Opening Question

Before we talk about AI writing — let's see where the class stands right now.


When you use AI to write something, how much do you edit the output?

I use it almost word-for-word — minimal changes
I edit lightly — fix a few words, adjust the tone
I rewrite significantly — the AI gives me a starting point
I only use AI for ideas — I write the actual words myself

Discussion: There's no wrong answer here — but by the end of today's class, think about where you want to land. The goal isn't to use AI less. It's to use it more deliberately, with more editing and more of your own voice in the output.

AI Writing Workflows

AI doesn't replace the writing process — it changes where your effort goes. Instead of staring at a blank page, you're editing, directing, and shaping output that already exists.


What AI Can Do in a Writing Workflow

Drafting — Generate a first version fast

Give the AI a clear brief and let it produce a full draft. The draft won't be perfect — that's not the point. The point is to get past the blank page. Your job shifts from writing to editing, which is almost always faster. Drafting is where most communication students start using AI, and it's one of the highest-value applications when done with a strong prompt.

Common uses: First draft of a press release, news story lede, show notes, ad copy, bio, caption set, social post series.

Editing — Ask AI to improve what you already wrote

Paste in your own writing and ask the AI to tighten it, fix grammar, improve clarity, or cut it to a specific length. This keeps your voice in the piece while using AI for the mechanical parts of editing. It's often more effective than asking AI to write from scratch, because it starts from your ideas and structure.

Here is a paragraph I wrote. Edit it for clarity and cut it to under 100 words without losing the main point. Do not change the tone. [your paragraph]
Tone-Shifting — Rewrite the same content for a different context

Take one piece of content and adapt it for a completely different channel or audience — without rewriting from scratch. A press release becomes a social caption. A formal bio becomes a casual podcast intro. A news story becomes an email newsletter blurb. AI can make these shifts in seconds; a human would spend thirty minutes on each one.

Here is a press release I wrote. Rewrite it as a 120-character Instagram caption for a college student audience. Keep the key facts. Use an informal, direct tone — no corporate language.
Summarizing — Condense long content into shorter formats

Paste in a long document — a transcript, a research report, a meeting recording — and ask the AI to pull out the key points, quotes, or facts you need. This is one of the most reliable AI use cases because the source material is already accurate; the AI is just reorganizing it, not inventing anything.

Here is a transcript from a 45-minute interview. Pull out the three most quotable moments and list the five most important facts the subject shared. Format as a bulleted list. [transcript]

AI Across Media Writing Types

These workflows apply across every professional writing format you'll encounter in communications:

Professional Bio Draft from a résumé; tone-shift for different audiences
Press Release Draft structure, then fill in accurate facts yourself
Podcast Show Notes Summarize transcript → edit for SEO and readability
Ad Copy Generate 5 variations; pick and refine the best one
Social Captions Tone-shift a long post into platform-sized pieces
Podcast Script Draft from an outline; record your own voice on top
News Story Lede Generate multiple angle options; choose and develop
Website Copy Draft per section; humanize for brand voice in edits
The rule across all of them: AI gives you the draft. You bring the facts, the voice, and the editorial judgment. No AI output should go anywhere without a human making final decisions about accuracy, tone, and appropriateness.

When AI Copy Fails

AI writing breaks down in predictable ways. Learning to recognize these failure modes is what separates a professional who edits critically from one who just pastes and submits.


Failure Mode 1 — Generic Voice

It could have been written for anyone. It wasn't written for you.

When a prompt lacks role, audience, and constraints, AI defaults to a safe, featureless middle ground. The result sounds like the opening paragraph of a Wikipedia article — technically correct but completely devoid of personality or purpose.

