You open an article because the headline promises a clear answer. The prose is fluent. The structure looks professional. The author sounds confident. Ten years ago, you might reasonably have assumed that another person researched and wrote it. Today, that assumption needs checking.
Generative AI can produce articles, reviews, commentary and social posts in seconds. Sometimes a person uses it as a tool: to suggest a structure, summarise supplied material or identify questions worth exploring. Sometimes the system produces most of the text. The visible result may look equally polished in both cases.
That does not make AI-assisted content automatically false. It does make authorship, responsibility and evidence harder to see. This is the information environment Harry Kemsley and Sean Corbett describe in Fact & Friction as Synthetic Reality.
The old assumption: a person wrote what you are reading
Human writing has never guaranteed accuracy. Newsrooms, experts and commentators can make mistakes, frame evidence selectively or publish before the facts are settled. The change introduced by generative AI is therefore not a fall from perfect human truthfulness. It is a change in production conditions.
A machine can generate plausible language at low marginal cost and extraordinary speed. That reduces one of the natural limits on how much material can be produced. When distribution systems already reward attention and engagement, synthetic content can enter the same loops explored in earlier Fact & Friction episodes: it can be recommended, repeated, emotionally packaged and encountered from several directions.
The Point of Friction is simple: we still often judge online material with instincts shaped by human authorship, while more of the environment can be produced or assisted by machines. Our habits have not fully adapted.
Why “spot the AI” is the wrong challenge
People often look for stylistic clues: generic phrases, uniform sentences, repetition, vague references to “experts” or an oddly polished tone. Those clues can prompt a closer look, but they cannot establish authorship. Human writers can be formulaic. AI output can be edited by a person. A skilled user can ask a model to imitate almost any voice.
Detector tools have limits too. They may be useful as one signal, but a score is not proof. Current Ofcom research also records uncertainty among children and young people about whether online material is real or AI-generated. The educational task is therefore broader than learning to catch visual glitches or verbal habits.
A better question is not merely, “Was this made by AI?” It is, “What would make this trustworthy?”
Check authorship, sources and corroboration
Start with the author. Is a person or organisation named? Can you find a credible body of work? Is there an editorial policy, corrections process or accountable publisher behind the page? Anonymous authorship is not automatically deceptive, but it gives you less to assess.
Then inspect the sources. A list of impressive-looking citations is not enough. Open them. Do the links exist? Do they support the claim being made? Are they primary or authoritative sources, or several summaries pointing back to one unverified assertion?
Finally, read laterally. Leave the page and see how independent, reputable sources describe the same event or claim. Repetition is not corroboration if every page traces back to the same origin. The practical discipline is to compare genuinely independent evidence.
Harry’s concise advice in the episode is: “Ask it to show you the sources.” The next step matters just as much: read those sources yourself.
Use AI, but keep a human responsible
The episode does not argue that people should reject generative AI. Harry and Sean describe a useful hybrid model: ask a system for a framework or a set of factors, then apply human knowledge, research and judgement. The person remains responsible for what is accepted, rejected and published.
That distinction is especially important when the subject affects health, money, safety, education or public decisions. Fluency can make weak information feel finished. Human responsibility means slowing the process down long enough to test it.
What teachers, parents and teenagers can do
Teachers can turn authorship uncertainty into a source-tracing exercise. Give pupils several accounts of the same event and ask them to identify the author, evidence, original source and editorial accountability. The goal is not to guess which paragraph “sounds like AI”; it is to build a defensible judgement.
Parents can investigate with a child rather than test them. Choose one surprising post and ask: Who made this? What is it trying to make us feel or do? Where else is the claim reported? Could any part have been generated or altered by AI? A shared inquiry builds confidence without implying that either generation is uniquely gullible.
Teenagers can treat verification as autonomy. Before reposting something that creates an instant reaction, pause long enough to check the account, search for the original source and compare one independent report. You do not need perfect certainty about authorship to decide that a claim has not yet earned your trust.
A three-minute Synthetic Reality check
| Check | Question |
| 1. Authorship | Who created or published this, and can you verify that identity? |
| 2. Evidence | Which sources support the main claim, and do the links say what the article says they do? |
| 3. Independence | Can you find separate reporting or primary evidence that does not trace back to the same origin? |
| 4. Incentive | Is the content trying to inform, sell, provoke, recruit or simply keep you engaged? |
| 5. Action | If the claim matters, what would you need to know before sharing or acting on it? |
Reflection and next step
The next time a headline makes you react instantly – with anger, curiosity or disbelief – pause. Did the evidence create that reaction, or did the packaging? Then ask the more useful question: what can you verify?
Listen to Fact & Friction for the full conversation, and explore Luminae’s work on information, judgement and agency at www.luminae.org.
Source and Editorial Note
The mechanism, examples and practical advice are derived from the Episode 4 recording and canonical transcript. The reference to current children’s uncertainty is external context from Ofcom’s Children and Parents Media Use and Attitudes Report 2025-6. The article does not claim that all AI-assisted content is false, that style reveals authorship, or that a detector provides proof.
From Fact & Friction
This Insight accompanies “Synthetic Reality: AI is Now Writing the News”.
Listen to the companion episode
Explore the connections
- Why Online Content Makes You React: How A/B Testing Shapes Headlines and Feeds
- Critical Thinking for the Next Generation
Further source reading
Associated Press: standards around generative AI
A newsroom example of verification and editorial responsibility, rather than a method that can reliably detect every AI-written text.
Explore the research
Academic research notes supporting this Insight are available on request. Use our contact form and tell us which topic interests you.