---
title: Can you rely on AI for email marketing and deliverability?
description: Can you rely on AI for email deliverability? Where it gets its information, six ways it fails in this field, what it is good at, and why expertise comes first.
---

[Skip to content](https://weareastral.co.uk/thevault/can-you-rely-on-ai-for-email-marketing-and-deliverability-only-in-proportion-to-how-much-you-already-know#main-content)

**FREE WEBINAR:** Email marketing in 2027: everything you need to know

[Sign up now](https://weareastral.co.uk/webinar/email-marketing-in-2027)

![astral.](https://weareastral.co.uk/hs-fs/hubfs/astral.%20new%20logo%20no%20background%20large.png?width=500&height=200&name=astral.%20new%20logo%20no%20background%20large.png)

- Email services
  
  Show submenu for Email services 
  
    - [Email Strategy Workshops](https://weareastral.co.uk/email-strategy-workshops)
    - [Email Audit](https://weareastral.co.uk/email-audit)
    - [Deliverability Audit](https://weareastral.co.uk/email-deliverability-audit)
- Email & CRM resources
  
  Show submenu for Email & CRM resources 
  
    - [Deliverability Hub](https://weareastral.co.uk/deliverability-hub)
    - [Email & CRM Vault](https://weareastral.co.uk/thevault)
    - [Email Copy & Psychology Guide](https://weareastral.co.uk/email-copy-psychology-guide)
    - [Email Design Handbook](https://weareastral.co.uk/email-design-handbook)
    - [Email Deliverability Guide](https://weareastral.co.uk/email-deliverability-guide)
    - [Email Health Check tool](https://www.emailhealthcheck.co.uk/)
    - [Deliverability Health Check tool](https://www.emaildeliverabilitycheck.co.uk/)
- [astral academy](https://weareastral.co.uk/academy)
  
  Show submenu for astral academy 
  
    - [Deliverability Certification](https://weareastral.co.uk/emaildeliverability/training)
    - [Email Design Masterclass](https://weareastral.co.uk/email-design-masterclass)
    - [Inbox Psychology Masterclass](https://weareastral.co.uk/email-psychology-masterclass)
    - [Free Deliverability Training](https://weareastral.co.uk/free-email-deliverability-training)

Open main navigation

Close main navigation

- Email services
  
  Show submenu for Email services 
  
    - [Email Strategy Workshops](https://weareastral.co.uk/email-strategy-workshops)
    - [Email Audit](https://weareastral.co.uk/email-audit)
    - [Deliverability Audit](https://weareastral.co.uk/email-deliverability-audit)
- Email & CRM resources
  
  Show submenu for Email & CRM resources 
  
    - [Deliverability Hub](https://weareastral.co.uk/deliverability-hub)
    - [Email & CRM Vault](https://weareastral.co.uk/thevault)
    - [Email Copy & Psychology Guide](https://weareastral.co.uk/email-copy-psychology-guide)
    - [Email Design Handbook](https://weareastral.co.uk/email-design-handbook)
    - [Email Deliverability Guide](https://weareastral.co.uk/email-deliverability-guide)
    - [Email Health Check tool](https://www.emailhealthcheck.co.uk/)
    - [Deliverability Health Check tool](https://www.emaildeliverabilitycheck.co.uk/)
- [astral academy](https://weareastral.co.uk/academy)
  
  Show submenu for astral academy 
  
    - [Deliverability Certification](https://weareastral.co.uk/emaildeliverability/training)
    - [Email Design Masterclass](https://weareastral.co.uk/email-design-masterclass)
    - [Inbox Psychology Masterclass](https://weareastral.co.uk/email-psychology-masterclass)
    - [Free Deliverability Training](https://weareastral.co.uk/free-email-deliverability-training)
- [Work with Beth](https://weareastral.co.uk/work-with-beth)

[Work with Beth](https://weareastral.co.uk/work-with-beth)

 11/2026

# Can you rely on AI for email marketing and deliverability?

![Picture of Beth O'Malley](https://app.hubspot.com/settings/avatar/2761c9d87ba551994b382ed8415ad538) [Beth O'Malley](https://weareastral.co.uk/thevault/author/beth-omalley)

Share: [Share this page on LinkedIn](http://www.linkedin.com/shareArticle?mini=true&url=https://weareastral.co.uk/thevault/can-you-rely-on-ai-for-email-marketing-and-deliverability-only-in-proportion-to-how-much-you-already-know) [envelope icon](mailto:?body=https://weareastral.co.uk/thevault/can-you-rely-on-ai-for-email-marketing-and-deliverability-only-in-proportion-to-how-much-you-already-know)

[← Back to the Vault](https://weareastral.co.uk/thevault)

 

#### Can I not just ask AI?

Yes, yes you can. But remember that AI is useful in proportion to how much you already know. If you understand how email works, it is a powerful assistant. If you do not, you cannot tell when it is wrong, which is precisely the situation in which you most need to.

