Key takeaways
Core updates do not punish sites. They re-sort them, and comparison content sits where the sorting is harshest.
The sites that held had one thing in common: they would be useful without the commission.
Losses concentrate on templated pages, not on the whole domain.
Recovery is a rewrite, not a retune — and the work has to start before the next rollout, not after it.
What moved
A core update is not a penalty. It is a re-sorting, applied over a week or two, and the sites that fall have usually not done anything new. They were being carried by signals the update stopped weighting as heavily, and comparison content — which is what most affiliate sites are — sits exactly where that re-sorting bites.
Across the affiliate portfolios we track, the movement was not evenly spread. Three things were consistent.
The loss landed on templates, not on domains. Sites did not lose uniformly. One page type — most often the operator review, sometimes the “best X for Y” list — fell while the rest of the domain held. That pattern is useful, because it points straight at the cause: something about how that template is built, repeated a hundred times.
Volatility ran for longer than the rollout. Positions moved, partially recovered, then moved again over the following fortnight. Reading results on day three and acting on them is how sites end up making changes that were not needed.
The winners were not better optimised. The pages that replaced the losers were frequently worse on the measures an audit tool reports — slower, fewer keywords in the heading, thinner internal linking. They were better on things no tool scores.
That last point is where every useful conclusion comes from, so it is worth separating out.
The pattern behind it
The sites that held through the rollout had one thing in common. Stated plainly, it is this: they would still be useful if the commission disappeared tomorrow.
That sounds like a slogan. It is not — it decomposes into four things that are all visible on the page.
Evidence of actual use
The surviving reviews showed that somebody had opened an account. Deposit and withdrawal timings with dates on them. Screenshots from the product rather than from the operator’s press kit. A note about what the verification flow was actually like. Two things the operator does badly.
The pages that fell described products in the operator’s own language, which means the only thing they added to the web was a link.
A reason the page exists beyond the keyword
Where a site had twenty near-identical pages covering overlapping queries, the update usually kept one and dropped the rest. Topical authority is not a count of pages on a subject; thin pages on adjacent terms compete with each other and dilute the thing they were meant to build.
Disclosure that is not hidden
Commercial relationships stated plainly, near the recommendation rather than in a footer. This has no direct ranking mechanism, and it correlates anyway — because a site willing to say how it makes money is usually a site doing the other things on this list too. It is the clearest single marker of the E-E-A-T signals the guidelines describe.
A link profile that does not need explaining
Sites with a large share of bought placements on low-traffic domains fared worse, and the ones carrying a visible footprint across a cluster of sellers fared worst of all. Links were not the trigger — but when a template loses its content advantage, a weak profile is what it falls back on, and there is nothing there.
None of these four is a trick that stopped working. They are the same four things that have described a good comparison site for a decade. What changes with each update is how much slack there is for sites that skip them.
What to do now
In the order it pays back.
First, separate the loss from the noise. Which URLs, which templates, which query types — compared against the same period last year, not against last month. Betting and gaming traffic is seasonal enough that a September drop can be a fixture list rather than an algorithm. This is a day’s work and it prevents a quarter of wasted rebuilding.
Second, fix the template, not the pages. If one page type fell, the problem is in how that type is produced. Rewriting forty reviews by hand is slower, more expensive, and leaves the same hole for next time. Change what the template requires — tested on, dated, two criticisms, disclosure above the fold — and then rebuild against it.
Third, merge what competes. Near-duplicate pages that split the same intent should become one page that answers it properly. The traffic does not divide; it concentrates, and usually on a total above what the pieces held separately.
Fourth, deal with the links you would not defend. Not a mass disavow, which does more damage than good in most cases. A specific list: rented placements that will vanish when the invoices stop, anything in a visible cluster, anything on a page selling eight links to eight industries. Replacing them is the piece on rented links and the seven checks.
Fifth, rebuild authority somewhere real. Coverage on publications that an update has no reason to discount. It is slower than buying placements and it is the part that holds through the next rollout — and increasingly it is also what gets a site named by answer engines, which are reading the same signals for a different purpose.
Two things about timing. Recovery is rarely visible before the next rollout, because the re-sorting that would recognise the work has not happened yet; sites we have rebuilt typically see the change land one or two updates later. And the work does not start after an update. A site that is already doing these four things treats a rollout as a week of noise rather than an event.
If the portfolio is large enough that this is a programme rather than a project, the sequence we run it in is on the affiliates page.
