Lesson Report:
Title
Algorithmic Creation, Engagement Formulas, and Disinformation Workflows
In this session, students moved from recognizing AI’s stylistic fingerprints in social media writing to dissecting engagement formulas in short-form video. The class then applied these insights to design low-cost, scalable workflows for polluting public discourse (disinformation simulation) around “walkable cities,� foreshadowing an upcoming discussion on democratic implications.
Attendance
– Students mentioned absent: 0
Topics Covered (chronological)
1) Admin: Feedback timeline
– Instructor noted delays returning feedback on a prior assignment and aimed to finish within two days.
2) Cold Open: Detecting AI-authored social posts (LinkedIn screenshot)
– Prompt: “Where did this message come from and why do you think so?â€� Focus on the author’s process rather than the platform.
– Indicators students identified for AI authorship:
– Formulaic structure: “It’s not about X; it’s about Yâ€� flips; list-making with micro-headings (e.g., “Know your edge:â€� followed by explanation).
– Tone and vocabulary: polished, aspirational/motivational register; recurring LLM words/phrases (“delve,â€� “shed light on,â€� etc.).
– Punctuation/formatting: flawless punctuation; frequent use of em-dashes where humans might use parentheses; overall “too cleanâ€� copy.
– Discussion: AI’s influence on human style (people adopting/avoiding “AI-soundingâ€� words and punctuation); expectation that models will increasingly adapt to individual voice-cloning and style-mimicry.
3) Framing the week: From algorithmic curation to creation
– Shift from last week’s focus (algorithms curating content for your “digital selfâ€�) to this week’s focus: AI systems and creators now producing the content we consume—and what that means for propaganda.
4) Activity Setup: Revisit short-form videos (TikTok/Reels/Shorts) from last week
– Task (written in chat for clarity):
– Re-open your five saved short videos.
– Select the one with the highest engagement (likes/comments/shares).
– Analyze the “formulaâ€� used to leverage the algorithm (hook, structure, overlays/captions, emotional triggers, watch-time engineering, controversy, etc.).
– Work time: ~10 minutes.
5) Share-outs and live analysis of examples
– Cross-platform access note: TikTok blocked locally; instructor used a VPN; some links failed; Instagram loaded slowly.
– Example A (comedic remix, “Bye Nancy! Bye Mom!â€� with a shark VO overlay):
– Hook via caption: suspenseful/teasing copy primes viewers to “watch to the end.â€� Long captions possibly help ranking.
– Pattern: unexpected payoff through a spliced gag; comedic incongruity keeps attention.
– Meta-observation: caption likely AI-written (em-dash, polished cadence); captions serve engagement goals more than description.
– Example B (feel‑good compilation):
– Emotional contagion: viewers mirror visible emotions (joy/tears).
– Compilation design: stitching multiple high-valence clips typically boosts retention (even if the clips are unrelated); this one used two clips but mimics the more common 5–6 clip format.
– Example C (luxury lifestyle/entrepreneur montage—Royal Rentals):
– Dual engagement strategy: aspirational flex (cars/jewelry/friends with businesses) plus controversy in comments.
– Algorithmic reality: all comments “countâ€�—supportive or critical—so engineered debate (viewers challenging authenticity/wealth claims) maximizes reach.
– Takeaway: maximum engagement often comes from eliciting argument either with the creator or between commenters.
– Example D (fully AI-generated “cats in bikinisâ€� genre):
– Novelty + cuteness + weirdness = attention; rapid, low-cost production via image/video generators.
– Comment-bait: “Is this real?â€� reliably spawns reply threads (sarcasm/explanations), amplifying distribution.
– Synthesis across examples:
– Recurrent engagement levers: suspenseful hooks; emotional peaks (joy/tears); jealousy/aspiration; controversy; novelty/absurdity; text overlays/captions; compilation/remix; “watch-to-endâ€� engineering.
– Strategic goal alignment: many creators design for the algorithm (retention, comments, shares) rather than for topic coherence.
6) Mini-lecture: Misinformation vs. Disinformation
– Definitions:
– Misinformation: false/inaccurate info shared without intent to deceive.
