Anchor text is the part of a link programme that is either specified in advance or assembled by accident, and the accident is expensive to undo. Nobody notices the profile drifting while it drifts. They notice eighteen months later, when a page that should rank does not and the only unusual thing about it is that forty percent of its inbound anchors are the exact term it is trying to rank for.
The fix is unglamorous: decide the distribution before the first pitch, write it into the brief, and report against it every month like any other budget.
The five anchor types, and what each is for
Branded. Your company name, with or without a suffix. This is what a publication uses when it mentions you naturally, which is why it dominates any profile that was earned rather than bought. It carries little topical signal and enormous credibility signal.
URL or naked. The address itself, or the domain. Common in reference lists, methodology notes and citations. Like branded, it is what real citation looks like.
Partial match. A phrase containing part of your target term inside a natural sentence — "their approach to inventory forecasting" when you are targeting inventory forecasting software. This is where most of the useful ranking signal lives, and it is the type a good writer produces without being told to.
Exact match. The target term and nothing else. The strongest signal per link and the one that looks manufactured fastest, because publications rarely write this way unprompted.
Generic. "Read more", "this study", "here". Contributes nothing topically and quite a lot to looking normal.
Proportions worth planning to
These are the bands we specify at kickoff for a B2B SaaS programme. They are not a law, and any individual profile can sit outside them for good reasons — but a profile that sits far outside them without a reason is usually one that was bought to a formula.
Branded: 35–45%. The single largest block. If yours is much lower, the profile does not look like a company being mentioned; it looks like a company being linked to on purpose.
URL and naked: 15–25%. Rises naturally if you publish research or tools, because citation-style linking uses addresses.
Partial match: 20–30%. The working block. This is what you actually spend the editorial effort on, because it does the ranking work while reading as prose.
Exact match: 2–5%. Deliberately scarce. Not because a threshold exists somewhere that trips a penalty — nobody outside Google knows where that is — but because scarcity is what earned profiles look like, and looking earned is the point.
Generic: 10–15%. Free camouflage that costs nothing to accept.
Why exact-match stays scarce
The argument for loading exact-match anchors is that they are the strongest signal available. That is true, and it is the reason to be careful with them rather than the reason to use them.
Consider how a real citation gets written. An editor links because a sentence needs support. The anchor is whatever words happen to be there — usually the company name, sometimes a descriptive phrase, occasionally an address. Almost never the precise commercial term the company would most like to rank for, because that phrasing serves the vendor rather than the reader.
So a profile heavy in exact-match anchors is describing its own origin. It is not that the pattern is forbidden; it is that the pattern is informative, and what it informs on is that somebody chose the words.
The second reason is practical. Exact-match anchors are the ones a publisher is most likely to change, strip or nofollow later during an editorial cleanup, which makes them the least durable thing to have spent money on.
Auditing the profile you already have
Before specifying a target distribution, measure the current one. Pull all inbound anchors to your target page, classify each into the five types, and calculate the percentages. An hour of work, and it changes the plan more often than not.
Three findings are common enough to expect.
The profile is already exact-match heavy from a previous programme. Then the next twelve months should be deliberately branded-and-generic heavy to dilute it, which means the new links will not look like they are "doing anything" on a report. This has to be explained at kickoff or it will be argued about in month four.
The profile is almost entirely branded, with no partial match at all. Common in companies that have never done link building and have been mentioned in the press. The gap here is not volume; it is that nothing in the profile tells a search engine what the page is about.
The anchors are fine but they point at the wrong URL. This one is worth checking before spending anything. If your documentation subdomain or an old blog post is absorbing the anchors meant for the product page, that is an internal linking and architecture problem, and no amount of outside acquisition fixes it. We ended an engagement over exactly this: seven months of good placements, all reinforcing a URL that could not convert.
Reporting against the model
The mechanism that keeps this honest is boring. A table in the monthly report with two columns: target percentage and actual to date. Five rows. Drift shows up immediately, which is the entire purpose.
One refinement worth making: report the distribution of the last three months separately from the all-time profile. All-time drifts slowly and hides a bad quarter. Trailing three months is where you see a supplier starting to lean on exact-match because rankings were flat and somebody asked why.
The other thing worth writing into the contract is who decides the anchor. If the agency proposes it and the client approves it, drift is a conversation. If the agency simply reports what it got, drift is a discovery — and by then it is twelve months of profile that has to be diluted rather than corrected.
The version that fits on a page
Audit what you have. Set the twelve-month distribution before the first pitch. Keep branded and URL anchors as the majority, put the editorial effort into partial-match phrasing that reads like prose, keep exact-match under about five percent, and take the generic anchors for free.
Then report actual against target every month and treat a drift the way you would treat a budget overrun: as something to explain in the month it happens, rather than something to discover in the annual review.