Inbox Placement Testing: How to Measure Where Your Email Really Lands

Inbox placement testing answers a question that delivery reports cannot: after a provider accepts your message, where does it actually go? A receiving server saying yes only means the message passed the door. It may then land in the inbox, in a Promotions tab, in spam, or in a quarantine nobody looks at. This guide shows us how to measure placement honestly, what seed lists can and cannot tell us, and how to turn noisy results into decisions.

Delivered is not inboxed

Most sending dashboards report a delivery rate. It is the share of messages the receiving server accepted without a bounce. That figure is useful, but it stops at the SMTP conversation. What happens next is decided by the mailbox provider after acceptance, using reputation, authentication, content, recipient history and many signals we never see.

Consider a campaign of 50,000 messages with 49,000 accepted. A 98 percent delivery rate looks excellent. If the provider then routes a large share of those messages to spam, the real reach could be far lower, and the report would still look healthy. The reverse also happens: a small bounce rate can hide good placement for everyone else. The two measures answer different questions, so we keep them apart.

Measure	What it tells us	What it cannot tell us
Accepted / delivered	The receiving server took the message	Which folder or tab it reached
Bounced	Rejected or failed, with a reason code	Anything about messages that were accepted
Inbox placement	The message reached the main inbox for a given mailbox	Whether a real recipient saw it
Spam folder placement	The message was filtered as unwanted	Why, without further evidence
Missing / not found	Accepted but never visible, possibly quarantined or dropped	Where it went
Opened / clicked	A person (or a machine) interacted	Reliable placement, since opens can be inflated by privacy features

Opens are a weak placement proxy for another reason. Some mail clients preload images, which can mark messages as opened without a human reading them. Our guide to Apple Mail Privacy Protection and open rate reliability explains why clicks and replies are safer to trust. For bounce failures, see SMTP bounce codes explained.

How seed-list testing works

A seed list is a set of test mailboxes at various providers. You send your message to those addresses along with, or just before, a real campaign. Then you check where each message landed: inbox, spam, a tab, or nowhere. The result is shown as a percentage by provider, for example 8 of 10 test mailboxes placed in the inbox.

The appeal is obvious. It is quick, it is repeatable, and it does not require waiting for subscribers to react. Used with realistic expectations it is a smoke detector. It tells us when something is very wrong, and it helps compare two versions of a message. It does not give a census of what happened to our subscribers.

Where seed results mislead

• Small samples. Ten mailboxes at one provider can swing by 10 percentage points when a single message moves. That is not a trend.
• No history. A new mailbox that never opened, replied to or moved your mail is a stranger to the filter. Real subscribers have a relationship with your sender, and filters personalize for that.
• Unnatural patterns. Seed addresses that appear on many senders lists, never log in, and receive mail from everyone can look unlike a normal inbox to a filter.
• Not representative of your list. Your audience may be mostly consumer Gmail and a few business domains, while the seed set mixes providers evenly.
• Different path. If tests are sent from a different IP, domain or stream than your production mail, the result describes that path, not your real one.
• Single-moment view. Placement can change within hours as complaints and engagement arrive. A test run at send time cannot see how the campaign performs later.

Because of these limits, we avoid presenting a seed percentage as the inbox rate. A fair statement is: on our test mailboxes, this template placed 9 of 10 in the inbox at provider A and 6 of 10 at provider B, and our control placed 9 of 10 and 9 of 10. That is evidence of a difference worth investigating, not a measured rate for the audience.

Panel data, real-user data and seed accounts

Placement can be observed from three different vantage points. Each has its own strengths and its own blind spots, and the honest approach is to know which one you are looking at.

Source	How it works	Strength	Bias or limit
Seed accounts	Test mailboxes you or a service control receive your mail	Direct view of folder placement, fast and repeatable	Small, synthetic, little history, not your audience
Panel data	Aggregated data from a group of consenting real users of a mailbox or toolbar program	Real people and real behavior	Panel members are a self-selected slice, coverage varies by provider and region
Provider-reported data	Dashboards such as Google Postmaster Tools, plus feedback loops from some providers	Comes from the provider itself, tied to your real traffic	Aggregated, delayed, thresholds apply, little detail on folders
Your own engagement data	Clicks, replies, conversions and unsubscribes by recipient domain	Real recipients of your actual mail	Opens are unreliable, low volume gives wide uncertainty

We recommend weighing them in this order when they disagree: provider-reported data and your own click, reply and complaint data first, panel data second, and seed accounts as an early-warning and comparison tool. If seeds show a drop but Postmaster, complaints and clicks are stable, we watch and retest. If all of them move the same way, we act.

Build your own small seed set of real mailboxes

A set of mailboxes we control has real advantages: it costs little, we know exactly how it is configured, and nobody else sends to it. The goal is not volume. The goal is a few realistic mailboxes that behave like ordinary recipients.