That is not an argument against using it, I use it every day. It is an argument about what it is for, and about where the knowledge has to come from first.

Also, remember to use your brain so your brain cells don't die xox

 

## And before anything else, how this post was made

I write every post in The Vault with AI, so it would be odd not to say so.

My process is the same every time; I speak the blog out loud. Everything I want to say, teach and to tell you, completely unstructured, in whatever order it arrives in my head. Then the AI helps me put it together: it suggests a structure, better headers and sections, tells me where a point needs expanding, and reads the draft against everything else I have published so it can link the pieces up.

Which makes it a very good assistant. The knowledge is mine: twelve years of recovering block listings and auditing emails, the positions, the client examples, the frameworks, and the judgement about what is true. What the AI does is help me craft it into something easy to read, which is a real job and one it does well.

It also gets things wrong in ways only I can catch: a headline nobody would ever search for, a framing that leaned on deliverability when the post needed to be broader, and, working from an AI-generated brief, a certificate price that was for the wrong product entirely quoted alongside a supplier who no longer issues them.

That division of labour is the whole answer to the question this post is asking. The expertise decides, the tool executes. Reverse the order and you get something that reads beautifully and is quietly incorrect.

 

## What this covers

1. Where AI gets its information about email, and why that matters here more than in most subjects.
2. Six ways it fails specifically at deliverability.
3. What it is good at, which is a lot.
4. How to use it without getting hurt.
5. And how to tell a human expert worth listening to from one who is not.

 

## Before you dig in, why don't you get access to RE:markable

  RE:markable is the weekly email about emails. Dropping the latest email marketing news, updates, insights, free resources, upcoming masterclasses, webinars, and of course, a little inbox mischief. 

 Subscribe to RE:markable

 

 

# The source

## Where AI gets its information, and why the Wikipedia comparison is too kind

People reach for Wikipedia as the comparison, and I understand why. Built by humans, assembled from what is published, broadly reliable but not something you would cite in court.

It is too generous. Wikipedia has citations, an edit history, talk pages where disagreements are visible, and flags on contested claims. You can see the seams. You can tell when something is disputed, thin, or two people arguing.

A language model gives you the confident average of the internet with the citations stripped out and the uncertainty smoothed away. Everything arrives in the same assured tone whether it is a settled fact, a vendor's marketing claim, or something that stopped being true in 2021.

 

### Which would matter less if the source material were any good

- The published corpus on deliverability is unusually poor. It is dominated by vendor content written to rank, recycled best practice, and listicles produced by people who have never diagnosed a placement problem in their lives.
- So the model learned the average of published opinion. Not the correct answer, the average. And the average of published opinion about email deliverability is wrong, which is the thing I have been saying for most of a decade.
- It also learned the volume, not the authority. A claim repeated across four hundred SEO blog posts carries more weight in training than the same subject covered correctly once by somebody who knows the subject.

# The failures

## Six ways it goes wrong, specifically in this field

 

One: it is confidently out of date, in a field that moves monthly

A sample of the last six months. Microsoft removed spam trap hit counts from SNDS in July 2026. Google retired Postmaster Tools v1 in September 2025, taking the IP and domain reputation metrics with it. Spamhaus began listing around a million cold email domains from late August 2026. Gmail Live started rolling out in October 2026.

Ask most models about any of that and you will get the previous world, stated with complete confidence and no indication that it might not know. A model does not experience the edge of its own knowledge. It has no sensation of a gap.

Two: the myths are over-represented

Do not use the word free, avoid exclamation marks, mind the spam trigger words. Twenty years of content marketing says so, so the model says so, and somebody new to email spends a fortnight rewriting subject lines instead of fixing their authentication.

Three: it cannot see your data, and deliverability is diagnostic

Opens have fallen. That has at least ten causes, and the only way to separate them is your own numbers split by provider, by opt-in source, by sending stream, read as a trend. An AI working from your description of a symptom will give you a confident, plausible fix for a problem you may not have.

Which is the expensive kind of wrong. Not obviously nonsense, just aimed at the wrong thing, consuming a quarter of somebody's effort while the real cause keeps going.

Four: it answers the question you asked

Always. Which sounds helpful and is the most underrated risk on this list, because the skill in deliverability is knowing which question to ask.

Ask how to improve open rates and you will get a competent answer about open rates, when the problem was that half your list never opted in. The diagnosis is the expertise, and you cannot prompt your way to a diagnosis you do not know you need.

