How to Make Text-to-Speech Read Only the Article

How to Make Text-to-Speech Read Only the Article

If your read-aloud tool announces the cookie banner, the navigation menu, and nine related-post headlines before it reaches the first sentence, the problem is what the tool is collecting, not the voice. There are five ways to fix it, and they escalate: read only the text you select, play a single block you point at, switch to main-content extraction, hide one offending element for good, or save a per-site skip list you never think about again. Start at the top of that list — most noisy pages are solved in one step.

Why does text-to-speech read menus, ads, and comments?

Because a web page has no reliable marker that says "the article is here."

A read-aloud tool has to guess. The simplest tools take everything textual in the page and queue it in document order, which is why you hear the skip-link, the nav, the newsletter box, the byline, the share buttons, and then finally the article. Smarter tools try to identify the main content region, but that guess can miss on pages that split the body across containers, load paragraphs after the initial render, or wrap the article in the same generic markup used for the sidebar.

Two things reliably make the guess worse:

• Layout-heavy pages. News homepages, documentation with a large navigation tree, forums, and dashboards have more non-article text than article text.
• Content that appears late. Comments, lazy-loaded sections, and single-page-app route changes can arrive after the tool has already built its queue.

Knowing this changes the fix. You are not looking for a better voice. You are looking for control over the queue.

This is a different problem from a *visually* cluttered page. If the page looks noisy while you read, a focused reading overlay helps — see listening to web pages without distractions. This post is about what ends up in the audio.

Fix 1: Read only the text you select

The fastest fix needs no configuration. Highlight the paragraphs you want, right-click, and choose the read-selection action. In Flow TTS that menu item is Read Selection.

This is the right tool when:

• You want one section of a long page
• The page structure is unusual and extraction is unlikely to work
• You are checking a single quote, abstract, or summary

The tradeoff is that it is manual every time. For a page you read once, that is fine. For a site you open daily, keep reading.

Fix 2: Play one block by pointing at it

A step up in convenience: turn on hover-to-play, and a play button appears near readable text blocks as you move the mouse over them. Click it and only that block is read.

In Flow TTS this is Show play button when hovering over text in Settings → Appearance. There is a nested option, Include more page text (sidebars, links, extra regions), which is off by default — leave it off if your goal is less noise, since it deliberately widens what counts as a hover target.

Hover-to-play is the best fit for scanning. You skim a documentation page, find the one paragraph that matters, and hear it without selecting anything or starting a full-page session.

Fix 3: Use main-content extraction instead of the whole page

If your tool offers a mode that extracts readable main content, use it as the default. In Flow TTS that is Auto mode, and it collects readable main content from the current page rather than everything on it.

Be honest about what this does. Auto mode is a large improvement over reading the entire page, but it is not a promise that every layout will be handled perfectly. On a clean article it usually gets you the body and little else. On a dense homepage or an unusual template, some junk can survive and some real text can be missed.

For most people, Auto mode plus one of the fixes below covers nearly everything they read.

Fix 4: Hide one specific element for good

This is the fix most people never find, and it is the one that solves the recurring annoyance.

When a particular element keeps getting read — a site's newsletter box, a persistent banner, a "trending now" rail — you do not have to accept it or configure a whole reading plan. Right-click that element and choose Hide this element when extracting. Flow TTS opens the Flow Templates panel with the element you right-clicked already seeded as a skip target, so you confirm and save rather than hunting for it in a picker.

The important part is what happens next. A skip can be applied back to Auto mode through the Also apply in Auto mode option, which removes skipped items from the extract and adds picked items that Auto missed. In other words, you are not switching to a different mode to get the benefit — you are teaching your normal mode to stop reading that thing.

Do this two or three times on the sites you actually use and the daily noise mostly disappears.

Fix 5: Save a reading path for sites you revisit

For pages where you want positive control rather than exclusions, save a reading plan. Flow TTS calls these Flow Templates: you pick the parts of the page that should be read, mark others as skipped, and reuse that plan when you return.

