Opinion

Why AI Speech Coaching Tools Count the Wrong Things

Amy Cuddy used 85 filler words in her TED talk. 72 million views. The entire speech coaching industry is optimizing for the wrong metric.

ReadRoom · March 2026 · 7 min read

The Filler Word Fallacy

Amy Cuddy said "um" or "uh" 85 times during her TED talk. Eighty-five. If you ran that talk through Yoodli, Poised, or any of the other AI speech coaching tools on the market right now, they would flag it as a disaster. A blizzard of red warnings. A confidence score in the basement.

That talk has 72 million views. It is one of the most shared, most cited, most life-changing presentations in the history of public speaking.

Nobody in that audience was counting filler words. They were feeling something.

Ken Robinson: 45 filler words. Hand in his pocket for 80% of the talk. Argued for a complete revolution in global education with the energy of someone chatting at a pub. 75 million views. The most-watched TED talk of all time.

Brené Brown: scored 6 on Confidence in our analysis, but 9 on Engagement and 9 on Impact. Her voice cracks. She admits to calling a therapist. 65 million views.

The data is clear: filler words do not predict impact. Not even close. The correlation between filler count and audience response, based on our analysis of hundreds of hours of speeches, is effectively zero. Some of the most powerful moments in public speaking history are riddled with hesitation, self-correction, and verbal stumbles.

So why does every speech tool on the market treat filler words as the primary metric?

What the Tools Actually Measure (and Why It's Comfortable)

Filler words are easy to count. Pace is easy to clock. Volume is a waveform. These are audio-only metrics. They require no visual understanding, no multimodal reasoning, no comprehension of what a face is doing while a mouth is talking.

That is why every speech tool measures them. Not because they matter most, but because they are cheapest to detect.

Consider the gap between what the tools report and what the audience actually experiences:

What Tools Measure
What Audiences Experience
Filler word count
Emotional resonance
Words per minute
Whether time flew or dragged
Volume variation (dB)
Whether they felt pulled in
Eye contact percentage
Whether they felt seen

These are not the same thing. They are not even measuring the same phenomenon. One column describes acoustic properties of a waveform. The other describes a human emotional response.

You can have perfect metrics in the left column and put an audience to sleep. You can have terrible metrics in the left column and make 72 million people share your talk with everyone they know.

None of these tool metrics capture the thing that actually separates a forgettable presentation from one that changes minds.

The Thing That Actually Matters: Congruence

Congruence is the alignment between what you say, how you say it, and what your body communicates while you say it. It is the single best predictor we have found for whether a speech actually lands.

Brené Brown: Confidence 6, Engagement 9, Congruence 7, Impact 9. When she breaks down talking about calling a therapist, her vulnerability IS the message. Her voice cracks. Her shoulders drop. Her face shows exactly what she is describing. And 65 million people lean in.

That is congruence. Not polish. Not performance. Alignment.

Simon Sinek at minute 17:08. He says, "He gave the I have a dream speech, not the I have a plan speech." For the first time in eighteen minutes, he stops pacing. His voice drops. He stands completely still. Confidence, Engagement, and Congruence all peak simultaneously. His body matches his words. That is what separates his best moment from his weakest.

At his weakest (minute 1:02), Sinek asks the audience to think about Martin Luther King while staring at the floor and adjusting his glasses. The words demand grandeur. The body says he is still warming up. Congruence score: 3. The audience can feel the mismatch, even if they cannot name it.

When your face says one thing and your words say another, audiences feel it. They do not know why. They just tune out.

The Coaching Note
A speaker with 100 filler words and perfect congruence will outperform a speaker with zero filler words and mismatched body language. Every time. The audience forgives hesitation. They cannot forgive inauthenticity.

The 93% Problem

Albert Mehrabian's research on emotional communication is often oversimplified, but the directional finding holds up: the vast majority of how we interpret a speaker's emotional state comes from nonverbal channels. Face, posture, gesture, tone. Not words.

Every major speech coaching tool on the market analyzes only the audio channel. The transcript. The waveform. The words and their acoustic properties.

Zero tools analyze the video. None of them watch the speaker's face. None of them track posture, gestures, or micro-expressions. None of them can tell you that your face went flat during the most important sentence of your pitch.

0
Speech coaching tools that analyze video
Every competitor is audio-only

Here is what the market looks like:

Tool Analysis Type Price
Gong Audio + transcript $100-250/user/mo
Yoodli Audio + AI roleplay $20/mo
Poised Audio widget $19/mo
Speeko Audio exercises $25/mo
Video analysis Face + body + voice Nobody does this

This is not a feature gap. It is a category gap. The entire industry has agreed, implicitly, that speech coaching means audio coaching. That the speaker's body does not matter. That what the audience sees is irrelevant.

That assumption is wrong. And it explains why people spend thousands on coaching, memorize every rule about filler words and pacing, and still cannot figure out why their presentations do not land.

What Changes When You Can Actually See

When you analyze video alongside audio, you find things that a transcript never could. These are the moments that lose deals and bore audiences, and they are completely invisible to every speech tool on the market.

You cannot find any of these in a transcript. You cannot detect them in a waveform. They exist only in the gap between what a person says and what their body reveals.

The Point

Audio tells you what was said. Video tells you what was meant. The audience always responds to what was meant.

This is what multimodal analysis reveals. Not more metrics. Better metrics. The ones that correspond to what a human audience actually experiences when they watch you speak.

Related reading

We Analyzed the 4 Most-Watched TED Talks of All Time. Simon Sinek scored 6/10 on confidence. Amy Cuddy scored 4/10. None scored above 7. Here is what the data actually shows.

Speech Analysis Tools in 2026: What Each One Actually Measures. Gong, Poised, Yoodli, Speeko, ReadRoom. Five tools that sound similar. They measure completely different things.

Amy Cuddy Used 85 Filler Words. 72 Million Views. We analyzed 8 speakers. The one with 2 filler words scored lowest on impact. The one with 85 scored highest. The data destroys a $6 billion assumption.

See What They Can't

Your voice is only part of the story. Upload your video and find out what the audience actually experienced.

Get Your Free Analysis

First analysis free. $19/analysis after beta.