Features
Part of Affiliate income for 2027: disclosures, real reviews and a launch plan that holds
Reading affiliate income metrics without fooling yourself
Affiliate income metrics: what clicks and commissions can support, a per-page ledger that makes comparison fair, and the decisions a number should trigger.
Every figure in an affiliate report is a lower bound on something you cannot observe. The chain between a recommendation and a recorded commission breaks in half a dozen ordinary ways, and none of them leave a trace in your dashboard.
That does not make measurement pointless. It makes it a different exercise from the one people think they are doing, which is why so much time gets spent optimizing against numbers that cannot support the weight put on them.
What to take away
- Your recorded results undercount your influence, always, by an amount nobody can tell you.
- Compare your own pages against each other over stable periods. Comparing against anyone else's published figures answers nothing.
- Let a number trigger a specific investigation, not a mood. Write the triggers down before you look.
What the numbers can and cannot support
What it can tell you
- Clicks
- That readers found the recommendation and acted on it
- Recorded sales
- That some purchases were attributed to you
- Conversion from click to sale
- A great deal about the merchant's checkout, pricing and availability
- Earnings per click
- A basis for comparing two programs sending traffic from the same page
- Reversal rate
- That buyers returned things, which is a content signal before a revenue one
- Any period total
- Direction, over several periods
What it cannot
- Clicks
- Whether they intended to buy, or whether the tag survived the next step
- Recorded sales
- How many you influenced that were credited elsewhere or not tracked at all
- Conversion from click to sale
- Much about your writing, which is mostly upstream of it
- Earnings per click
- A verdict on a page, since it moves with the merchant's rate and mix
- Reversal rate
- Which specific recommendation misled them, unless you look
- Any period total
- Anything at all, over one period
Two habits follow from this table. Attribute changes to the merchant before attributing them to yourself, because most of the mechanism sits at their end. And never read a single period, because holds and reversals mean the current one is provisional by construction.
What each metric can support
The figure
- Clicks
- Readers acted
- Recorded sales
- Some attributed
- Conversion
- Merchant checkout
- Earnings per click
- Compare same-page programs
- Reversal rate
- Content signal
- Period total
- Direction over periods
Can tell you
- Clicks
- Intent or tag survival
- Recorded sales
- Influenced but untracked
- Conversion
- Your writing
- Earnings per click
- Verdict on a page
- Reversal rate
- Which rec misled
- Period total
- Anything in one period
Cannot tell you
- Clicks
- Recorded sales
- Conversion
- Earnings per click
- Reversal rate
- Period total
The ledger that makes comparison possible
Program dashboards are organized around programs. Your decisions are about pages. Keep your own record, one row per page, updated on the same schedule.
Per-page ledger columns
- Page and decision it helps
- Programs it links to
- Readers, clicks, sales, reversals
- Hours spent since publication
- Last verification date
- One line on what to change
Keep the program's own statements beside this record rather than only the totals you copied out of them, because the statements are the evidence behind every figure you will later report. That is the reasoning in the Internal Revenue Service's recordkeeping guidance, and it costs nothing to follow.
The hours column is the one nobody keeps and the one that changes decisions. A page that earns steadily but needs rework every quarter can be a worse use of your time than one that earns less and needs nothing, and you cannot see that anywhere else.
Running a fair comparison
Most comparisons in this business are unfair in ways that are easy to fix.
- Use the same period boundaries every time,and know whether a sale belongs to the click date or the payment date. Pick one, write it at the top of the ledger, and never change it.
- Compare pages that sit at the same kind of moment.A page someone reads while deciding and a page they read out of interest are not comparable, however similar the subjects.
- Hold the traffic source constant.The same page performs differently for readers arriving from a search than from your own audience, and mixing them hides both.
- Know your noise floor.With a small number of sales, ordinary variation looks exactly like a result. If you cannot say roughly how much a figure moves in a normal period, you cannot tell a signal from a wobble.
- Do not compare across categories.Different products, prices and buying cycles produce different numbers for reasons that have nothing to do with the writing.
Why no published figure answers your question
The question is always "is this good". Published averages, rate lists and shared screenshots feel like the answer and are not, for reasons worth understanding.
The figures are self reported by people with an interest in the impression they create. Who reports is not random, and the disappointing results are the quiet ones. The underlying arrangements differ in every respect that matters: category, price, attribution window, exclusions, hold period, traffic mix and the merchant's own checkout.
Almost none of them state whether the number is before or after reversals, which is the difference between a figure and a guess. What a study should disclose about itself is not a matter of opinion. The American Association for Public Opinion Research's professional standards set it out, and a shared screenshot meets none of it.
The problem's general form, and why it applies to every published creator income figure, is in income benchmarks. The practical answer matches the rest of this work: your own series, stable definitions, long enough for a trend.
That argument is in career metrics. It applies here with more force because the external comparison set is unusually bad.
Decisions a number should trigger
Write these down before you look, so that the reading produces an action rather than a feeling.
- Plenty of clicks, very few sales, over several periods.Look at the merchant, not the page: checkout, availability in your readers' countries, price changes, and whether the products you feature are excluded from the program.
- A reversal rate that stands out.Treat it as a content problem first. People returning things means the page set the wrong expectation, and that is worth more attention than the lost commission.
- Almost no clicks on a well read page.The recommendation is probably buried, or there was no decision on that page to begin with.
- A large share of income from one page or one merchant.A concentration finding, not a success. Decide whether you are comfortable with what happens if it stops.
- Income holding up on a page you have not checked in a year.Straight into the audit queue, because it is still recommending something in your name. The twelve checks in the affiliate audit are the fix.
- Hours climbing on a page whose earnings are flat.Consider retiring it. Maintenance load is the real limit on this income line.
Cadence
Monthly, add the rows and note the direction. Do not act.
Quarterly, run the comparisons and apply the triggers. This is the meeting with yourself where things actually change.
Annually, ask the structural questions: concentration, total maintenance load, and whether this line still deserves the share of effort it is getting. That last question belongs with the wider decision about where commissions sit against everything else, which is what the affiliate framework is for.
Common questions
Should I use a tracking tool to see more of the chain?
It can consolidate reporting and let you change a destination in one place, which is genuinely useful. It cannot recover what was never tracked, and no tool will tell you about the reader who bought on a different device three weeks later.
Is a page with no recorded sales a failure?
Not necessarily. It may sit before the decision rather than at it, and it may be doing real work that shows up on another page. Judge it on whether it answers the question well, then check whether the recommendation is even findable.
How long before a page's numbers mean anything?
Long enough for the hold period to close and for several periods to accumulate. Anything read sooner is provisional, and any conclusion drawn from two periods is a story.
What single number should I watch?
Reversals, relative to your own history. It is the closest thing to a measure of whether your recommendations are actually right, and it is the one that tells you about the reader rather than about the merchant. The full account of why the chain loses data on the way to you sits in the guide to affiliate income.