The controls worth knowing:

• Read, Skip, and Repeat. Repeat matches similar elements, which is how you cover every card in a repeated list with one pick.
• Read methods. Normal reads the element's text; Dynamic extract re-scans a region on each run, which suits comments and feeds; Read everything inside takes all text in the selection.
• Live-root picks. Select a parent region instead of clicking every paragraph.
• URL scoping. Apply a plan to a whole domain, one page, or URLs containing a substring, so a site's article template and its homepage can behave differently.

Reach for this on the handful of sites you open constantly. The full walkthrough is in how Flow Templates create custom reading paths.

Which fix should you use?

• A one-off page where you want one section: select the text and use Read Selection.
• Scanning for the single relevant paragraph: hover-to-play.
• A normal article you will read once: Auto mode on its own.
• One element that keeps getting read on a site: hide it from extraction.
• A site you open every day: save a Flow Template.

Work down the list only as far as the problem requires. Building a reading plan for a page you will never open again is wasted effort.

What still goes wrong

Control over the queue does not solve everything.

• Redesigns break saved selectors. A plan built on page structure needs updating when a site rebuilds its markup. Edit the template or re-pick the targets.
• Some pages resist extraction. Complex PDFs, paywalled articles, canvas-rendered text, and heavily dynamic pages may need manual selection or another format entirely.
• Tables, equations, and code want eyes. Getting them out of the audio queue is usually the better outcome, not getting them read correctly.
• Late-arriving content can miss the queue. Sections that load after playback begins may not be included.

If a name or acronym is read incorrectly rather than the wrong text being read, that is a separate fix — see how to fix text-to-speech pronunciation.

Where Flow TTS fits

Controlling what gets read is the reason Flow TTS exists rather than a feature bolted onto a voice. All five fixes above are built in: Read Selection from the right-click menu, hover-to-play from Settings → Appearance, Auto mode for main-content extraction, "Hide this element when extracting" for one-off offenders, and Flow Templates with Auto-mode patching for sites you revisit. The in-page player adds pause, skip, speed, and an optional node scrubber for moving between readable nodes when that preference is enabled.

Speech synthesis runs in the browser for supported local models, so page text used for audio does not need to go to a cloud TTS API. Accounts, licensing, usage, and model downloads still use network services, which is why it is not a fully offline or zero-cloud tool — the on-device explainer covers that boundary. Playback requires an account set up with an emailed one-time code. Free usage is 10,000 characters per day, and premium unlocks unlimited characters and Glass Mode.

New to listening in the browser? Start with how to listen to web pages in Chrome, then Getting Started with Flow TTS for the first session.

FAQ

How do I make text-to-speech read only the article?

Use main-content extraction rather than whole-page reading, then remove whatever survives. In Flow TTS that means Auto mode, plus right-clicking any element that keeps getting read and choosing "Hide this element when extracting."

Why does my read-aloud extension read the navigation menu?

Because it is queueing all text on the page instead of identifying the article, or its main-content detection failed on that layout. Switch to an extraction mode and add a skip for the offending element.

Can I stop text-to-speech from reading comments?

Yes. Mark the comments container as a skip and apply it in your normal reading mode. For comment sections that load new content, a region pick with Dynamic extract handles the changing contents.

Can I make a tool read only part of a page every time?

Yes, with a saved reading plan. Flow Templates store Read and Skip targets per domain with URL scoping, so the same page type behaves the same way on every visit.

Is reading a selection the same as reading the article?

No. A selection is a manual, one-time choice. Extraction and saved plans are repeatable. Use a selection for pages you read once, and a plan for sites you open often.

How do I try this in Chrome?

Install Flow TTS from the Chrome Web Store, finish account setup, run Auto mode on the page that annoys you most, then right-click the first thing you did not want to hear and hide it from extraction.