1. Choose providers by audience. Look at your list by recipient domain and create mailboxes at the providers that cover most of your volume. Add a business mailbox on a domain that uses its own filtering if you send to companies.
2. Give each mailbox a history. Use it normally for a few weeks before relying on it: send and receive ordinary mail, reply to some messages, move a few from spam to inbox, and keep the account logged in on a regular basis.
3. Subscribe like a real person. Sign up through your own form with double opt-in so the mailbox is a genuine subscriber. Our guide to sign-up forms and double opt-in describes the flow.
4. Engage in a human way. Open some messages, click some links and reply to some. Do not automate heavy, uniform engagement, which looks artificial and can distort the picture.
5. Keep notes on each account. Record the provider, age, settings, and any filters or rules. A change in an account itself can explain a change in results.
6. Respect privacy and terms. Use mailboxes you own or have explicit permission to use. Do not use other peoples accounts without consent, and follow each provider terms of service.

Keep the set small, perhaps 3 to 5 mailboxes at each major provider, and treat each one as an individual observation rather than a data point in a percentage. A result such as 4 of 5 is a prompt to look at which one differed and why.

Gmail tabs, Promotions and what they mean

Gmail sorts many messages into tabs such as Primary, Promotions, Social and Updates. A marketing newsletter in Promotions is not a filtering failure. It is how many recipients expect commercial mail to appear, and plenty of subscribers read it there. What matters is whether your messages reach the places where your recipients will find them.

• Tab placement is personal. The same message can reach Primary for one recipient and Promotions for another, depending on how each uses the mailbox.
• Postmaster Tools does not report tab placement. We observe it only on our own seed accounts and indirectly through engagement.
• Moving a message between tabs and replying to it are signals the user gives, and they can influence how future mail from that sender is sorted.
• Account notices, receipts and password messages belong in the stream described in separating transactional and marketing streams, and they should not look like promotions.
• Do not chase the Primary tab with tricks. Content that imitates personal mail while being bulk marketing tends to hurt trust and complaint rates.

Our practical rule is to track tab placement on seeds as context, then judge success by clicks, replies and conversions from Gmail recipients. If Gmail engagement is healthy, the Promotions tab is not a problem we need to solve.

First-party signals: Postmaster data, complaints and engagement

Seeds show a few mailboxes. The signals below come from your real traffic and real recipients, and for that reason they carry more weight. We set up the Google tool first, as described in Google Postmaster Tools: setup and how to read your Gmail reputation.

Signal	Where we get it	What a change may mean
Gmail spam rate and domain reputation	Google Postmaster Tools	A rising spam rate or falling reputation points to complaints or poor list quality at Gmail
Authentication pass rate	Postmaster Tools, DMARC aggregate reports	A drop suggests a DNS or sending-path change that may push mail to spam
Complaint rate by provider	Feedback loops and your delivery events	A rise means recipients are marking mail as spam, often before placement tests show it
Bounce and deferral reasons	Your delivery logs	Throttling or policy blocks show a provider reacting to volume or reputation
Clicks and replies by recipient domain	Your analytics and inbound replies	A domain-specific fall hints at placement trouble at that provider
Unsubscribe rate	Your unsubscribe handling	A spike can mean the audience no longer expects the mail

For details on measuring complaints see spam complaint rate and feedback loops, and for a wider set of metrics see email deliverability metrics: what to measure. For DMARC visibility, DMARC aggregate reports show whether authentication is passing across all sources.

Designing a fair placement test

A test only helps when it isolates one change. Most bad conclusions come from tests where two or three things changed at once, or where timing differed. We keep the design deliberately dull.

1. Same content. To compare infrastructure or timing, hold the content constant. To compare content, hold everything else constant.
2. Same time. Send the test and the control within the same few minutes. Reputation and filter state shift during the day.
3. Same path. Use the same sending domain, IP pool, authentication and stream you use for production mail, so the result describes your real path.
4. Include a control. Send a known good message, such as your last stable template, to the same seeds. If both drop together, the cause is probably reputation or provider-wide. If only the new one drops, suspect the content.
5. Repeat. Run at least two or three sends before trusting a difference. One result is an anecdote.
6. Record everything. Keep the subject, template version, send time, stream, authentication results and any list or volume changes alongside the results.

Test log: template comparison (invented example)
Date: 2026-10-14   Stream: marketing   Sending domain: mail.example.com
Send time: both messages within 4 minutes of each other

Message A (control, last stable template)
  Provider 1: inbox 4, promotions 1, spam 0, missing 0
  Provider 2: inbox 4, promotions 0, spam 1, missing 0
  Auth results in headers: spf=pass dkim=pass dmarc=pass

Message B (new template with large hero image and 3 links)
  Provider 1: inbox 3, promotions 2, spam 0, missing 0
  Provider 2: inbox 1, promotions 0, spam 3, missing 1
  Auth results in headers: spf=pass dkim=pass dmarc=pass

Note: Provider 2 differs between A and B. Repeat on two more sends before changing the template.