– Disinformation: deliberately false/misleading info shared to deceive, confuse, or sow conflict.
– Student examples:
– COVID-era soda “remedyâ€� myth (misinformation).
– Government/organizational black PR against critics (disinformation).
7) Group Simulation: Disinformation-for-Hire brief (walkable cities)
– Roles: Students as a consulting firm contracted by “City Planning and Development, LLCâ€� (lobbying group).
– Objective: Pollute the public conversation about walkable cities—not to persuade, but to create confusion, toxicity, and mistrust so fewer citizens engage.
– Constraints: Very small budget (~$100), one month.
– Deliverable: An automated content workflow that can:
– Prompts: 1–3 reusable LLM prompts for posts/scripts/captions/comments tuned to controversy, confusion, and cross-fire.
– Visuals: AI-generated images/videos aligned with the prompts (speed and low cost prioritized).
– Scaling: Produce ≥100 unique pieces/day (e.g., prompt variations, style templates, asset remixing).
– Distribution: Seeding plan across accounts/platforms to stimulate arguments and maximize comment activity.
– Breakouts: 5 rooms, ~13 minutes. Note: Present group strategies at the start of Thursday’s class.
8) Closing and upcoming materials
– Readings assigned (short; skim acceptable):
– Case study: “Alice Donovanâ€� — a fabricated journalist linked to Russian intelligence, used to seed divisive blog posts (2016).
– “Soon the Supply of Disinformation Will Be Infiniteâ€� — on AI-driven scaling of disinformation creation/distribution.
– Next steps preview: Use the simulation and readings to examine democratic consequences and potential countermeasures.
– Reminder: Next video journal due Sunday; details to be discussed Thursday.
Actionable Items
Urgent (before Thursday)
– Post both readings and links on the LMS: “Alice Donovanâ€� case and “Soon the Supply of Disinformation Will Be Infinite.â€�
– Clarify Thursday share-out: Each group should come ready with a concise plan (1–2 slides or a short doc) covering prompts, visuals, scaling, and distribution. Provide a template to standardize submissions.
– Return overdue feedback or communicate a firm return date if still pending.
– Resolve video access logistics:
– Provide a fallback plan for TikTok (e.g., request Instagram/YouTube mirrors, or ask students to screen-record clips).
– Test screen-sharing on a single monitor or preset the correct display to avoid window mix-ups.
This week
– Publish the Video Journal assignment details (prompt, length, criteria, submission method) and rubric; remind of Sunday due date.
– Create a shared class list of “engagement formulasâ€� derived from today’s examples (hook types, emotional triggers, controversy tactics, caption conventions) to reference during the disinformation simulation.
Longer-term
– Plan a follow-up segment on AI-style detection limits and best practices for authenticity (how to write “like a humanâ€� without triggering AI tells; ethical guidelines).
– Consider a brief tool demo next week: low-cost ways to generate/scale assets responsibly (and how platforms detect/limit coordinated inauthentic behavior).
Homework Instructions:
ASSIGNMENT #1: Group Strategy Prep — Disinformation-for-Hire Workflow (Walkable Cities)
You will consolidate your breakout-room work into a concise, ready-to-share plan that shows how you would use AI and algorithmic “engagement formulas� to pollute the public conversation on walkable cities with maximum impact, minimum cost, and minimal human labor—connecting directly to today’s analysis of formulaic, AI-sounding language, emotional hooks, controversy engineering in comments, and short-form video tactics.
Instructions:
1) Reconnect with your breakout group.
– Gather the notes you started in class on the “disinformation for hireâ€� brief for City Planning and Development, LLC.
– If you missed class, join the group you were assigned to or coordinate with classmates to review their notes.
2) Define the goal and constraints in one sentence each.
– Goal: Create confusion, argument, and mistrust around walkable cities (reduce healthy engagement with the topic rather than persuading to one side).
– Constraints: Budget roughly ≤ $100; one-month campaign; design an automated workflow that can output ≥ 100 unique pieces/day.
3) Draft 1–3 AI prompts for text content (posts, captions, comments, video scripts).
