How to Remove Unwanted Bold Text from AI Drafts in 2026

You’ve just pasted a perfectly drafted response from your AI assistant into Outlook, and suddenly your email looks like a ransom note. The font sizes are mismatched, the background has a subtle gray hue that only shows up on mobile, and—worst of all—every third word is bolded, making you sound like you’re shouting random nouns at your client.

We have arrived in 2026, and while Large Language Models (LLMs) can now write poetry in the style of 17th-century pirates, they still haven't figured out that professional emails shouldn't look like aggressive marketing flyers. If you spend your day managing communication, you know that ai drafts are only 80% of the work; the remaining 20% is fighting the formatting cleanup battle to make the text look human.

The Technical Reason AI Loves Bold Text

To understand how to remove bold text effectively, you have to understand why it’s there in the first place. It isn't a stylistic choice made by the AI because it thinks your email needs pizzazz. It is a byproduct of Markdown rendering.

Most AI models generate text in Markdown. In Markdown, wrapping a word in double asterisks (**word**) renders it as bold. Because LLMs are trained on vast datasets of technical documentation, instructional guides, and web content where skimming is prioritized, they are statistically biased toward emphasizing key entities—dates, names, prices, and action items.

When you copy that text from a browser interface (like ChatGPT or Claude), you aren't just copying characters. You are copying a complex HTML Document Object Model (DOM) tree.

The "Paste as Plain Text" Fallacy

The most common advice for formatting cleanup is to use "Paste as Plain Text" (usually Ctrl+Shift+V or Cmd+Shift+V). If you are writing a three-sentence email, this works. If you are sending a project update, this is a productivity killer.

When you paste as plain text, you nuclear-blast the formatting. Yes, the unwanted bold text is gone. But so are your hyperlinks. Your bulleted lists turn into sad little hyphens or asterisks that don't indent correctly. Your paragraph breaks might double up or disappear entirely.

You are then forced to manually re-link every URL and re-bullet every list. In 2026, spending five minutes re-formatting an email that took the AI ten seconds to write defeats the purpose of using the AI.

Why Outlook is the Final Boss of Formatting

If you use Gmail or Apple Mail, you might experience minor font discrepancies. If you use Outlook (especially the desktop app on Windows), you are dealing with a rendering engine that hasn't fundamentally changed its philosophy in two decades.

Outlook uses the Microsoft Word rendering engine to display HTML emails. This is distinct from a web browser engine like WebKit (Safari) or Blink (Chrome). The Word engine is notoriously hostile to modern CSS.

When an AI draft includes bold text, it often wraps it in a <span> tag. Outlook looks at that tag and frequently interprets it as a distinct "run" of text with its own set of properties. This is why you often see the "Frankenstein Font" effect:

1. The unbolded text adopts your default email font (e.g., Aptos or Calibri, 11pt).
2. The bolded text retains the font family from the browser (e.g., Söhne or Roboto, 12pt).

You end up with a sentence that looks disjointed because the bolded words are physically taller and shaped differently than the rest of the sentence.

The Hidden "Ghost Box" Background

A secondary issue with ai drafts that use heavy bolding is the background attribute. Often, the CSS attached to the bolded text includes a background color definition.

In Light Mode, background-color: #ffffff (pure white) looks invisible. However, when your recipient opens that email in Dark Mode on their phone, the email client inverts the text color but often respects the background color.

The result? Your recipient sees white text on a black background, except for the bolded words, which appear as black text inside jarring white boxes. It looks unprofessional and broken. To achieve clean writing 2026 standards, you must strip the background attributes associated with the bold tags, not just the weight itself.

Counterintuitive Fixes That Make It Worse

In an attempt to fix these issues quickly, many users try shortcuts that actually compound the problem.

1. The "Clear Formatting" Button

Most rich text editors have a "Tx" button to clear formatting. In Outlook, this is the eraser icon.

• The problem: It is too aggressive. It often reverts the font to a system default (like Times New Roman) rather than your set signature font. Furthermore, it wipes out hyperlinks in some versions of Outlook Web, treating the anchor tag <a> as a format to be cleared.

2. The Format Painter

You highlight a "clean" word, click the paintbrush, and drag it over the bold word.

• The problem: This applies the new style on top of the old style. In the HTML code, you are nesting spans (<span class="clean"><span class="dirty">Text</span></span>). Visually, it might look okay on your screen. But because you've increased the code complexity, the email is more likely to break when sent to a different client (e.g., from Outlook to Gmail).

