Gmail Help Me Write 2026: New Custom Refine Prompts Explained

For two years, Gmail's AI writing assistant could draft an email in seconds but argued with you the moment you wanted to change it. You got three or four polite buttons — Polish, Formalize, Shorten — and if none of them matched what you actually meant, your only option was to delete the draft and start over. In 2026 that finally changed. Help me write now hands you a prompt bar where you can type an instruction in your own words, and Gemini rewrites the draft on the spot. Tell it "add a deadline of Friday," "make the second paragraph warmer," or "mention the invoice number," and the edit lands instantly — no more wrestling a canned button into doing something it was never built for.

This upgrade rolled out to Gmail through July 2026 and reached most accounts by the third week of the month. It is a small interface change with an outsized effect on how quickly you can go from a rough idea to a sent email. This guide walks the whole feature end to end — what Help me write custom refinement does, how the prompt bar works, exactly how to use it on desktop and mobile, the prompts that get the best results, how undo and redo let you iterate safely, who actually has access, and the privacy and quality limits nobody puts in the marketing copy. Whether you send a handful of careful emails a week or draft outreach at scale, this is the field guide to Gmail's most useful AI update of the year.

What Changed in Gmail Help Me Write in 2026

The headline change is deceptively simple: you can now edit an AI-generated draft by typing a plain-language instruction instead of picking from a fixed menu. When Help me write first shipped, refining a draft meant tapping one of a few preset chips — Polish, Formalize, Shorten, Elaborate, or Rephrase — and hoping the result nudged the text in the direction you wanted. Those presets were useful for broad strokes but useless for anything specific. There was no button for "sound less apologetic," no button for "add the meeting time," no button for "keep it friendly but firm." The 2026 update replaces that ceiling with a prompt bar that accepts any instruction you can phrase.

Google framed the change as moving from rigid options to flexible, natural-language editing, and in practice that is exactly how it feels. The preset buttons are still there for quick, one-tap changes, but underneath them sits a free-text field. Type "add a missing detail to the second line," "include a deadline for the request," or "make this more confident," and Gmail updates the draft immediately. The feature turns Help me write from a one-shot draft generator into an actual editing partner you can iterate with, the way you would give notes to a human assistant.

This matters because the first draft an AI produces is almost never the one you want to send. The value of a writing assistant was always going to live in the revision loop, not the initial guess — and until this update, that loop was broken. You either accepted a generic draft, hammered it with a preset that overshot, or gave up and wrote it yourself. Custom refinement closes that gap and makes the whole feature worth turning on for the first time for a lot of people who tried it once in 2024, found it too blunt, and never came back.

How the Custom Refine Prompt Bar Works

When you generate a draft with Help me write, the result appears in a small panel with a text field at the bottom — the prompt bar. That field is where the new power lives. Instead of only offering you buttons, Gmail invites you to describe the change you want in a sentence. You type the instruction, submit it, and Gemini rewrites the draft in place, keeping everything you did not ask it to touch and altering only the part you specified.

The mechanics are worth understanding because they explain why the results are so much better than the old presets. When you tapped "Formalize," the model rewrote the entire message toward a formal register, often flattening a good sentence you liked in the process. When you type a targeted instruction like "make only the closing line more formal," the model has a precise scope and a precise goal, so it changes less and preserves more. Specificity is the whole trick: the narrower your instruction, the more surgical the edit, and the less likely Help me write is to rewrite something you were happy with.

Each instruction is treated as a follow-up on the current state of the draft, not the original prompt. That means you can stack edits — refine once, look at the result, then refine again based on what you now see. The draft carries forward between instructions, so "add a deadline" followed by "now make it sound less demanding" both apply to the same evolving message. This conversational, layer-by-layer editing is the core of what makes the 2026 version genuinely useful rather than a novelty, and it is the same interaction pattern behind the broader wave of AI features arriving across the inbox, which we cover in our overview of the Gmail AI Inbox.

From Three Preset Buttons to Unlimited Instructions

To see how large the jump is, compare what refining a draft used to require against what it takes now. The old model forced your intent through a handful of fixed doors; the new one lets you say what you mean.

The old presets have not disappeared — Polish, Formalize, and the rest are still one tap away when a broad change is all you need. But they are now the fast lane rather than the only lane. For anything that requires knowing what your email is actually about, the prompt bar wins every time, because a button cannot know that your recipient prefers bullet points or that you forgot to mention the attachment. With Help me write custom refinement, you supply that context in a sentence and let the model act on it.