Authentication passing on both messages tells us the difference is not an SPF or DKIM fault. If you need to verify these results in the raw message, our guide to reading email headers shows where to look.

Worked example: reading mixed signals

Suppose we send a weekly newsletter to 60,000 subscribers, and the audience is mostly at two large providers. One Tuesday our seed check shows a change. We then look at the other signals before deciding.

Signal	Previous 4 weeks	This Tuesday
Seed inbox placement, Provider 1 (5 mailboxes)	5, 5, 4, 5	2
Seed inbox placement, Provider 2 (5 mailboxes)	4, 5, 5, 5	5
Control message, Provider 1	5, 5, 5, 5	5
Complaint rate at Provider 1	0.05 percent	0.06 percent
Clicks at Provider 1 (per 1,000 delivered)	14 to 16	15
Delivery rate overall	98.8 percent	98.9 percent

The first reaction might be alarm: only 2 of 5 seeds at Provider 1 placed in the inbox. But the control message to the same seeds placed 5 of 5, complaint rate is flat, and clicks from real recipients are normal. Provider 2 looks fine. The delivery rate is unchanged, which is expected, since it says nothing about folders.

Our reading: the new newsletter content may be triggering filtering on those particular seed accounts, but real recipients are not showing harm. We do not make a drastic change. We repeat the test with the same content next send, review the template for new elements such as a different link domain or a larger image, and keep watching clicks. If the next two sends show the same seed drop and clicks begin to slip, we act on the content. If seeds recover, we log it as noise.

Now imagine the opposite pattern. Seeds look fine, but complaint rate at Provider 1 doubles to 0.12 percent over two sends and clicks fall by a third. Here the real-user signals are what matters, and we would act, probably by tightening the audience to recent engagers, even though the seed test was clean. This is why we weigh real-user data above seeds.

When to act and when to wait

Placement data is noisy, so the main discipline is deciding in advance what counts as a signal. We suggest simple rules, written down and agreed by the team, rather than judging each result in the moment.

Pattern	Our response
One seed result changes, everything else is stable	Log it, retest on the next send, take no action
Seeds drop for new content only, control is stable	Review content, links and images, test variants against the control
Seeds and control both drop at one provider	Check authentication, blocklists, complaint data and recent volume or list changes for that provider
Seeds, complaints and engagement all worsen together	Treat as a real reputation problem: reduce volume to engaged recipients, pause risky segments, investigate the cause
Seeds are fine but real engagement or complaints worsen	Trust the real-user data and act, since seed accounts may not reflect your audience
Messages missing everywhere	Check for rejections, deferrals or authentication failures in logs before assuming filtering

When the trouble is real, the walkthrough in why are my emails going to spam is the next step. Change one thing at a time and give each change a couple of sends before judging it. Fixing five things at once makes it impossible to know which one worked.

An ongoing monitoring routine

Good monitoring is boring and regular. It combines a light weekly review with a few extra checks around risky moments, and it records what was seen so patterns become visible over time.

Weekly

• Review Postmaster Tools for spam rate, reputation and authentication.
• Review complaint rate, bounce reasons and deferrals by provider.
• Review clicks and replies by recipient domain, ignoring opens as a standalone measure.
• Run the seed set with the control message and note the result in the log.

Before launches and changes

• Run a design-controlled seed test before large campaigns, new templates, new domains or infrastructure changes.
• Check authentication and alignment on the exact stream you will use.
• Plan ramp-up for volume increases, as covered in peak season email volume planning.

Monthly

• Review the log. Did the routine catch the problems that mattered, and were any alerts false alarms?
• Refresh the seed set: check the accounts still work, still have a normal history and still match your audience mix.
• Adjust thresholds for action based on what you learned.

Placement monitoring checklist

• Delivered, bounced and inboxed are reported as separate numbers.
• Seed results are described as observations about test mailboxes, not as the true inbox rate.
• A small seed set of real mailboxes exists for each major provider in the audience.
• Seed mailboxes have history, are used normally, and are owned by us or used with permission.
• Every test includes a control message and uses the same content, time and sending path.
• Authentication results from the headers are recorded with each test.
• Postmaster Tools is set up and reviewed weekly.
• Complaint rate, bounce reasons and clicks are tracked by recipient domain.
• Thresholds for action are written down, and no change is made on a single result.
• A test log records what was sent, when, and what changed.
• The routine is reviewed monthly and the seed set is refreshed.

Further reading and sources

• Google: Email sender guidelines
• Google: Email sender guidelines FAQ
• Google Postmaster Tools help
• RFC 5321: Simple Mail Transfer Protocol
• RFC 7489: DMARC
• RFC 5965: An Extensible Format for Email Feedback Reports