– Use prompts that reliably produce:
– “Listyâ€� structures with mini-headlines (mirroring what we noticed about LLMs’ list tendencies).
– Emotional/aspirational tones or deliberately polarizing frames (like the lavish-lifestyle flex video that provoked arguments in the comments).
– Hook lines that force watch-to-end behavior (e.g., “This is so diabolical—wait for it,â€� or a tantalizing “not about X, but about Yâ€� structure).
– Include at least one prompt tailored for generating reply-comments that seed controversy (Instagram/TikTok/YouTube comments that invite arguments count the same for the algorithm, as we discussed).
4) Plan your visuals and formats for speed and volume.
– Specify which tools you’d use (e.g., an AI image/video generator of your choice) and what types of visuals you’ll generate (cute/absurd compilations, emotional reveals, wealth/status flexes, outrage bait, stitched feel-good clips, etc.), drawing on examples we analyzed.
– Decide your aspect ratios (9:16 for Reels/Shorts/TikTok) and on-screen text/captions. Note that longer captions can boost reach and that “wait-for-itâ€� overlays drive retention.
5) Design the automation/batching workflow to hit 100+ pieces/day within budget.
– Describe how you’ll:
– Batch-generate scripts/captions/comments with LLM prompts (parameter variations for uniqueness).
– Batch-generate visuals (template prompts plus small prompt tweaks).
– Assemble short videos quickly (reuse music, quick templates, consistent end-cards).
– Include a rough time and cost estimate showing you can stay under ~$100 for the month.
6) Specify distribution and comment-engineering.
– Identify 2–3 platforms (e.g., Instagram Reels, TikTok, YouTube Shorts) and your cross-posting process.
– List your hashtag/keyword approach and how you’ll seed first-comments to spark debates (remember: arguments in comments supercharge reach).
– Note any basic account hygiene (rotating accounts, varying bios, posting cadence) to minimize takedowns.
7) Add a brief ethical note (2–3 sentences).
– Acknowledge the democratic risks (confusion, erosion of trust) we’re studying; this exercise is for analysis/awareness, not endorsement.
8) Prepare to share at the start of Thursday’s class.
– Deliverable: Either a one-page outline with headings (Prompts, Visuals, Scaling, Distribution, Budget) or a 3–5 minute group walkthrough covering those points.
– Have 1–2 concrete examples ready (sample prompt + sample caption/overlay + sample visual concept).
ASSIGNMENT #2: Read and Prepare — “Alice Donovan� Case + “Soon the Supply of Disinformation Will Be Infinite�
You will read/skim two short pieces that connect our in-class findings about AI-shaped writing, emotional hooks, and controversy tactics to real-world disinformation operations and to the near-future of automated, infinite content supply; you’ll come prepared to discuss how these readings map onto the workflow you designed.
Instructions:
1) Access the two short readings posted for this week:
– “Alice Donovanâ€� case (a fake reporter producing divisive political content).
– “Soon the Supply of Disinformation Will Be Infinite.â€�
2) Skim both thoroughly (about 15–20 minutes each).
– Focus on: the actors’ intent (disinformation vs. misinformation), tactics used to pollute conversations, the role of automation/scale, and why these tactics reduce meaningful public engagement.
3) Take brief notes.
– For each article, jot 3–5 bullet takeaways and 1 quote or example that stood out (include a short note on why it’s revealing or effective).
4) Connect readings to today’s lesson.
– In 3–5 sentences per article, answer:
a) Which engagement “formulas� we identified (lists, emotive hooks, engineered controversy, long captions, watch-to-end structures) show up in these cases?
b) Which parts of the tactics could be scaled today with inexpensive AI within a ~$100 budget?
c) What signals might help ordinary users detect such content?
5) Prepare to contribute at the start of Thursday’s class.
– Bring your notes and be ready to share 2 discussion points and 1 question that link the readings to your group’s disinformation workflow.
6) Optional enrichment (recommended):
– Save one current example from your feed that looks AI-generated or formula-driven (screenshot or link) and be ready to connect it to a tactic from the readings.