Why Affiliate Reviews Fail to Rank: Indexing, Intent, Depth, Links

Diagnose in the right order

When an affiliate review does not rank, almost everyone starts debugging at the wrong end. They rewrite the intro, add a table, chase backlinks, expand it by six hundred words. Usually it does not work, because the failure was three layers below where they started looking.

There is a correct order, and it is strictly sequential. Each layer makes the ones above it irrelevant.

• Layer 0 — Is the page indexed at all? If Google does not have the page, nothing about the page matters.
• Layer 1 — Does the page match the query's intent? If the SERP wants a different artifact, a better version of the wrong one does not help.
• Layer 2 — Is the page substantively competitive? Only meaningful once 0 and 1 are clear.
• Layer 3 — Do internal links tell Google what the page is? The last thing to tune, and the first thing most people tune.

Layer 0: the page is not indexed, and you did not notice

This is the failure mode nobody writes about, because it is embarrassing and invisible from inside your own analytics. A site with no traffic looks the same whether the cause is weak content or a crawl that stopped months ago.

We know this precisely, because it happened to us. On a site we run, every page carried a canonical tag pointing at a different hostname — a preview domain left over from the platform the site was built on. Every URL was, in effect, telling Google: I am a copy, the real version lives elsewhere. Google believed it, because that is what a canonical tag is for. It classified the site as a duplicate and stopped crawling. The result: three indexed pages against a site of more than a thousand URLs, and 35 impressions with zero clicks over twenty-eight days. No penalty, nothing anyone would call an SEO problem — one wrong attribute in the head of every page.

The most instructive part was not the canonical itself. It was one number in the coverage report: zero pages in "Crawled — currently not indexed." That bucket holds pages Google fetched, evaluated, and declined to index. Zero pages in it meant nothing had ever been rejected on quality. Meanwhile more than a hundred sampled URLs sat in "Discovered — currently not indexed": Google knew they existed, from a sitemap it had read without errors, and never fetched them.

That distinction is the whole diagnosis. "Crawled — currently not indexed" is a content verdict: Google read the page and passed, so now, and only now, do rewrites matter. "Discovered — currently not indexed" is a crawl verdict: Google never read it, so rewriting is pure waste.

Before touching a word of copy, check three things. The canonical tag on a live page as Google sees it — fetch the rendered page and read the tag, not your template source. Ours was wrong on every URL for months without one alert firing. Whether canonicals cross hostnames: staging domains, preview builds, platform subdomains, www versus apex. Any cross-host canonical claims another site owns your content. Last crawl date across twenty sampled URLs; if the newest is months old, you do not have a ranking problem.

The less dramatic version is crawl budget: publish hundreds of thin, near-identical pages and crawling gets slower site-wide. But do not delete pages reflexively. We nearly made that mistake, and the coverage report stopped us — if nothing was rejected on quality, deleting pages destroys good URLs and fixes nothing. Measure before you cut.

Layer 1: intent mismatch

Once the page is being crawled, the next question is whether it is the kind of page the query wants. This is where most affiliate reviews fail.

The classic version: a commercial-intent query answered with an informational article. Someone searching a product name plus "pricing" has a wallet open and one question. They receive two thousand words opening with the history of the category. The inverse happens just as often and is harder to spot — a research-stage query answered with an aggressive comparison that pushes a decision on a reader three weeks away from making one.

Diagnosing this takes ten minutes and no tools. Look at what actually ranks, not what should: if eight of the top ten are single-product reviews and you published a roundup, you have your answer. Check for community results, because forum threads mean the query wants lived experience, and a page assembled from vendor documentation cannot compete. Read the first hundred words of the top result for what it assumes about the reader — if it names a specific decision and yours defines a category, you are addressing different people.

Tooling shortens the mechanical half. Pulling the ranking set and extracting heading structures is what something like Surfer is built for, and doing it by hand wastes an afternoon. The read itself is not automatable: no coverage score tells you that a SERP full of forum threads wants lived experience rather than a well-structured document, and a tool that scores your page highly against that set is confidently measuring the wrong thing.

When there is a mismatch, the fix is rarely a rewrite. It is usually a second page: keep the informational article for the research-stage query it serves, publish a tighter decision-focused page for the commercial one, and link them. One document serving both intents serves neither — which is how most bloated affiliate reviews got bloated.

Layer 2: thin in the way that matters

"Thin" does not mean short. It means the page contains nothing you could not get from the vendor's own site in five minutes. A four-thousand-word page assembled from marketing copy is thin; a nine-hundred-word page containing one real test result is not. Comparison pages fail this more often than reviews.

• Sequential reviews wearing a comparison's clothes. Product A for eight paragraphs, then Product B, then a summary. Nothing is ever held against anything. A real comparison names a dimension and evaluates both products against it at once.
• No disqualification. The page never says who should not buy. This is the strongest signal of first-hand evaluation, and its absence is the clearest tell of its opposite.
• Unverifiable claims. "Users report", "widely considered" — attribution to a source that does not exist. The reader who has actually evaluated the product spots these instantly, and that is the reader you are trying to convert.

Layer 3: internal links

Internal linking is the layer people reach for first and understand least. Three failures dominate.

• Orphan pages. Published, sitting in the sitemap, with no inbound internal link from anywhere. Crawl your own site and list every URL with zero inbound links; that list is usually longer than expected.
• Flat architecture. Every page linked only from a paginated blog index, so related pages never reference each other. Build category hubs that link to their members, and have members link back and sideways.
• Generic anchors. "Read more", "click here". Anchor text describing the target's subject is one of the few signals you fully control.

The opportunity affiliate sites waste: comparison pages that name a product without linking to that product's own page. Decide internal links at brief time and the graph builds itself. One caution: internal linking amplifies whatever is already there. If Layer 0 is broken, a perfect link graph moves nothing between pages that do not exist as far as Google is concerned. This is why it belongs last.

The ten-minute triage

Before your next rewrite, run this in order and stop at the first failure. Is the exact URL indexed, and which bucket does it sit in? What does the rendered canonical say? When was it last crawled? What page type ranks for the query? If it is not what you published, the fix is a different page, not a longer one. Does the page contain one fact unavailable on the vendor's site? Does anything link to it?

The reason to work in this order is not tidiness. A failure at zero makes every measurement above it meaningless, and you will spend months producing careful work against feedback that is noise. We spent that time. The content was never the problem, and no amount of writing would have revealed it. One line in the head of every page was — and the only reason we found it is that we looked at the crawl data before the prose.