There is a subtler benefit too. Because you can now describe edits precisely, you can start from a rougher, faster first draft and shape it afterward, rather than trying to get the initial prompt perfect. Many people found the original feature frustrating because a vague first prompt produced a vague draft with no good way to fix it. Now the workflow flips: generate something quick, then steer it with two or three targeted instructions. That is a far more forgiving way to write, and it plays to what large language models are genuinely good at.

How to Use Custom Refinement on Desktop and Mobile

The feature works the same way conceptually across platforms, though the exact taps differ. Here is how to reach the prompt bar on each one.

• Desktop (web): Click Compose, then click the Help me write pencil icon in the toolbar of the new message window. Type a prompt to generate a draft, and when it appears, use the text field beneath it to enter a refine instruction. Submit it and the draft updates in place.
• Android: Open the Gmail app, tap Compose, then tap the Help me write icon (the pencil with a spark) inside the draft. Generate your text, then type your custom instruction into the refine field that appears below the draft and tap to apply it.
• iPhone and iPad: Tap Compose, tap the Help me write icon, generate a draft, and use the prompt bar at the bottom of the assistant panel to type your refinement. The edit applies to the current draft without discarding it.

A practical tip: give one instruction at a time. It is tempting to type "make it shorter, add a deadline, and change the greeting" in a single go, but the model handles a single clear instruction far more reliably than a stacked list. Refine, read the result, then give the next note. This one-change-at-a-time discipline is the single biggest difference between people who find Help me write magical and people who find it frustrating. If you feed it a paragraph of mixed demands, you get a muddled rewrite; if you feed it one crisp instruction, you get a clean edit.

Another habit worth forming: read every draft before you send it. The prompt bar makes editing so fast that it is easy to fire off a message you only skimmed. AI-generated text is fluent and confident even when it is subtly wrong about a name, a date, or a commitment you never actually made. Treat the output as a strong first draft that you own and are responsible for, not as a finished email the machine signed for you.

12 Custom Refine Prompts That Actually Work

The quality of a refine depends almost entirely on the instruction you give. Vague prompts produce vague edits; specific prompts produce specific ones. Here are twelve instructions that consistently get good results, grouped by what they fix.

• Tone: "Make this friendlier without sounding unprofessional."
• Tone: "Sound more confident and less apologetic — remove the hedging."
• Length: "Cut this to three sentences and keep only the ask."
• Length: "Expand the second paragraph with one concrete example."
• Detail: "Add that the deadline is end of day Friday."
• Detail: "Mention that I've attached the signed contract."
• Structure: "Turn the middle section into a short bulleted list."
• Structure: "Move the request to the top so it isn't buried."
• Context: "This is a reply to a complaint — make it more empathetic."
• Context: "Write this as a follow-up, since I already emailed once."
• Clarity: "Replace the jargon with plain language."
• Clarity: "Make the subject line clearer and more specific."

Notice that every one of these names both a target (tone, a paragraph, the deadline) and a direction (friendlier, shorter, empathetic). That combination is what gives the model enough to work with. When a refine goes wrong, it is almost always because the instruction was missing one of those two halves — "make it better" tells the model nothing, while "make the closing line more decisive" tells it exactly what to change and how. Keep a few of these phrasings in mind and Help me write becomes dramatically more predictable.

You can also use the prompt bar to correct the model's assumptions rather than just polish its prose. If the AI drafted a cold introduction but you are actually replying to someone you know well, a single instruction like "we're already friends, make it casual" resets the entire register. This is where custom refinement earns its keep: it lets you fix the model's misread of the situation in one sentence, instead of deleting a draft that got the facts of the relationship wrong.

Undo, Redo, and Iterating Without Losing Your Draft

One of the quieter but most important parts of the 2026 update is that you can undo and redo refinements. Every edit you apply is reversible, so experimenting with an instruction no longer risks destroying a version you liked. If you ask Help me write to shorten a draft and it cuts a line you wanted to keep, you can step back to the previous state and try a more precise instruction instead. This safety net is what makes aggressive iteration practical.

