How to choose anchor text for bought links

• Branded: the company or site name. The bulk of any natural profile.
• Topical (partial match): a phrase about the subject that is not the exact target keyword. “their guide to margin loans” for a page targeting “margin loan rates”.
• Exact match: the target keyword verbatim. Present in natural profiles, rare.
• Generic: “here”, “this article”, “read more”.
• Naked URL: the address itself.

There is no universal ratio, and anyone quoting one to two decimal places is guessing. The shape that recurs in profiles that rank is: branded and naked URL together the majority, topical a substantial minority, generic present, exact match a small slice. The direction of the mistake is always the same, too much exact match, so the safe default leans further toward branded and topical than feels efficient.

The better question is not “what mix?” but “what mix given what the page already has?” A page whose existing links are 40% exact match needs the next ten to contain none; a brand-new page with no links can afford a topical anchor or two early.

An order takes an anchor plan, a distribution across those classes, rather than a single anchor per line. Campaign memory records the anchors on every order for a target, classifies them, and reports the running mix, so a follow-up tranche rebalances rather than repeats. The target profile call classifies the page’s existing anchors from the backlink index when you give it your brand and keyword, so the plan starts from the real baseline rather than from zero.

"anchor_plan": { "branded": 0.4, "topical": 0.3, "generic": 0.15, "naked_url": 0.1, "exact": 0.05 }

Same anchor, same landing page, many new referring domains inside a short window, on publishers that link out to far more domains than link in. Each element is individually defensible; together they describe a paid campaign in a way an algorithm can learn. Vary the anchor, spread the tranche across weeks with a standing order, and let the quality tests keep the saturated publishers out.

Less than it did, more than zero. The observable effect of exact-match anchors has shrunk with each core update while the penalty for over-optimisation has not; the asymmetry is the whole argument for a conservative mix. We report rank movement after placement as observed, never caused, and anchor class is one of the fields that will be in the outcome data once there is enough of it to say anything.