Five: it invents specifics, and the specifics are what break things

- **Thresholds, syntax, and which provider does what.** A plausible-looking SPF record with one too many lookups. A DMARC tag that does not exist. A complaint threshold attributed to the wrong provider.
- **All of it delivered in the same tone as the correct parts.** Which is the problem. There is no change in register when it stops knowing and starts filling in.

Six: it cannot do the part of the job that is hardest

Most real deliverability work is not technical, it is organisational. Getting sales to stop prospecting from the main domain. Persuading somebody senior to accept a smaller list. Getting IT to publish a DNS record in under three weeks. Deciding who wins when three teams all want to send on Thursday.

No tool does any of that. It requires somebody in the building with evidence, credibility and the patience to have the same conversation four times.

# The upside

## What it is good at, and I would not work without it

A post that told you AI was useless would be dismissed by everybody who uses it daily, including me. Used properly it is the biggest productivity change of my career.

- Drafting, once you know what you want to say. The thinking is yours. The first draft does not have to be.
- Explaining something you half understand. At your level, as many times as you need, without anybody watching.
- Structuring research you have already gathered. Turning a pile of notes, transcripts and sources into something with a shape.
- Writing the query to pull the data. Which gets you to the diagnosis faster, as long as you know what you are looking for.
- The repetitive production work. Alt text, subject line variants, first-pass QA, checking a template against a list of requirements.
- And working through documented specifications. Record syntax, provider requirements, standards. Reliable, as long as you verify against the primary source rather than its summary of it.
- As a thinking partner, when you already know enough to catch it. Which is the whole point, and the condition every item on this list depends on.

 

# The practice

## How to use it without getting hurt

1. Never let it be your only source on anything current. Provider requirements, thresholds, tool capabilities and enforcement all change faster than training data. Check the provider’s own documentation, every time, for anything you are going to act on.
2. Ask it to show you where a claim comes from. And then go and look. If it cannot point you at a source, treat the claim as a hypothesis rather than a fact.
3. Use it on your data, not instead of your data. Paste in your own numbers and have it help you interrogate them. That is a completely different activity from asking it what usually causes this.
4. Never let it near your DNS without verification. Authentication records are the one place a plausible-looking error has immediate and total consequences.
5. Notice when it agrees with you too easily. If you ask whether your plan is sensible, you will usually be told it is. Ask it to argue the opposite case and see whether the objection lands.
6. And treat fluency as a warning rather than a reassurance. The most dangerous output is the one that reads perfectly and contains a confident number nobody has checked.

 

# The humans

## Which humans, though, because plenty of them are wrong too

I am not going to argue that anybody with a pulse beats a model. Most of the bad deliverability content the AI learned from was written by humans. The question is which humans.

- Do they show their working? Somebody who tells you what they checked, and where, is giving you something you can verify. Somebody who gives you a rule is asking you to trust them.
- Will they say I do not know? The single most reliable marker of expertise in this field, and the one thing a language model structurally cannot do. Mine comes up more often than clients expect.
- Do they change position when the evidence changes? Email changes constantly. Anybody giving you the same advice they gave in 2019 has stopped paying attention.
- Do they have anything at stake? Somebody who will still be there in six weeks when the listing arrives has a reason to be right. A tool has no exposure of any kind.
- And are they selling you the diagnosis or the cure? Be careful with anybody whose recommendation is always the thing they happen to sell. Including me, which is why most of what I publish is free and why the first thing I tell people is usually to send less email.

### Where to get it from people

- Primary sources, which are not people but are authoritative. Google, Microsoft, Yahoo and Apple publish their sender requirements. Spamhaus publishes its reasoning. Read those rather than somebody's summary of them.
- Practitioners who publish their mistakes. The useful ones write about what went wrong and what they did about it, not about seven tips for better open rates.
- Industry bodies and the standards community. Where the people who build the filtering talk directly to the people who send the mail.
- And somebody who has seen your kind of problem before. Pattern recognition across many programmes is the thing neither a model nor a smart beginner can substitute for.

# The argument

## Why learn it, when you could just ask

Because the value of the tool depends entirely on what you bring to it, and that is not a slogan. It is the mechanism.

- Knowledge is what lets you spot the wrong answer. Every failure above is invisible to somebody without a working model of how email arrives. With one, they are obvious in seconds, and the tool becomes powerful.
- Diagnosis cannot be delegated. An AI cannot look at your sending, your sources and your infrastructure and tell you which of ten causes you have. You, or somebody you hire, has to do that.
- And the organisational work needs a credible human. The person who can show the board what placement is doing to revenue is the person who gets the decision made. That credibility comes from understanding the subject, not from access to a tool everybody has.

