Start

Attribution tags

Tags are the unit of measurement in AffiliateOS. Every link belongs to one, every conversion comes back carrying one, and earnings are grouped by them. This page explains how the plumbing works across five networks with five different sub-ID conventions.

What a tag is

A tag is a slug you choose per campaign, placement, or conversation: yt-desc-2026-07, newsletter-roundup, chat-session-8341. Slugs are lowercase letters, digits, hyphens and underscores, up to 64 characters. Tags can carry a description and freeform metadata.

You rarely create tags explicitly: create_link accepts any tag name and creates it on first use.

How a tag round-trips

When a link is minted, AffiliateOS generates a unique sub-ID for it and maps that sub-ID to your tag. The sub-ID rides the network’s own tracking parameter, and when the conversion syncs back hours or days later, the sub-ID on the transaction is resolved back to the tag.

Network	Outbound parameter	Returns on transactions as
Awin	`clickref`	`clickRefs.clickRef`
Impact	`subId1`	`SubId1`
CJ	`sid`	`shopperId`
Rakuten	`u1`	`u1`
Amazon	`ascsubtag`	no earnings feed (see the Amazon guide)

One convention on your side, five on theirs. That translation is the point of the product.

Commission lifecycle

Every transaction moves through four statuses:

Status	Meaning
`pending`	reported by the network, not yet validated
`approved`	validated; will be paid in a future payout
`reversed`	declined or clawed back (returns, fraud, cancelled orders)
`paid`	included in a payout you received

Validation is slow in this industry: networks typically confirm commissions 30 to 90 days after the click, matching merchant return windows. Judge a tag by its approved and paid totals, and treat fresh pending numbers as a leading indicator only.

Reading earnings by tag

get_earnings_summary (or GET /v1/earnings/summary) groups by tag by default and returns pending, approved, reversed and paid totals per tag, currency-converted. Grouping by day, network or program answers the other common questions.

For agents, the useful pattern is a feedback loop: mint links under distinct tags per placement, wait for data, read the summary, and shift effort toward the tags that convert. The MCP reference shows the exact calls.