How to Remove Bold Text Without Breaking Structure

If you want to maintain your sanity, you need a workflow that removes the styling (bold, italics, background colors) but preserves the structure (links, lists, headers).

Method 1: The "TextEdit" Intermediary (Mac Only)

This is a manual workaround for Apple users that offers more control than plain text pasting.

1. Open TextEdit.
2. Set the format to "Rich Text" (not Plain Text).
3. Paste your AI draft.
4. Select all (Cmd+A).
5. Use Cmd+T to show fonts, and manually set the weight to "Regular."
6. Copy and paste into your email.

• Verdict: Reliable but slow. It keeps links active but strips the bold weight effectively.

Method 2: The Browser Console Injection (Advanced)

For those comfortable with tech, you can scrub text right in the browser before copying it.

1. Right-click the AI response and select "Inspect."
2. In the Console, you can run a script to replace <strong> tags with their inner text, but this is overkill for daily emails.

Method 3: Intelligent Parsing (The PasteClean Way)

This is where specialized tools come in. To truly clean the text, you need a parser that acts as a filter between the AI and the email client.

The logic works like this:

1. Ingest HTML: Accept the messy DOM tree from the AI.
2. Whitelist Tags: Keep <a>, <ul>, <ol>, <li>, <br>, and <p>.
3. Blacklist Attributes: Strip all style="...", class="...", and id="..." attributes. This removes the specific font-family definitions that cause the Outlook mismatch.
4. Flatten Emphasis: specific to the goal of removing bold, the parser identifies <strong>, <b>, and span style="font-weight:..." tags and unwraps them, leaving the plain text content in place.

A Concrete Before and After

Let's look at a real-world example of how this impacts readability.

The Raw AI Draft (Pasted directly):

Critique: It feels frantic. The bolding implies that the reader is too stupid to know that "spreadsheet" and "Friday" are important nouns.

The Cleaned Version:

Critique: This reads like a professional human wrote it. The emphasis is conveyed through the sentence structure, not visual noise.

Strategic Bolding: When to Add it Back

Once you have stripped the AI's algorithmic bolding, you should manually add emphasis back in—but sparingly. Clean writing 2026 isn't about the absence of formatting; it's about intentional formatting.

Use bold text only for:

• Deadlines: "Please submit by Tuesday at 5 PM."
• Monetary Figures: " The total cost is $4,500."
• Key Actions: "You must approve the invoice before we proceed."

Do not bold:

• Product names (unless it's the first mention).
• Jargon.
• The names of people involved.

Prompt Engineering to Prevent Bolding

Can you just ask the AI to stop? Sometimes.

As of 2026, models have gotten better at following negative constraints, but they still hallucinate formatting because their training data is so heavily weighted toward Markdown.

If you want to try to fix this at the source, append this to your system instructions or prompt:

• "Output plain text only. Do not use bolding, italics, or markdown headers. Keep bullet points and hyperlinks."

However, in my testing, this has a failure rate of about 30%. The model will often slip back into bolding proper nouns or numbered lists. It is usually faster to let the AI generate its natural output and then clean it, rather than fighting the model's inherent token prediction patterns.

The Checklist for Perfect Email Formatting

Before you hit send on that AI-assisted email, run through this mental checklist:

1. Uniform Font: Highlight the whole paragraph. Does the font name in your toolbar go blank? That means you have mixed fonts (e.g., Arial and Roboto). Select the whole text and explicitly click your desired font.
2. Link Check: Hover over your hyperlinks. Did the cleaning process break them? (If you used PasteClean, they are safe. If you used Notepad, they are gone).
3. Color Reset: If you pasted from a dark mode interface, select all text and click the "Automatic" color swatch to ensure you aren't sending white text to a white-background user.
4. The Squint Test: Squint at your screen. Does the text look like a uniform block of gray, or are there weird dark patches? If you see patches, you still have unwanted bold weighting.

We are likely years away from a universal clipboard standard that handles HTML-to-RTF conversion perfectly across all platforms. Until Microsoft rewrites the Outlook rendering engine (don't hold your breath) or AI models stop obsessing over Markdown, you need a dedicated workflow to sanitize your text.

Don't let your tools dictate your tone. Strip the bold, keep the links, and send emails that look like