### Which is what the Certification is built to do

- You run a full audit on your own email marketing as you go. Your data, your domains, your sources, your placement. The exact thing no AI can do for you, because it cannot see any of it.
- It is CPD certified, with live sessions. You can ask me the question behind your question, which is the part that matters and the part a prompt cannot reach.
- You finish with a diagnosis and a prioritised plan. Not a summary of what the internet thinks about email.
- And afterwards, AI becomes properly useful to you. Which is the outcome I would claim. Not that you will stop using it. That you will finally be able to tell when it is wrong, and get real value out of it the rest of the time.

[See the Email Deliverability Certification →](https://weareastral.co.uk/emaildeliverability/training)

 

 

## A quick win:

*Take one piece of email advice an AI has given you recently and go and check it against the provider's own documentation.*

*Somewhere between a third and a half of the time, in my experience, you will find it is out of date, attributed to the wrong provider, or stated with more certainty than the source supports.*

## The conclusion

We have drifted into treating AI as the place knowledge lives, and for email it is not, because what it learned from was mostly content written to rank rather than content written to be right.

It is an extraordinary tool for doing things. It is a poor place to find out what is true in a field that changes every month, cannot be diagnosed without your own data, and is half organisational anyway.

So learn the thing. Then use the tool on top of it, hard, and get far more out of it than the people who skipped the first step. The expertise is what makes the tool worth having, and that order does not reverse.

### Further reading from The Vault:

- [The State of Email Deliverability in 2026](https://weareastral.co.uk/thevault/the-state-of-email-deliverability-in-2026.-and-what-i-think-happens-in-2027)
- [Synthetic Engagement Is Now a Blocklistable Offence](https://weareastral.co.uk/thevault/synthetic-engagement-is-now-a-blocklistable-offence.-what-validitys-heatwave-means-for-anybody-using-a-warm-up-tool)
- [Spam Traps: Every Type Explained](https://weareastral.co.uk/thevault/spam-traps-every-type-explained-how-they-end-up-on-your-list-and-how-to-find-out-whether-you-have-one)
- [Gmail for Email Marketers: Every Feature That Decides Whether You Land](https://weareastral.co.uk/thevault/gmail-for-email-marketers-every-feature-that-decides-whether-your-email-lands-gets-read-or-gets-unsubscribed-from)
- [What I Say vs What the Internet Says About Email](https://weareastral.co.uk/thevault/email-marketing-best-practice-is-mostly-wrong.-what-the-internet-says-what-i-say-and-why-we-have-all-been-taught-email-backwards)
- [Who Owns Email Deliverability?](https://weareastral.co.uk/thevault/who-owns-email-deliverability-not-it)

 

## Like this blog? You'll love RE:markable

 RE:markable is the weekly email about emails. Dropping the latest email marketing news, updates, insights, free resources, upcoming masterclasses, webinars, and of course, a little inbox mischief.

 Subscribe to RE:markable

 

## Email, CRM and HubSpot Support

I help marketers and businesses **globally** improve, design and fix their email, CRM, and HubSpot ecosystems, from strategy through to execution.

**My services include:**

- Email marketing strategy, audits, training, workshops, and consultancy
- CRM strategy and enablement
- Full HubSpot implementations, optimisation and onboarding through my agency

If you’re looking for experienced external support (and lots of enjoyment along the way), this is where to start.

[Work with Beth](https://weareastral.co.uk/work-with-beth)

 

[← Back to the Vault](https://weareastral.co.uk/thevault)

[Email Marketing](https://weareastral.co.uk/thevault/tag/email-marketing), [Marketing Strategy](https://weareastral.co.uk/thevault/tag/marketing-strategy)

![a. final -2](https://weareastral.co.uk/hs-fs/hubfs/a.%20final%20-2.png?width=100&height=100&name=a.%20final%20-2.png "a. final -2")

Copyright © Astral Digital ™. All Rights Reserved By Astral Digital.  
Registered company number 1460594

```json
{
  "@context" : "https://schema.org",
  "@type" : "BlogPosting",
  "author" : {
    "@type" : "Person",
    "name" : "Beth O'Malley",
    "url" : "https://weareastral.co.uk/thevault/author/beth-omalley"
  },
  "dateModified" : "2026-11-17T12:53:54.616Z",
  "datePublished" : "2026-11-17T12:53:54.000Z",
  "headline" : "Can you rely on AI for email marketing and deliverability?",
  "mainEntityOfPage" : {
    "@id" : "https://weareastral.co.uk/thevault/can-you-rely-on-ai-for-email-marketing-and-deliverability-only-in-proportion-to-how-much-you-already-know",
    "@type" : "WebPage"
  },
  "publisher" : {
    "@type" : "Organization",
    "logo" : {
      "@type" : "ImageObject",
      "url" : "https://weareastral.co.uk/hubfs/astral.%20new%20logo%20no%20background%20large.png"
    },
    "name" : "Astral Digital"
  }
}
```