AI Output (No Context Given)
"We are excited to announce the launch of our new communications program, designed to provide students with the skills they need to succeed in today's dynamic media landscape. This innovative curriculum will prepare graduates for a wide range of exciting career opportunities."
After Human Editing
"Susquehanna University's new Media Production track puts students in front of cameras, microphones, and real clients from day one — not just in a classroom. Sixty percent of graduates have landed jobs or internships before they walk across the stage."
Fix: Add audience, tone, and specific constraints to the prompt. Then treat the output as a first draft — not a final product. Replace every vague word ("dynamic," "innovative," "exciting") with a specific fact.
Failure Mode 2 — Wrong Facts

It sounded right. It wasn't.

AI does not look things up. It generates plausible-sounding text based on patterns in its training data. When you ask it about specific people, dates, statistics, quotes, or events it wasn't trained on — or trained on inaccurately — it will confidently produce wrong information. This is called hallucination, and it has ended careers.

Real pattern: A student asks AI to write a press release about an upcoming campus speaker. The AI invents a quote from the speaker, an award the speaker never won, and a publication credit that doesn't exist. The student submits it without checking. The speaker's assistant calls the university communications office to complain.
Fix: Never ask AI to generate facts it couldn't know. Provide the facts in your prompt and instruct it to use only what you give it. Use placeholders: "Mark any quotes as [QUOTE TK] — do not invent quotations." Verify every number, name, and date before anything goes out.
Failure Mode 3 — Missing Context

It answered the question you asked, not the one you needed answered.

AI takes your prompt literally. If your brief is incomplete, the output will be too. The AI won't ask what you meant — it will fill the gaps with its best guess, which is often the most generic option available. Context that feels obvious to you is invisible to the model.

Prompt Missing Context
Write show notes for my podcast episode about social media.
"In this episode, we explore the world of social media and its impact on modern communication. We discuss various platforms and how they affect our daily lives, featuring insights from industry experts..."
Prompt with Context
You are writing show notes for "The Comm Brief," a podcast for college communications students. Episode 7 is a 28-minute conversation with a former ESPN social media editor about TikTok strategy for sports brands. Audience: comm students, ages 18–24. Write 150-word show notes with: a one-sentence hook, three bullet-point takeaways, and a closing line pointing listeners to the guest's LinkedIn. Do not invent any biographical details.
Output: Specific, usable, structured — exactly what a podcast page needs.
Fix: Assume the AI knows nothing about your show, your organization, your audience, or your goals. Tell it everything. If you find yourself thinking "it should know that" — that's the context you forgot to include.

Class Activity

Pick your track and your task. Use Claude, ChatGPT, or Gemini to create something real. We'll review your output together — look for the three failure modes you just learned.


Step 1 — Choose your communication track
Step 2 — Choose your task

Your Starting Prompt

This prompt is already built using the seven-element framework from Lesson 2. Copy it, paste it into Claude, ChatGPT, or Gemini, and add your own details where you see [brackets].

While you work, watch for the three failure modes:
  • Generic voice — Does it sound like it could be for any organization?
  • Wrong facts — Did it invent anything you didn't tell it?
  • Missing context — Did it misunderstand or ignore part of your brief?
Try this after your first output: Add one sentence at the end of your prompt — "Before you begin, ask me 3 to 5 questions that would help you produce a better result." See what the AI thinks is missing from your brief.
Choose a track and a task above to see your starting prompt.

Authorship & Voice

If AI wrote 80% of your press release, who authored it? This is no longer a hypothetical question — it's one your future employer may ask you directly.


"If AI wrote most of it, does it still have your byline?"

Authorship has always meant more than just typing the words. It means being accountable for the accuracy, fairness, and intent of a piece. When AI generates your content, you are still responsible for every word — even the ones you didn't write.

The question isn't whether you used AI. The question is whether you can stand behind the work as if you wrote every word yourself. If the answer is no — you haven't done enough editing.


What Does Disclosure Actually Look Like?

These are real scenarios you will face in internships, freelance work, and staff positions. Industry norms are still forming — but the direction is clear: more transparency, not less.

80%
AI-drafted press release

You briefed the AI, it wrote the structure, you filled in real quotes and facts, you edited the voice. Standard in PR now. Disclosure varies by firm — but know your firm's policy before you submit.

