Latest / Marketing Automation with Fexingo: HubSpot, Marketo, and Automated Customer Journeys / Why Your Email Subject Lines Fail an AI Filter
Transcript
- Lucas: Luna, I want to start with a number that stopped me this morning. Gartner published a report last week — it says that as of the second quarter of 2026, ai powered inbox filters are burying roughly 67 percent of marketing emails before a human being ever lays eyes on them. Luna: Sixty-seven percent? That's almost two out of three emails that never reach the inbox. That feels like a crisis for anyone running automated campaigns. Lucas: It is. And the really interesting part is that the filters aren't just looking for obvious spam keywords anymore. They're analyzing patterns — subject-line length, punctuation frequency, even the ratio of promotional language to neutral language. If your subject line looks like it was written by a marketer, the filter flags it. Luna: So the more you optimise for opens, the more you get penalised by the filter. That's a brutal feedback loop. Lucas: Exactly. One example the report highlighted: subject lines that start with a number — '5 Ways to,' '3 Tips for' — those used to be best practice. Now they're one of the strongest predictors of filtering. The AI has been trained on millions of emails, and it's learned that pattern equals promotional. Luna: So what's actually working? If the old playbook is broken, what are brands doing that still gets through? Lucas: A couple of brands that are consistently beating the filter are Bombas and Away. Bombas, the sock and apparel company, they've shifted to what they call 'behavioral subject lines' — instead of 'Your perfect spring socks are here,' they'll send something like 'You left these in your cart last Tuesday.' It references a specific action, not a generic promotion. Luna: Right, because the filter sees a reference to a user's actual behavior and classifies it as transactional, not promotional. Transactional emails get a much higher pass rate. Lucas: Precisely. The Gartner data backs that up — transactional subject lines have a pass rate above 90 percent. Promotional ones are down around 30 percent. So Bombas is essentially turning every promotional email into a semi-transactional one by tying it to a real event. Luna: And Away? They're luggage, right? Lucas: Yeah, travel gear. Away took a different approach. They looked at their subject-line data and realised that any line containing an exclamation mark was being filtered at a rate of almost 80 percent. So they removed all exclamation marks from their automated flows. Just like that. They rewrote their entire sequence with plain, declarative sentences. Luna: No exclamation marks. That's such a simple change, but I bet it hurt to do. Marketers love exclamation marks. Lucas: They do. But Away saw their inbox placement rate jump from 58 percent to 83 percent within a month. Their open rate actually dropped slightly — but their click-through rate went up, because the people who did see the email were the ones who actually wanted it. Luna: So a smaller, more engaged audience. That's a better outcome for the business, even if the vanity metric of open rate takes a hit. Lucas: Exactly. And this is where I think the automation angle gets really interesting. Most marketing automation platforms let you set up A/B testing on subject lines. But the test is usually about which line gets more opens. What you really need to test is which line gets past the filter and into the inbox. Luna: But most platforms don't report on filter pass rates, do they? You can see open rates, but you can't see how many emails were filtered before they even had a chance. Lucas: Not natively, no. But there are third-party deliverability tools that plug into HubSpot, Marketo, Salesforce Marketing Cloud — tools like Everest or InboxAlly — that give you that data. And once you see it, the patterns are pretty stark. Luna: Give me a concrete pattern that listeners can check in their own campaigns today. Lucas: Sure. The single strongest filter trigger, according to the Gartner data, is what they call 'subject-line asymmetry' — that's when the subject line uses a different tone or format than the sender's historical average. If a brand usually sends plain, short subject lines and suddenly sends a long, punctuation-heavy one, the filter flags it as anomalous. Luna: So consistency itself is a deliverability strategy. The AI learns your brand's normal voice, and any deviation looks suspicious. Lucas: Right. So the fix isn't just about what you write — it's about how consistently you write. Bombas's subject lines, for example, all follow the same structure: a user action, a time reference, and a simple verb. 'You viewed this item yesterday.' 'Your discount expires soon.' 'We restocked your size.' Every single one fits that pattern. Luna: That kind of discipline is hard to scale across a whole team. How do they enforce it? Lucas: They built a subject-line template library inside their automation platform — HubSpot, in their case — and locked the flows so that no one can edit the subject line without approval. It's a process constraint, not a creative constraint. Luna: Interesting. So they treat subject lines like code, not copy. You don't just write a subject line; you follow a function. Lucas: Exactly. And I think that's the shift we're going to see more of. As AI filters get smarter, the creative freedom marketers had is going to narrow. But the trade-off is better deliverability, which means more actual humans seeing your message. Luna: I want to ask about one more thing: what about personalisation? Like using the recipient's first name in the subject line. Does that still help, or is the AI now filtering for that too? Lucas: That's a great question. The Gartner report had a specific finding on that. First-name personalisation in the subject line — 'Hey Luna, check this out' — actually increases the filter probability by about 12 percent when combined with any promotional language. The AI sees the name as a pattern matching a known spammer technique. Luna: So personalisation, which was supposed to be the holy grail, is now working against us in some cases. Lucas: In certain contexts, yes. But if the personalisation is inside the body of the email, it still has a positive effect on engagement. The subject line is the one place where the filter scrutinizes every character. Luna: So the rule of thumb might be: keep subject lines plain, short, and consistent with your brand's established voice, and move all the personalisation into the preview text or the body. Lucas: That's exactly the takeaway. And if you want a simple test to run this afternoon, pick your top-performing automated email from last month, rewrite the subject line as a short, declarative sentence with no punctuation except a period at the end, and run it as a split test against the original. Measure inbox placement rate, not open rate. I think you'll be surprised. Luna: I'm actually going to do that with one of our own flows. And speaking of the show — if these marketing conversations have sparked something you've actually used in your own work, there's a quiet way to keep this going. Lucas: Yeah. A small group of listeners already chips in monthly at buy me a coffee dot com slash fexingo. No ads, no sponsors — just listener support covering the production costs. It's a simple setup, and it means we can keep digging into topics like this without any outside pressure. Luna: If that's something that fits your budget, we appreciate it. If not, no problem — the episodes stay free either way. Now, back to subject lines: one other pattern I want to flag is the use of urgency language. Lucas: Right. 'Act now,' 'limited time,' 'only a few left' — those phrases are basically a red flag to the AI. Gartner found that subject lines containing any urgency word have a filter rate above 75 percent. The irony is that genuine urgency — like an actual expiration date — often gets caught in the same net. Luna: So what's the workaround? If you genuinely have a time-sensitive offer, how do you communicate that without getting filtered? Lucas: Bombas handles this by putting the deadline only in the preview text, not the subject line. The subject line just says 'Your code is waiting.' The preview text says 'Expires in 24 hours.' The filter sees the subject line as neutral, and the preview text isn't scored as heavily. Luna: That's clever. Use the subject line to earn the inbox placement, then use the preview text to drive urgency. The filter doesn't penalize the preview text the same way. Lucas: Exactly. And that's the kind of structural thinking that the best marketing automation teams are adopting. It's not about being more clever with words. It's about understanding the rules of the game the AI is playing and working within them. Luna: Alright, so the big takeaway for anyone listening: check your subject-line patterns against your brand's historical average, remove exclamation marks, avoid urgency words in the subject line, and move personalisation to the preview text. And test for deliverability, not just open rate. Lucas: That's the checklist. And I'd add one more thing: look at your own email data from the past six months. Do you see a correlation between subject-line format changes and sudden drops in open rate? That's often a deliverability issue, not a content issue. Once you start looking, the patterns are hard to miss.