TL;DR
Clarity of delivery is the strongest predictor of whether your text, video or talk gets a response from the audience. Stronger than the topic, stronger than charisma, stronger than the novelty of the idea. And it's measurable: an LLM reliably rates the clarity and structure of any text, producing scores that predict real human behavior: views and likes.
Intuitively, we think the winner is whoever has the more interesting topic or the more charismatic delivery. The data says otherwise. The brain physically turns away from unclear text: it's cognitive friction, an unpleasant effort. When an explanation is muddled or the structure falls apart, people don't think "hard but valuable." They lose the thread and leave. Conversely, easy-to-follow text produces a faint but steady positive signal: it feels good to think along with the author.
The method is simple: take any text, run a clarity-and-structure scoring prompt, get two scores on a 10-point scale, iterate. For stability, do several runs and average them. Works on articles, scripts, presentations, posts, pitches.
Want this for your own case? Apply this study to your task: AI adapts its method to what you are working on and writes the prompt.
Method Blueprint
STEP 1: Take a text — article, video script, presentation, pitch → prepared text
STEP 2: Paste it into the scoring prompt → get X,X (two numbers: clarity and structure)
STEP 3 (optional): Repeat 3–5 times → average for a stable result
STEP 4: Rewrite the weak spots → run the prompt again → compare the scores
All in one chat. One request = one scoring cycle.
Example in Practice
Task: Jake runs a YouTube channel about personal finance. He's written a script for a video called "How to Pay Off Your Mortgage Early." He senses something's off but can't tell what. He wants to check before filming.
Prompt:
You are an expert in evaluating the quality of public talks and texts.
Rate the following text on two criteria:
Clarity of delivery (1–10): how easy it is to understand the ideas, explanations and concepts —
no confusion, no unnecessary hedging, no unexplained terms.
Structure and logical flow (1–10): how coherently the text is built —
is there a clear thread, does the thought get lost, does the text lead the reader from the start to the conclusion.
Rate the text only — ignore voice, gestures, editing.
Give the scores without explanations.
Format: X,X (for example: 8,7)
[SCRIPT TEXT]
Result:
The model returns two numbers, say 6,5. That's a signal to get to work. If you want to know why it scored that way, ask separately: "What exactly is lowering the clarity? Point to specific places." The model will break it down piece by piece. After editing, run it again. If the score climbs to 8,7, the script has become clearer and will most likely hold viewers longer.
Why It Works
An LLM sees what used to be unmeasurable. Old readability formulas (like the Flesch index) counted word and sentence length. But "cognitive friction" isn't about word length. It's about whether the text follows a logic, whether the transitions between thoughts are explained, whether the thread gets lost. An LLM picks up exactly that: semantic coherence and the architecture of the argument. That's why LLM scores predict real audience behavior better than classic metrics.
Cognitive load works against you. When a person has to "decode" a text, filling in transitions and reconstructing the logic, the brain spends energy on decoding instead of content. That's physically unpleasant. Subconsciously it turns into a negative verdict on the source: "not interesting," "boring," "not for me." The author thinks people didn't get it. In reality, they got tired.
Prompt control levers: - Remove "without explanations" → you get a detailed breakdown of each criterion, not just numbers. Useful on the first iteration. - Add a third criterion for your task, for example "Persuasiveness of arguments (1–10)" for a pitch or "Accessibility for beginners (1–10)" for educational content. - Run it 5 times and take the average → the score stabilizes. An LLM varies a bit from request to request; an average across several runs is more reliable. - Compare versions: first draft vs. after edits. The difference in points is an objective measure of progress.
Prompt Template
You are an expert in evaluating the quality of texts and public talks.
Rate the following text on two criteria:
Clarity of delivery (1–10): how easy it is to understand the ideas, explanations and concepts —
no confusion, no unnecessary hedging, no unexplained terms.
Structure and logical flow (1–10): how coherently the text is built —
is there a clear thread, does the thought get lost, does the text lead the reader from the start to the conclusion.