50%
AI-assisted news story

AI summarized documents or structured the lede. You reported, verified, and wrote the voice. Many outlets now require a disclosure line. The AP has a standing policy — look it up before your first internship day.

30%
AI-edited your own draft

You wrote it; you asked AI to tighten the grammar and cut 100 words. Most style guides don't require disclosure at this level — but your organization might. When in doubt, ask.

5%
AI-generated ideas only

You brainstormed with AI but wrote everything yourself. Generally no disclosure required — this is no different from bouncing ideas off a colleague. But don't let "I used it for ideas" become a convenient defense when you actually used more.


Discussion Questions

For the class: What does your future employer expect you to disclose? If you were the editor, what would you need to know about how a story was produced before you published it?
For journalists: The AP Stylebook now has AI guidelines. SPJ (Society of Professional Journalists) has published ethical guidance. These aren't optional reading — they're professional expectations. How should disclosure look in a byline?
For PR students: A client hires you to write their press materials. They assume human authorship. Do you have an obligation to tell them AI helped? What if the client specifically requests no AI use?
The one rule that holds across every scenario:

If you wouldn't put your name on something a human had written for you without reading and editing it — you shouldn't put your name on something AI wrote for you without reading and editing it either. The byline is yours. The responsibility is yours.

Glossary — Lesson 3

New terms introduced in this lesson on AI writing workflows, authorship, and professional media production.


Writing & Workflow Terms
Tone-Shifting

Asking an AI to rewrite existing content in a different register, formality level, or style for a new channel or audience. Tone-shifting is one of the most practical AI writing skills — the same core content can become a formal press release, an Instagram caption, an email newsletter blurb, or a podcast teaser without starting over. The key is specifying the new audience and platform precisely, not just saying "make it more casual."

Multi-Tool Workflow

A production process that uses more than one AI tool in sequence to complete a project. For example: use NotebookLM to analyze a research document → feed key findings into Claude to draft a press release → use Canva Magic Write to adapt the release into social graphics copy. Multi-tool workflows let you use each tool for what it does best, rather than forcing one tool to do everything. They also distribute your dependence, which matters as individual tools change or go down.

Iterative Refinement

The process of improving AI output through successive follow-up prompts, treating each response as a step toward the final product rather than an all-or-nothing result. Professionals rarely accept a first AI draft unchanged. They refine the voice, correct the facts, adjust the structure, and push back when the output misses the mark — often across three to five rounds of prompting before the content is ready to use.

Authorship & Ethics Terms
Voice

The distinctive personality, perspective, and style that makes a writer's work recognizable. Voice is what AI most consistently fails to replicate — it defaults to a generic, safe register that sounds like no one in particular. Developing your own voice as a writer is one of the most durable professional skills you can build, precisely because it's the hardest thing for AI to mimic. When you edit AI output, your goal is to put your voice back into it.

Byline

The line crediting the author of a news story, article, or media piece, typically appearing at the top: "By [Name]." A byline is not just a name — it is a claim of authorship and accountability. When you put your byline on a piece, you are telling readers that you are responsible for its accuracy, fairness, and integrity. The byline question is at the center of AI disclosure debates in journalism: if AI wrote most of it, what does your byline actually mean?

AI Disclosure

A statement attached to a piece of content indicating that AI tools were used in its production and describing how. Disclosure norms vary widely by outlet, firm, and platform — but they are evolving rapidly toward greater transparency. In academic work, Susquehanna University requires disclosure on any AI-assisted assignment. In professional practice, check your organization's policy before assuming any level of AI use is acceptable or that none requires disclosure.

Hallucination

When an AI model generates text that is factually incorrect but presented with the same confidence as accurate information. Hallucination is not a bug that will eventually be fixed — it is a structural property of how large language models work. They predict likely-sounding text, not verified facts. In communications work, hallucination is a professional liability: an invented quote, a wrong date, or a fabricated credential can cause real harm to real people. Every AI output must be fact-checked before it is used.

Lesson 3 · Moving from Ideas to Media