The workflow this enables looks like real editing. You generate a draft, refine it, and if the result overshoots, you undo and rephrase your instruction. Because the draft state is preserved across steps, you are never gambling your whole message on a single risky prompt. In the old preset-only world, hitting the wrong button could flatten a draft you had already massaged into shape, and there was no clean way back. Undo and redo remove that fear entirely, which in turn makes people bolder about asking for exactly what they want.

In practice, the most effective pattern is a short loop: refine, evaluate, and either accept, undo, or refine again. Two or three passes usually gets a draft to "good enough to send." If you find yourself on the sixth or seventh instruction, that is a sign the initial prompt was too vague or the email is genuinely complex enough that you should write the tricky part yourself. The tool is a fast path to a solid draft, not a substitute for knowing what you want to say — and the undo history is there precisely so you can explore without consequences.

Who Gets It: Rollout, Editions, and Admin Controls

Custom refinement in Help me write began rolling out in July 2026 across both Rapid Release and Scheduled Release domains, with the rollout expected to complete by around July 20, 2026. That means the feature reached most eligible accounts within a couple of weeks of the announcement, which is why a lot of people noticed the prompt bar appear without warning in mid-to-late July.

Access depends on having Gemini in Gmail available on your account. For Google Workspace users, the feature is on by default when two conditions are met: Gemini for Workspace in Gmail is enabled, and Workspace Intelligence access to Gmail is enabled by the administrator. When both are in place, custom refinement is available to end users automatically — there is no separate switch each person has to flip. Here is who gets it and how:

For administrators, the controls live in the Google Workspace Admin console under the Gemini and Workspace Intelligence settings. If your organization has decided not to expose generative AI in Gmail, disabling those settings removes Help me write and its custom refinement along with them. Conversely, if your team wants the feature but does not see it, the fix is usually on the admin side rather than the individual one. If you would rather keep AI out of your inbox entirely, our guide on how to turn off Gmail AI covers the opt-out paths in detail.

Limits, Privacy, and What the AI Still Gets Wrong

Custom refinement is a real improvement, but it is not magic, and treating it as infallible is how people send embarrassing emails. The model still hallucinates details when you ask it to "add" something it has no way of knowing — invent a phone number, guess a date, or state a commitment you never made. The prompt bar will happily insert a plausible-sounding fact that is completely wrong. Every instruction that adds content is an instruction you must verify before sending, because Help me write optimizes for fluent, confident prose, not for accuracy.

There are privacy considerations too. To refine your draft, the feature sends the text you are working on to Google's Gemini models for processing. Within Workspace, that processing is governed by your organization's data protections and enterprise commitments, but it still means your draft content is being handled by an AI system rather than staying purely on your device. For genuinely sensitive material — legal language, confidential figures, anything under compliance rules — think about whether you want it passing through an AI editor at all. The convenience is real, but so is the fact that the content leaves your compose window.

It is also worth remembering that AI features reading and rewriting inbox content have opened new attack surfaces that did not exist before. As assistants parse, summarize, and edit email, malicious instructions hidden inside messages can attempt to hijack what the AI does — a risk we examine in our analysis of the Gmail Gemini prompt injection attack. Custom refinement operates on your own draft rather than on incoming mail, which limits that specific exposure, but the broader lesson holds: the more the AI touches your email, the more carefully you should read what it produces.

How to Turn Help Me Write On or Off

Not everyone wants an AI editor one tap from every compose window, and Gmail makes it possible to dial the feature up or down. If you have Gemini access and the prompt bar simply is not appearing, the most common causes are a mobile app that needs updating, an admin who has not enabled Workspace Intelligence, or a personal account that lacks Gemini in the first place. Updating the Gmail app and confirming your Gemini status resolves the majority of "it's missing" complaints.

To reduce how present the feature is, you have a few levers:

• Just ignore it: The Help me write icon only acts when you tap it. If you never invoke it, it never drafts or refines anything — there is no automatic rewriting of what you type by hand.
• Turn off smart features: Gmail's smart and AI features can be scaled back in Settings under the smart features and personalization controls, which affects the broader suite of assistants.
• Admin-level disable: On Workspace, an administrator can disable Gemini in Gmail entirely, which removes Help me write and custom refinement for the whole organization.

If your goal is a cleaner, less AI-heavy inbox overall rather than just removing one button, the settings that govern the wider redesign are worth a look — our walkthrough of the Gmail AI Inbox explains which toggles control which features, so you can keep the assistants you like and mute the ones you do not. Custom refinement is opt-in by nature: it does nothing until you deliberately ask it to, which makes it one of the easier AI features to live alongside even if you are skeptical of the trend.

