Measuring deliverability: what you can and cannot know

No instrument exists that reports your true inbox placement rate. Not one. Every number you have ever been shown for it was an estimate, and most of the numbers senders stare at daily (delivered rate, open rate) measure something else entirely. Measuring deliverability honestly means knowing what each available instrument can actually see, where it goes blind, and how to combine several partial views into a picture you can run a sending operation on.

Delivered is not inboxed

Start with the number your ESP shows in big friendly type: delivered, usually somewhere north of 98%. Here is what it actually measures. When your ESP's server hands a message to Gmail, the two servers hold a short SMTP conversation, and Gmail ends it with a code: accepted or rejected. Delivered means accepted. That is the entire claim.

Where the message went after acceptance is invisible to your ESP. Gmail routinely accepts a message and files it straight into spam; the sending server sees the same success code either way. So a campaign can report 99% delivered while a third of it sits in spam folders, and your ESP is not lying to you. It is reporting the only thing the protocol lets it see. The two quantities have names worth keeping separate in your head: delivery rate is acceptance, inbox placement is the folder, and no ESP dashboard on the market shows you the second one.

Open rates stopped meaning opens

For years the workaround was the open rate. If placement dropped, fewer people saw the mail, so fewer opened it, and the open rate fell with it. Crude but usable. Then Apple broke it.

An open is recorded by a tracking pixel: a tiny invisible image loaded from the ESP's server when the message is displayed. Apple Mail Privacy Protection, shipped with iOS 15 in 2021, preloads those images through Apple's own proxy servers for mail arriving in Apple Mail, whether or not a human ever looks at the message. Every such recipient registers as an open. On a list with many Apple Mail readers, a large share of your recorded opens are machines.

Two consequences. Comparing your open rate to a benchmark or to anyone else's is now meaningless, because the Apple share of every list differs. And absolute open rates flatter you. What survives is the trend of your own list against itself (the prefetching is roughly consistent, so a real drop still shows as a drop) and clicks, which are harder to fake, though corporate security scanners do follow links, so B2B click data carries its own ghosts.

The instruments worth having

Three sources give real signal, and all of them are free.

Google Postmaster Tools is Gmail's own view of your domain. You verify ownership with a DNS record and get charts of your domain reputation (graded in four bands; high is the goal, bad means the filter has made up its mind), your spam complaint rate, authentication success and delivery errors. The complaint chart is the exact number the 2024 bulk sender rules hold you to: stay under 0.3%, and ideally under 0.1%. Two honest caveats. It covers Gmail traffic only, and low-volume senders often see empty charts, because Google only draws them once your daily volume crosses a threshold it does not publish.

Microsoft SNDS (Smart Network Data Services) is the Outlook-side equivalent, with a catch: it is organized by IP address rather than domain, and you must prove control of the IPs to register them. If you send through Mailchimp or any other shared ESP infrastructure, the IPs are theirs, so SNDS is closed to you. It earns its keep for people running their own servers or dedicated IPs, where it shows complaint volume and spam-trap hits, plus a color grade for how your traffic was treated.

DMARC aggregate reports are the third instrument: the large receivers that participate (Google and Microsoft among them) will mail you a daily summary of which servers sent mail as your domain and whether it authenticated. They measure authentication rather than placement, and they see every sender using your name, including the ones you forgot about. They get their own lesson: reading DMARC reports.

Seed testing, honestly

A seed test sends your campaign to a panel of mailboxes the testing service controls at dozens of providers, then reports which folder each copy landed in. It looks like the direct placement measurement everything above is missing. It is not, for one structural reason: seeds are not your audience.

Placement at the big providers is personalized. A recipient who opens you weekly and a recipient who deleted your last ten sends can receive the same campaign into different folders. Seed mailboxes have no history with you at all, which makes them a systematically strange sample of exactly the variable that matters. Panels also get discovered over time, because they behave like no human ever does, and filters have little reason to treat them as representative.

Where seeds still earn something: catching a categorical block. If every seed copy at one provider folders while other providers inbox, that is a real signal about that provider. As a precise percentage ("87.4% inbox placement"), treat seed results as theater. We deliberately built Inbox Decay without a seed panel for this reason; I would rather show you a verifiable layer than an authoritative-looking guess.

The layer you can measure with certainty

Under all the estimation sits one layer that is binary and public: the records themselves. Your SPF record parses or it does not. There is one of them, or there are two. The DKIM key at a selector exists or it does not, and its length is a fact. DMARC is published or missing, and its policy is right there in the text. Anyone can read this layer from outside with plain DNS queries, no cooperation from you required, and it is where a startling share of real-world deliverability failures actually live.

That observation is the spine of how we measure things. The free health check reads that outside-in layer for any domain. For the part DNS cannot show, meaning what receivers actually computed about your mail, we grade real emails you send to a test address, because a delivered message carries the receiver's verdicts in its headers. What we see monitoring domains, week after week, is people agonizing over the unmeasurable (content tweaks, send-time folklore) while something perfectly measurable, like a second SPF record pasted in by a well-meaning colleague, sits broken in their DNS.

Assembling a working picture

The practical stack for most senders costs nothing. Postmaster Tools for the Gmail view. DMARC aggregates for the authentication view. Outside-in record checks for the configuration view, and your own engagement data split by recipient domain for the trend view. That last one is underrated: group opens and clicks by the recipient's mailbox provider (gmail.com addresses versus outlook.com versus everything else) and a provider-specific reputation problem stops hiding inside a blended average. A cliff at Gmail while Microsoft holds flat is a diagnosis, and it took you one spreadsheet filter.

None of these instruments is the whole truth. Together they are close enough to run a professional operation on. The Part 3 checkpoint is next; the forensic craft of reading email headers, where the receiver's own record of each message enters the kit, waits in the diagnosis part of the course.

Terms from this lesson

• delivery rate - the share of messages the receiving server accepted during the SMTP conversation. It says nothing about which folder the message landed in.
• inbox placement - which folder a delivered message actually reached. No sender-side dashboard can see it directly.
• tracking pixel - a tiny invisible image loaded from the sender's server when a message is displayed, used to record opens.
• Apple Mail Privacy Protection - an Apple Mail feature, shipped in 2021, that preloads remote images through Apple's proxies. It records an open whether or not a human read the message.
• Google Postmaster Tools - Google's free dashboard showing your domain reputation and spam complaint rate for mail sent to Gmail.
• SNDS - Microsoft's Smart Network Data Services, an IP-level view of how your traffic to Outlook mailboxes is treated. You must control the sending IPs to use it.
• seed testing - sending a campaign to a panel of test mailboxes at many providers to see which folder each copy lands in. Limited, because seeds have no engagement history with you.

Check yourself

1. Your ESP reports a campaign as 99% delivered. What does that actually mean?

99% of messages reached the inbox99% of recipients opened the messageThe campaign passed authentication at 99% of providers99% of messages were accepted by the receiving servers

2. Why did open rates stop being a reliable placement signal?

Apple Mail Privacy Protection preloads tracking pixels and records opens no human madeProviders no longer allow images in emailOpen tracking was banned under the 2024 bulk sender rules

3. Why is Microsoft SNDS closed to most senders on a shared ESP?

It only covers Gmail trafficIt costs too much for small sendersIt is organized by IP address, and shared IPs belong to the ESP, not to youIt requires a DMARC policy of p=reject

4. What is the structural weakness of seed testing?

Seed tests are forbidden by most providers' termsSeed mailboxes have no engagement history, and placement is personalizedSeeds only exist for Gmail