Plain Text vs HTML Email: Which Passes 4 Filters?

Every founder who has sent a cold email has asked the same question at 11pm: should this be plain text or a designed HTML email? The honest answer is that you are asking the wrong question, because “spam filter” is not one gate. It is at least four, running in sequence, checking different things, and a message can pass three of them and still die on the fourth.

What we set out to change

The plain-text-vs-HTML debate online is a mess of contradictory advice, and the reason is not that anyone is lying. It’s that different sources are measuring different filters. One well-cited comparison shows plain-text cold emails outperforming designed HTML by roughly 15–25% in reply rate in aggregate sends, while a separate finding from the same space reports plain text emails being marked spam more often than “high-quality” HTML in certain samples. Both can be true at once, because “reply rate” and “marked as spam” are outcomes of different filter layers, and “designed HTML” and “quality HTML” are not the same artifact as the generic marketing template most tools default to.

The deeper pattern: authentication is now table stakes, not a nice-to-have. As of February 2024, Gmail and Yahoo require senders of 5,000+ daily messages to authenticate with SPF, DKIM, and DMARC, support one-click unsubscribe, and keep their reported spam complaint rate under 0.3%, with enforcement escalating further from November 2025 onward. A perfectly worded plain-text email from an unauthenticated domain fails before a human or a classifier ever reads a word of it.

The third pattern is that filters disagree with each other. Global average inbox placement across all providers sat at 87.25% in 2025, meaning roughly one in eight legitimate commercial emails never reaches an inbox at all, but that average hides a wide spread: Gmail placed 89.8% of mail in the inbox while Outlook placed only 75.6%. The same email, the same list, the same sender: different outcomes depending purely on which mailbox provider’s filter stack it hit.

The idea / how Raisy fixes it

Raisy treats “will this land” as a pipeline question, not a copywriting question, because that is what the evidence says it actually is. Three principles follow.

Name the filter you’re optimizing for, because there isn’t just one. A message is evaluated at the network/gateway level (is this connection and sender IP trustworthy at all), at the mailbox-provider level (does this look like wanted mail to a model trained on billions of messages), at the inbox-tab-classification level (is this “Primary” or “Promotions”), and at the client level (has this specific recipient’s own rules or training flagged senders like you before). Optimizing format for one layer can actively hurt you at another, heavy HTML can read as more “legitimate marketing” to an automated bulk-sender classifier while simultaneously reading as less personal to the human deciding whether to reply.

Authentication is the floor, not a layer to skip past. Every send Raisy handles for a founder passes a compliance and authentication check before it queues, the same principle as the human-approval gate applied to the sending pipe itself: nothing goes out that would trip the first filter on arrival.

Format follows intent, not habit. A first-touch investor email should read like a person wrote it, plain text or lightly formatted multipart, because that is both what research on cold outreach supports and what recipients say they want (two-thirds of recipients in one industry survey said they don’t mind AI assistance in a cold email as long as it still reads human; the same logic extends to formatting). A newsletter or product update is a different intent with a different filter risk, and multipart MIME (a plain-text version bundled with the HTML one) is the standard middle path there.

How it works (the technical layer)

• Network / gateway filtering. Before content is even parsed, the receiving mail server checks the connecting IP’s reputation, reverse DNS, and rate of connection , this is where a brand-new sending domain with no history gets throttled or greylisted regardless of what the email says.
• Authentication. SPF, DKIM, and DMARC are checked next. Gmail’s own guidelines set the bulk-sender bar explicitly: authenticate, keep spam complaints under 0.3% (target under 0.1%), and support one-click unsubscribe, miss any of these and the message can be rejected outright, independent of format.
• Mailbox-provider machine-learning classification. Gmail alone runs its spam classifiers, built on TensorFlow, across on the order of 15 billion messages a day, weighing domain and IP reputation, authentication consistency, and, critically: recipient engagement signals (opens, replies, deletes-without-reading, spam reports) that Gmail feeds back into the model. This is the layer where format matters most, and where it cuts both ways: a tracking pixel is itself a request to an external server, and dropping it entirely correlates with meaningfully higher reply rates in aggregate data (7.4% vs 4.4%, a 68% relative lift, across a 31-million-email 2025 sample), though a pixel is a weak signal on its own and the effect is measured at the sequence level, not proven as a single- message causal mechanism.
• Content and structure heuristics. Independent of the ML layer, rule-based systems like the open-source SpamAssassin explicitly score image-to-text ratio and image-only bodies, its published rule family fires around an 0.8 image-to-body-area ratio and scales up for image-heavy, low-text messages. This is a legacy from spammers historically hiding spam-trigger words inside images to dodge text scanners; it now penalizes any heavily designed HTML email that leans on graphics over text, format aside from intent.
• Client-side rules and personal filtering. The last gate is local: Outlook and Apple Mail on desktop support per-user client-side rules layered on top of whatever the provider already decided, and this layer is invisible to the sender entirely. It is trained (or configured) per recipient, one inbox at a time.

What the user actually sees

A founder using Raisy’s investor outreach flow doesn’t choose “plain text or HTML” from a dropdown and hope. The system already defaults first-touch investor emails to a lightly formatted, mostly-plain structure, because that is the format that survives the machine-learning engagement layer (recipients treat it as correspondence, not a campaign) without failing the authentication or gateway layers, since every send already goes out authenticated from a warmed domain per the compliance gate. The founder sees one send button; underneath it, four separate checks already ran.

Why this matters beyond convenience

The plain-text-vs-HTML framing treats deliverability as a single knob a founder can turn correctly if they just pick the right template. The evidence says otherwise: inbox placement varies by 14 points between major providers on the exact same message, DMARC enforcement sits at barely half of top domains even in 2026 (and most of that half is still monitor-only, not enforced), and the layer that most decides outcome, recipient engagement, is invisible to the sender until after the fact. Founders don’t need a stronger opinion on HTML; they need each of the four gates handled correctly, in order, before the question of wording even comes up.

Every number above reads from one committed data file, download it directly: raisy_email_format_deliverability_data.xlsx.

What this post does not answer: it does not report Raisy’s own live inbox-placement numbers (the compliance and authentication gate is verified pre-send; production placement data is a follow-up once real campaigns run at volume, per the outreach-cost post); it does not test a controlled plain-text-vs-HTML split on identical send infrastructure (the cited comparisons are industry aggregate data, not a single controlled experiment); and it does not cover client-side filter training in detail, since that layer is per-recipient and outside any sender’s visibility by design.

See your compliance and authentication status before your first investor email, raisy.team