Rate only the content of the text.
Give the scores without explanations.
Format: X,X (for example: 8,7)
{text}
What to put in {text}: a video script, article, pitch, post, email, explanation of a concept: any text where readability and structure matter.
🚀 Quick start: paste into the chat:
Here's a prompt for scoring the clarity of a text. Adapt it to my task: {your task}.
Ask me questions to fill in the fields.
[paste the template above]
The LLM will ask what type of text you're scoring and whether there are specific criteria for your audience, because "clarity for an investor" and "clarity for a teenager" are different things, and the prompt can be tuned more precisely.
Limitations
⚠️ Text only, not delivery: The prompt scores the transcript: what is said and how it's put into words. Intonation, pace, charisma, gestures are out of scope. For video or live talks that's an important blind spot.
⚠️ Relative scale: A 6 doesn't mean "bad in absolute terms." The LLM compares the text against its internal idea of a "good text." What matters more than the absolute number is the trend: how the score changes after edits.
⚠️ Doesn't work on very short texts: A paragraph or a tweet is too little context to judge structure and logical flow. The method shines on texts of 300–400 words and up.
⚠️ The LLM varies from request to request: One run might give 7,6, another 6,8. If you need a stable score, run it 3–5 times and take the average.
How They Studied It
The idea was simple: take a platform with thousands of talks and real engagement metrics, and check what predicts views and likes. A team from Bar-Ilan University and Michigan State University collected 1,239 TED Talks from 2006–2013 and ran each transcript through an LLM 50 times to get a stable average score for clarity and structure. In parallel they pulled like and view data from YouTube.
Then came regression: how much clarity predicts engagement beyond the factors already accounted for: video length, topic, science vs. non-science format, publication date. Clarity gave a significant boost to the model's predictive power: β = .339 for likes and .314 for views. That's a lot: the full model explained 29% of the variance in likes and 22.5% in views.
A curious detail: the effect turned out to be domain-independent. It didn't matter whether it was a physics lecture or a personal story; clarity worked the same everywhere. Another surprise: traditional readability metrics (word length, sentence length) lost to the LLM score, because "cognitive friction" comes not from simple words but from semantic coherence. Finally, the authors tracked TED's evolution: over time, the clarity of talks went up and the spread narrowed as the platform standardized. That indirectly confirms clarity is a trainable skill, not an innate gift.
Adaptations and Extrapolations
1. Iterative polishing instead of a one-off score
🔧 Technique: a "score → edit → rescore" loop → objective progress
Use the prompt not as a final check but as an iteration tool. After each run, ask: "Which three places hurt clarity the most?", fix exactly those, then score again. That way you see concrete gains, not just a feeling that "it got better."
2. Comparing two versions
🔧 Technique: A/B testing on texts → find out which version is objectively clearer
I'll give you two versions of the same text.
Rate each on the same scale: clarity (1–10) and structure (1–10).
Then explain what exactly sets them apart on these criteria.
Version A: {text_1}
Version B: {text_2}
Handy when you've written two versions of a headline, an intro or an explanation and can't decide which to use.
3. Carrying it over to team feedback
🔧 Technique: an anonymous clarity check on other people's texts → concrete feedback without matters of taste
Instead of "your text is unclear," say "I ran it through scoring: clarity 5/10, structure 4/10, here are the specific spots." The LLM makes feedback objective: not an editor's opinion, but a measurement.
Resources
Paper: Computational Analysis of Speech Clarity Predicts Audience Engagement in TED Talks
Authors: Roni Segal, Matan Lary (Bar-Ilan University, Israel), Ralf Schmaelzle (Michigan State University, USA), Yossi Ben-Zion (Bar-Ilan University)
Builds on: cognitive load theory (Sweller, 1988), the concept of processing fluency (Reber et al., 2004), and the AI lecture-scoring methodology from the authors' previous work (Zion et al., 2025)