Why the Account Behind Your AI Draft Still Matters

Here is the part the feature announcements never mention: a beautifully refined email is worthless if it lands in the spam folder. Help me write can make your message clearer, warmer, and more persuasive, but it has no effect whatsoever on whether Gmail's filters trust the account sending it. Deliverability is decided by sender reputation, authentication, and account history — none of which an AI editor touches. You can polish a draft to perfection and still watch it vanish into spam if the inbox behind it is brand new, unwarmed, or flagged.

This is why serious senders separate the writing problem from the sending problem. The words are only half the equation; the account is the other half. An established, phone-verified inbox with a genuine sending history carries the trust signals filters reward, so a well-written message is judged on its content rather than dismissed on a suspiciously fresh sender profile. If you are scaling outreach — where every email might be AI-assisted and volume is high — the reputation of your sending accounts determines whether any of that careful drafting ever reaches a human. Our aged, phone-verified Gmail accounts are built for exactly this: inboxes that land, so the effort you put into the message is not wasted at the door.

The broader principle carries across every corner of email in 2026. AI tools like custom refinement make the visible work — writing — faster and easier, which only raises the importance of the invisible work: deliverability, authentication, and account trust. A great draft from a distrusted account still fails. A solid draft from a well-aged, properly maintained inbox arrives, gets read, and gets a reply. Help me write handles the first mile; the account behind it handles the last one — and both have to work for the email to matter.

Frequently Asked Questions

What is the new custom refinement feature in Gmail Help me write?

It is an upgrade that lets you edit an AI-generated draft by typing an instruction in your own words, instead of choosing from fixed preset buttons like Polish, Formalize, and Shorten. You type a follow-up prompt such as "add a deadline" or "make the second line friendlier," and Gemini rewrites the draft in place, changing only what you asked for. The feature rolled out through July 2026 and completed for most eligible accounts by around July 20, turning Help me write from a one-shot draft generator into an interactive editing partner.

Do I still have the old Polish, Formalize, and Shorten buttons?

Yes. The preset chips remain available for quick, one-tap changes when a broad adjustment is all you need. The custom prompt bar is added underneath them, not in place of them, so you can tap a preset for speed or type a specific instruction for precision. In practice the presets become the fast lane for broad edits, while the prompt bar handles anything that requires knowing what your email is actually about — such as adding a detail, changing a single line, or fixing the tone of one paragraph.

Can I undo a refinement if the AI changes something I wanted to keep?

Yes. The 2026 update supports undo and redo for refinements, so every edit is reversible. If an instruction overshoots — for example, shortening a draft cuts a line you liked — you can step back to the previous version and try a more precise prompt. Because the draft state is preserved across each instruction, you can experiment freely without risking the version you already have. This is what makes iterative editing practical: refine, evaluate, and either accept, undo, or refine again in a short loop until the draft is right.

Is Gmail Help me write custom refinement free, and who can use it?

Access depends on having Gemini in Gmail available on your account. For Google Workspace users, custom refinement is on by default when Gemini for Workspace in Gmail and Workspace Intelligence access are enabled by the administrator. On personal Gmail accounts, it appears wherever Help me write is offered as part of your Gemini access. If Gemini is disabled by an admin, or your personal account does not include it, the feature is not available. When it is missing despite having access, updating the Gmail app or confirming your admin settings usually resolves it.

Does using Help me write affect whether my email lands in the inbox?

No. Custom refinement improves the wording of your message, but it has no effect on deliverability. Whether an email reaches the inbox or the spam folder is determined by sender reputation, authentication, and account history — factors an AI editor does not touch. A perfectly refined message sent from a new or low-trust account can still be filtered as spam. To make sure your writing actually gets read, pair good drafting with an established, warmed sending account that carries the trust signals Gmail's filters reward.

Ready to make sure the account behind your carefully written email is built to land in the inbox — not the spam folder? Whether you are protecting a primary account or scaling AI-assisted outreach that needs real deliverability, explore our aged, phone-verified Gmail accounts made for exactly that. Have a question about Help me write, Gemini, or getting your mail delivered? Message us any time on Telegram at @mixgmail — we answer real questions from real users every day.