draft-chuang-feedback-id-header-00 landed on 16 November 2026: the first IETF draft for the complaint header Gmail accepts through Postmaster Tools. It is read right to left, the last field must never be empty, and DKIM has to cover it — three rules that silently delete data when senders get them wrong.
On 16 November 2026, Wei Chuang and Allen Robinson of Google published draft-chuang-feedback-id-header-00, Feedback-id Email Header Field. It is an individual Internet-Draft: no working group, no RFC stream, IESG state “I-D Exists,” expiring 20 May 2027. Nothing in it changes how any mailbox provider treats your mail this week.
What did change is that the header now has a definition. Feedback-ID has been accepted by Gmail’s Postmaster Tools for years, but the only description of it was a help-centre page written for senders. The draft turns that page into a specification with MUSTs in it, adds the privacy reason the mechanism exists, and states limits the help page leaves implicit — the four-field budget, the right-to-left evaluation, the signing order. That last set is where senders get it wrong, and the errors are quiet. The header stays legal; the dashboard just stays empty.
The draft opens on the problem Feedback-ID was built to solve. Traditional feedback loops generate a per-complaint message using the Abuse Reporting Format, and those reports “can expose individual recipient actions to senders.” A mailbox provider can instead compute aggregated complaint rates per identifier and surface only the ones that look bad. Gmail did exactly that first, per the draft’s own aside: the mechanism was “originally deployed by Gmail for its recipients via Postmaster Tools.”
That is the whole design. You tag your traffic, the receiver aggregates complaints against your tags, and you see rates rather than people. Every structural rule below follows from it — including the ones that frustrate marketers who wanted per-subscriber complaint data.
The header is Feedback-ID: a:b:c:SenderId. The three leading fields are OPTIONAL and hold up to three identifiers; SenderId is REQUIRED, and the draft recommends 5 to 15 characters that stay consistent across your mail stream.
§4.2 formalizes the rule Gmail senders have had to infer from the help page, and it makes it a MUST:
A receiver MUST evaluate the Feedback-ID header field for data aggregation by reading from the rightmost field to the left, aggregating data for a maximum of four Identifiers, beginning with the SenderId field. Any further fields to the left of the fourth field MUST be ignored for aggregation purposes.
draft-chuang-feedback-id-header-00, §4.2Two consequences hide in that sentence. A fifth field is not an error — it is invisible. If a relay appends its own identifier to the left of your four, nobody rejects the message and no data appears for that field. And an empty SenderId means the receiver MUST NOT generate aggregate data at all, while an empty optional field only costs you that field’s dimension.
The grammar in §3.1 is stricter than most templates. An identifier is one or more characters drawn from letters, digits, -, _, ., +, / and = — enough for structured tokens and appended hashes, not enough for Spring Sale 2026, spaces, commas or an @ sign. Leading optional fields may be omitted (b:c:SenderId, c:SenderId), and an intermediate field may be left empty (a::c:SenderId), but that empty field still occupies one of your four slots. Padding with :: buys nothing and costs a dimension.
Most lifecycle teams do not write this header. Their platform does, and the platform’s choices consume fields. Amazon SES is the clearest published case: fields a and b are yours via the ses:feedback-id-a and ses:feedback-id-b EmailTags, field c “is used by SES to identify the account and region used to send the emails” and “cannot be overridden by the user,” and the required field is not yours at all — “the SenderId part is always AmazonSES and cannot be overridden.”
So on SES, of the four dimensions the draft imagines — campaign, customer, mail type, sender — two are yours, one identifies the AWS account and region, and the sender dimension is a string every SES customer shares. AWS also limits senders to “a single verified header value per traffic stream,” which it spells out as the reason the header cannot carry a per-recipient value.
If you send through any other ESP, this is the question to put in writing: which of the four fields do you set, can I set the rest, and is the header added before you DKIM-sign the message? The answers decide whether your campaign taxonomy survives contact with the receiver.
§4.1 is the part that quietly kills FBL data for senders whose pipelines look fine in a header dump. The header MUST be added before signing and MUST be covered by the DKIM signature of a domain the sender owns or controls — meaning it appears in the h= list, not merely in the message. §5 repeats it as a receiver rule: a mailbox provider MUST ignore any Feedback-ID not covered by a valid DKIM signature from a domain under the sender’s control, and MUST restrict aggregated reporting to verified owners of that domain.
The failure mode is an ordering problem, not a configuration flag. Add the header in a milter, a gateway or a relay that runs after your DKIM signer, and h= will not name it; the receiver discards it and reports nothing. The message authenticates, delivers, and produces no complaint telemetry — which reads as “our complaint rate is fine” rather than “we are not being measured.”
The rest of §4.1 is infrastructure you should already have: a single verified header per message (SHOULD), the signing domain registered in the postmaster tool, sending IPs published in SPF (MUST), PTR records that resolve (MUST). Senders MAY oversign the header, which per RFC 6376 §8.15 stops a downstream hop from inserting its own.
Read §4.2’s reporting rules before building a dashboard on this. Reports are generated per identifier, independently, and only for identifiers that surface an unusual rate — Google’s wording is that identifiers “with an unusual spam rate and that might cause deliverability issues will be reported,” and that reports appear only when an identifier is present in a certain volume of mail and in distinct user spam reports, for @gmail.com recipients alone.
Three limits fall out of that. There is no denominator, so there is no baseline to trend. Small segments never appear, because a threshold on distinct complainers cannot be met by a few thousand sends. And per-subscriber attribution is not merely absent but prohibited by construction — which is why §4.2 forbids identifiers unique across every message, names Message-ID as the anti-pattern, and tells senders not to repeat a value across fields, since cross-field repetition merges unrelated identifiers into one bucket. Appending the hash of one identifier to another is the sanctioned way to join two dimensions.
Four things, each with a template consequence.
The draft carries no working-group endorsement, and it asks IANA for nothing yet: its §6 registration is explicitly a placeholder, “not yet approved or ready for IANA action.” Adoption is a receiver’s decision, and one receiver implementing a draft does not bind the next. If a working group takes it up, the MUSTs on senders still only bite where a provider chooses to enforce them.
The useful reading is narrower than the headline. For Gmail, nothing changed this month except that the rules you were already measured against are now written in normative language you can quote at your ESP. What would change the picture is a second provider adopting the same header: the four-field budget would stop being a Gmail quirk and become a shared constraint on how you name your mail.
Pull the raw source of one message your platform sent last week and read the Feedback-ID line field by field. Count the colons: more than three, and your leftmost tag is being discarded. Check the last field is not empty. Check the characters — no spaces, no @. Then find feedback-id in the DKIM-Signature h= list. If that last check fails, you have been sending without complaint telemetry for as long as the header has been in your template, and the fix belongs in a ticket to your ESP rather than in your campaign brief.
Written by the team at LMR — a senior lifecycle marketing studio working in Iterable, Braze, Klaviyo and the tools you already run.
Start a project ↗