27) and skip the holistic review. The questions also reward close reading of an unfamiliar site, which is the opposite of what works on the modern web.

What works, Caulfield argues in the June 2019 hapgood.us post that formalized SIFT, is four moves welded to web techniques. Not a checklist of questions. Four verbs.

The unifying idea is recontextualizing — reconstructing the context the web so often strips away. The cognitive skill under it is lateral reading, which Sam Wineburg and Sarah McGrew documented in their 2019 Teachers College Record study of how professional fact-checkers evaluate online content. Fact-checkers don't read an unfamiliar site closely; they open new tabs and leave the site to investigate what the rest of the web says about it. Undergraduates and even professors tend to do the opposite — they read vertically, scanning design, domain, and the internal logic of the page. Lateral reading is faster and more accurateCaulfield 2019.

For the workshop, the practical move is to teach the four moves as verbs and to anchor them under the ACRL frame Authority Is Constructed and Contextual. The frame's concept statement is that information resources reflect their creators' expertise and credibility, and that authority is constructed by community recognition and contextual to the information need. SIFT is the operational layer under that frame; the frame is the why, SIFT is the how.

The 2026 complication is that the locus of evaluation has shifted. When students first encounter a source, increasingly they're not landing on a webpage. They're getting a synthesized answer from an AI tool and then maybe clicking a link. Caulfield's two 2026 Substack posts — "Look before you leap" (April 3, 2026) and "A SIFT Rebuild (in progress)" (March 29, 2026) — propose a draft "SIFT for AI" re-ordering:

1. Stop and get it in. Get the claim into the LLM in the fastest, most convenient way; don't over-engineer the prompt.
2. Investigate the evidence. Ask evidence-focused follow-ups. Gather and categorize context.
3. Find better sources. Ask the LLM for sources suited to the task — most academic, most recent, context on unknown organizations.
4. Track it down. Click the links, hover, verify that what the LLM surfaced is represented accurately. Get out of the LLM to check the LLM.

Caulfield explicitly frames this as a "parallel track" draft, not a replacement for the original four moves. The original still works for students who live in a link-based web; the rebuild is for students who live in an LLM-mediated webCaulfield 2026.

The single cleanest classroom evidence for why the AI layer matters is the March 2026 peer-reviewed audit by Cristo León and Michelle Kudelka at NJIT. They ran four AI tools (Gemini institutional, ChatGPT free GPT-4o, ChatGPT paid GPT-5.2 with Extended Thinking, Perplexity Academic) on a single canonical research prompt, twice each, in two retrieval postures: natural-language prompting (the default student workflow) and asking the tool to generate a Scopus-format Boolean query that the librarian then executed in Scopus. Sixteen audit runs total, top-k=20. The headline numbers: natural-language prompting returned 101 of 160 items (63.1%); Boolean translation lifted that to 129 of 160 (80.6%). Three failure patterns recurred in the natural-language runs — missing DOIs, non-resolving DOIs, and DOIs that resolved to a different work than the citation metadata claimed. Crucially, per-item correctness did not imply stability of the evidence set: Gemini's natural-language paired runs had a citation Jaccard of 0.00 and a DOI Jaccard of 0.057. The same query, run twice, returned essentially non-overlapping citation sets.

The audit's decision rule, written for student-facing inquiry, is workshop-ready: natural-language prompting is appropriate for exploratory term generation and early topic orientation. The evidence set used for screening, coding, and citation should come from database-executed Boolean retrieval with documented query logic and DOI-level verificationLeón & Kudelka 2026.

Caption: PRISMA-style corpus-cleaning workflow from León & Kudelka 2026, Figure 1. 220 records identified across all runs → deduplication and screening → 170 records identified → +27 LIS-expert records (n=197) → title/abstract screening → full-text eligibility → 37 articles included post-quality appraisal. The diagram shows how the librarian-augmented corpus was assembled; the comparison between librarian baseline and AI tools is in the paper's tables, not in this flowchart.

What the workshop borrows from this is two simple moves students can do at their own laptops in two minutes:

1. Ask the AI to help you build the search string. "I'm researching the impact of social media on body image among Generation Z. Give me 8 search terms I could combine, and 3 Boolean queries of varying breadth." Use the AI's output to seed your database search.
2. Then run the database search yourself, in Scopus / Web of Science / your library's article database, and verify each citation. Click the DOI. Does it resolve? Does the landing-page title match the citation? Does the abstract actually support the claim the AI attributed to the article?

This is the SIFT-for-AI workflow compressed: get the AI to draft the search, then track it down inside the database yourself. The AI is the search-string generator. The database is the verification surface.

Wake Forest's ZSR Library blog, in an August 2026 reflection on how their LIB100 course has changed in the AI era, calls this shift the move from "search-first" to "answers-first." Students now find too much information too easily, and the bulk of new instruction has to move to evaluating answers the student didn't assemble themselves. Their directional observation: hallucination rates in popular models have dropped (their classroom rate is now under 10%), but residual metadata errors persist — wrong DOIs, wrong volume / issue numbers, wrong author names. The verification method they teach is mundane: cross-check the citation in Google Scholar and verify the basic facts. They also name three research-specific AI tools — Scite, Consensus, and Google Scholar Labs — that search verified academic indexes rather than the open web, and that "reduce if not completely eliminate the hallucination problem" for academic searchKaufman & Ellis 2026.

The Collier 2024 master's paper at UNC, which surveyed 50 LibGuides on generative AI from U.S. academic institutions, found that most of the guides offered uncritical guidance and didn't anchor AI instruction in the ACRL framework. The LibGuides that did anchor in the framework gave students a stronger critical lens, especially under "Searching as Strategic Exploration" — where GAI is useful for keyword and search-string construction and for background context, and not useful for sensitive questions, complex synthesis, or verifiable citations. The lesson for the workshop is that the ACRL framework already carries AI literacy; you don't need a new framework, you need to apply the existing one with disciplineCollier 2024.

