Affiliate Content After AI: What Still Deserves a Link (and What Doesn't)

Affiliate content did not die in 2026. The cheap version is the one a model can emit a hundred times: a “best X for Y” page with public specs, a stock photo, and a link. Readers can smell it. Thin or scaled pages may rank poorly, or violate spam policies, when they are created mainly to manipulate rankings. Google’s helpful-content page still draws a line between people-first pages and that kind of automation. Affiliate pages are not a separate banned class. Confirm any traffic story against your own Search Console rather than a third-party “X% of sites dropped” recap.

The living form is narrower and more expensive to produce: a comparison a reader can check, written by someone who used the thing, with dates on the numbers and a clear “not for you” line. That is also the only form this site would eventually put a link on. Istar is still in draft mode — 37 guide files as of 23 August 2026, 36 of them draft: true — and the pipeline in AI Content That Actually Passes Review exists so a future affiliate note sits under a page that already survived an independent fact pass.

Three content types, revalued

Hands-on comparison (still alive). Two or three named services, one job, numbers you measured or opened on an official page this month. Example: this site’s transcription pair did not ask “which is the best tool.” It asked when free, subscription, and per-minute human review each pay off, and it separated Rev AI from Rev’s human-reviewed product after an auditor caught the merge. Judgment test: if you removed every adjective, would a table of dated facts still make the decision?

Pure listicle (usually not worth a link). Ten logos, one paragraph each, no test, no “who should skip this,” no date on the price. Judgment test: could another site generate the same page from marketing copy in an afternoon? If yes, do not attach a tracking link. You are competing with the vendor’s own page and with every other model that read the same spec sheet.

Vertical depth (expensive, sometimes worth it). A single niche done properly — bookkeeping for a two-person studio, contract review before a freelancer signs, local models on a 32 GB Mac. Judgment test: did you do the work the reader is about to pay for, or sit with someone who did? Depth without experience is just a longer listicle.

The operational rule does not need a viral thread: if you cannot show the work, do not monetize the page.

Verifiability is the moat

Three habits do more than a better prompt.

Date every number. “$20 a month as of 23 August 2026, checked on the vendor pricing page” is a claim. “Affordable” is not. When the vendor moves the price — OpenAI’s Luna cut, DeepSeek’s pricing table as of 23 August 2026, Notion’s current AI help page — a dated sentence can be updated. An undated one becomes a quiet lie.

Test it yourself, or say you did not. A screenshot of your own usage meter beats a hero image of a laptop. If you only read the docs, write “docs-only, not a hands-on test.” Readers and raters are asked to look for experience. Pretending is worse than a short, honest limit.

Write the counterexample. Who should not buy this. When the cheaper model is enough (cost-per-task routing). When a free tier is a trap (free-tier dependency). A page that cannot name a loser is an ad.

If a link pays, say so near the link. For U.S. consumer audiences, the FTC endorsement guides ask for a clear disclosure. For Google Search, mark paid links with rel="sponsored". Check the rules that apply to your readers’ countries.

Who should run affiliate pages, and who should not

Fit. You already publish in a niche you practice, you can re-check prices on a calendar, and you are willing to recommend nothing when the honest answer is “skip it.” A freelancer who uses three tools a week and writes about those three. A small studio that will keep a page updated when a plan changes.

Poor fit. A new domain whose only plan is “100 AI reviews this month.” A personal brand built on being universally positive. A YMYL topic (health, credit, legal outcomes) where you have no credential and no first-hand process. Google’s quality materials put extra weight on trust for those topics; a model-written supplement stack with an affiliate link is the wrong bet.

Brand risk. Istar’s job is a clear desk, not a directory. A guide that already went through refine → validate → independent audit → repair on FAIL → revalidate can later carry a quiet, disclosed link. A page that failed the auditor on mixed denominators should not grow a “buy” button until the math is fixed. Affiliate revenue on a broken comparison is how a small site trains readers to leave.

A “should this page have a link?” audit

Take ten published or nearly published URLs. For each, score 1 or 0:

1. A named person or site process used the product, or the page clearly says it did not.
2. Every price or quota has a check date and a source URL.
3. There is a who-it’s-not-for paragraph.
4. The page would still help if every affiliate link 404’d.
5. An independent pass has cleared facts and logic (see the review system).
6. Updating this page when the vendor changes price is someone’s job, with a next review date.

This is Istar’s starting editorial heuristic, not a ranking law. Treat 5–6 as “a link is allowed,” 3–4 as “rewrite before monetizing,” and 0–2 as “do not attach a tracker.” Whether to noindex or merge a weak page is a separate call: look at search intent, overlap, and Search Console, not this score alone.

Then look at the set. If almost none of the pages clear a 5, you do not have an affiliate program yet. You have a content debt. Spend the next stretch raising a few pages over the line — original screenshots, dated cards, a counterexample — instead of generating a pile of new outlines.

AI made the cheap page free to produce and easy to detect. The click that remains is the one a reader can defend after they open the vendor’s own site. Earn that click with a comparison you would still publish if the link paid nothing. If you would not, do not hang a tracker on it.