Google Made the Best Text-to-Speech Voices on the Planet. Then Didn't Put One in Gemini.

This is the part that gets me. Google has WaveNet. Google has Cloud Text-to-Speech with over a hundred neural voices in forty languages. Google made the Pixel phones that read your screen aloud with a voice so natural that my mom thought it was a person calling her. Google has spent billions on speech synthesis technology. And then Google built Gemini — their flagship AI — and shipped it without a read-aloud button on the web.

You get a text response. On screen. That's it. Read it with your eyeballs like it's 1997.

I asked Gemini to explain how the EU AI Act would affect my startup's chatbot deployment. It wrote 1,600 words covering risk classification tiers, transparency requirements, high-risk system obligations, and the enforcement timeline through 2027. Dense, specific, genuinely useful. I read about a third of it and then alt-tabbed to Slack because my brain hit a wall. The information was there. My capacity to absorb it through a screen at 3pm on a Wednesday was not.

There's a particular irony in Google having the most advanced voice technology on earth and not using it on their own AI product. The mobile app has a voice conversation mode, sure — same as ChatGPT's. But if you're using Gemini in a browser and you have a 1,200-word written response in front of you, there is no button that says "read this to me." You're on your own.

CastReader is the button Google didn't build. It's a free Chrome extension. Install it, open Gemini, and every AI response gets a small speaker icon. Click it. The response reads aloud in a natural AI voice while each paragraph highlights on the page. A floating player appears at the bottom — pause, resume, speed control, paragraph skip. The whole thing takes one click and zero configuration.

I want to be specific about why Gemini benefits from this. Gemini's responses have a particular structure that makes them harder to read than, say, a blog post. They use nested formatting — headers within the response, inline bold text, code snippets, sometimes even tables. Your eyes have to constantly switch between reading modes: prose, then a code block, then a list, then more prose. It's cognitively expensive. Listening to it linearizes the experience. One stream of information, one modality, no visual switching. My retention of Gemini's EU AI Act explanation went from "I think there were four risk tiers?" (reading) to actually being able to explain the timeline at a team meeting the next day (listening on my walk home).

The technical reason this works smoothly on Gemini is that CastReader has a specialized extractor built specifically for Gemini's page structure. Gemini's DOM is not a standard web page. It's a Material Design application with dynamically rendered conversation threads, response drafts, image generation cards, Google Search grounding citations, feedback buttons, and model selector chips all woven into the same container. A generic screen reader or text-to-speech tool would grab everything — the "Show drafts" button text, the "Google it" suggestion chips, the timestamp, the model name. CastReader's Gemini extractor identifies which elements are the actual AI response content and which are interface chrome. It pulls just the answer text, preserves paragraph structure, and sends clean content to the voice engine.

I use Gemini a lot for research because of the Google Search grounding feature. Ask Gemini about a current event or a factual question and it often pulls in real-time search results, citing sources inline. These responses tend to be information-dense — multiple facts, dates, statistics, all packed into flowing paragraphs. Reading this on screen is like reading a Wikipedia article that someone compressed by 40%. Listening is the decompression. My brain processes each fact sequentially, at the pace of the voice, without the temptation to skip ahead.

My girlfriend uses Gemini for recipe ideas. She asks for dinner suggestions based on what's in the fridge and Gemini gives her four or five detailed recipes. Each one is maybe 200 words — ingredients, steps, timing. She used to scroll through them on her phone while standing in the kitchen. Now she clicks the CastReader button and listens while she preps. "It's like having a sous chef who actually explains the reasoning," she told me, which is probably the nicest thing she's ever said about a Chrome extension.

The voice quality is the part that makes or breaks this. I tried using Chrome's built-in speech synthesis on Gemini responses once. The voice read "sauté" as "sawtay" and "mise en place" as "mize en place" and I turned it off before it finished the first paragraph. CastReader uses neural AI voice that handles emphasis, pauses, and pronunciation the way a human reader would. Not perfect — it still occasionally stumbles on highly technical terms — but the difference between it and browser TTS is the difference between listening to a podcast and listening to an automated phone tree.

Here's the setup. Go to the Chrome Web Store. Install CastReader. No signup. No email. No credit card. It is free and I mean that in the actual sense — not free-for-500-characters, not free-with-ads, not free-for-7-days. Free. Open Gemini, ask a question, wait for the response, click the speaker icon on the response. That's it.

It works on ChatGPT and DeepSeek and Claude and Kimi too. Same specialized extraction, same one-click flow. Each platform has its own extractor because each platform has its own DOM structure and its own quirks. You don't configure any of this. CastReader detects the site and picks the right extraction strategy automatically.

Google will probably add a read-aloud button to Gemini eventually. They have all the technology. They just haven't done it yet. In the meantime, CastReader exists, it's free, and it turns Gemini from something you read into something you listen to. That 1,600-word EU AI Act explanation? Seven minutes of listening on my walk home. Every word absorbed. Zero eye strain. The irony of needing a third-party extension to add voice to a Google product continues to amuse me, but at least the solution works.