For the live session, the practical script is:

"When you ask an AI tool for sources, two things will happen. First, it will give you plausible-looking citations. Second, some of those citations will be wrong — sometimes completely invented, sometimes real but with the wrong DOI, the wrong volume, or the wrong author. The fix is the same one you'd use without AI: track the citation to the original. Click the DOI. Open the abstract. See whether the abstract actually says what the AI said it said."

That's one paragraph, one move, and it covers the bulk of what first-year students will encounter.

6. Wrap-Up, Limits, and the Next-Time Checklist

The last fifteen minutes of the workshop belong to the paired exercise. Each pair has a sub-question they've been working on under the Topic 2 example. Now they run the full loop themselves: pull concepts, brainstorm terms, combine with AND/OR, run the search, pick one good source and one they rejected, and be ready to defend both choices out loud.

The "good source / rejected source / why" structure is doing two jobs at once. It is the workshop's assessment, but it is not a quiz — you don't grade the answer, you ask the question "how did you decide?" The answer to that question is the move you're trying to teach: lateral reading, source verification, the SIFT-style traceback to the original. The pair that can defend why they rejected a source is the pair that has internalized the source-evaluation frame. The pair that can defend why they kept a source — pointing to the DOI that resolved, the metadata that matched, the author with subject expertise — has internalized the verification moves.

The exercise also closes the loop on the Authority Is Constructed and Contextual frame from ACRL. Students who can articulate why they decided are doing what the frame says information-literate people do: evaluating sources based on context and information need, not on a checklist of surface features.

After the share-back, spend the last five minutes on what this workshop didn't do, and what to change next time. Be honest about the limits. Three honest points to make:

1. The flipped effect is medium, not transformative. Lee and Song 2026 give you r = .352 on the teaching side and r = .303 on the learning side. Flipping helps; it doesn't fix. The effect moves up when pre-class materials and in-class activities are diverse, which is why the home video and the live hands-on are both doing real work. It also moves up when you give students autonomy in what they search for, which is why the paired exercise uses their own sub-question.
2. The 2026 evidence base predates the AI search shift. None of the 121 studies in the Lee and Song meta-analysis specifically cover what happens when students' first move is to ask an AI tool. The AI verification block in Topic 2 is your best current judgment about how to teach the moved locus of evaluation, not something the meta-analysis certifies. Next term, audit what your students actually did with the AI tools and adjust.
3. The next-term changes are small and concrete. You don't need to rebuild the workshop. You need to:
   • Add one explicit AI verification move to the home video. A 30-second clip showing "ask AI for a search string, then run it in the database, then click the DOI" is enough.
   • Diversify the home video. Pair the existing walkthrough with one screenshot tour and one short knowledge-check question. Arnold-Garza 2014 found that quizzes help signal that the video is required.
   • Move the SIFT move into Topic 2, not just Topic 2's tail. Students are already running searches; have them pause at the second result and ask "how do I know this is real?" before moving on.

The Tufts Tisch Library redesign is the long-arc reassurance: that same async-plus-sync flipped model has been running at Tufts since summer 2020, across multiple cohorts, and Ferguson explicitly framed it as "a sustainable program for the future." If the first run of your workshop is uneven, you don't have to throw the design out. You iterate on the same shapeFerguson 2023.

The single metric to track, if you track only one, is the share-back quality. The share-back is where you see whether students can articulate why they rejected a source. If that share-back is mostly "it didn't look official" or "the website was weird," the next-term change is to drill the lateral-reading move. If the share-back is "I clicked the DOI and it resolved to a different paper, so I rejected it," the workshop has done its job.

References

1. Bergmann, J., & Sams, A. (2012). Flip Your Classroom: Reach Every Student in Every Class Every Day. ISTE/ASCD. — Publisher page
2. Lee, J., & Song, H.-D. (2026). What makes flipped learning effective? A meta-analysis of teaching and learning factors. Educational Technology Research and Development. — Springer DOI
3. Arnold-Garza, S. (2014). The Flipped Classroom: Assessing an Innovative Teaching Model for Effective and Engaging Library Instruction. C&RL News 75(1), 10–13. — DOI
4. Stellwagen, Q. H., Rowley, K. L., & Otto, J. (2022). Flip This Class: Maximizing Student Learning in Information Literacy Skills in the Composition Classroom. Journal of Library Administration 62(6), 731–752. — DOI
5. Ferguson, J. S. (2023). Flipping the (COVID-19) Classroom: Redesigning a First-Year Information Literacy Program during a Pandemic. portal: Libraries and the Academy 23(1), 145–168. — DOI
6. Association of College & Research Libraries (2016). Framework for Information Literacy for Higher Education. American Library Association. — Official page
7. Caulfield, M. (2019). SIFT (The Four Moves). hapgood.us, June 19, 2019; with 2026 AI-era updates Look before you leap (April 3, 2026) and A SIFT Rebuild (in progress) (March 29, 2026). — Blog post
8. Collier, J. C. (2024). Generative AI Instruction in Academic Libraries: An Analysis of LibGuides and Information Literacy Principles. UNC SILS Master's Paper. — DOI
9. León, C., & Kudelka, M. (2026). Auditing GenAI Literature Search Workflows: A Replicable Protocol for Traceable, Accountable Retrieval in Student-Facing Inquiry. AI in Education 2(2), 8. MDPI. — DOI
10. Kaufman, A., & Ellis, E. (2026). Not the LIB100 You Remember: Evaluating Output in the AI Era. Z. Smith Reynolds Library "Beyond the Desk," Wake Forest University, August 6, 2026. — Blog post