Anchor text: natural distribution vs over-optimization

When I pulled the anchor text profile of 400 domains in the last audit I ran at Lucas S.A., one pattern shouted: sites penalized in 2025 had, on average, 38% exact match anchors pointing to money pages. The ones that grew had 6%. That is not coincidence, and it is not 2012 Penguin crawling out of the grave. The algorithm got better at reading intent in linkjuice, and concentrated exact match became a nail-head flag. Before touching any off-page strategy, it pays to recalibrate what looks natural in 2026.

Healthy distribution, based on what I see across dashboards: 40-55% branded ("Lucas SA", "lucas.com.br"), 20-30% naked URL or junk ("click here", "read more"), 15-25% partial match or topical, and only 3-8% exact match. When you invert that, Google does not need manual review, just an embeddings classifier comparing your anchor cloud against the niche median. I cross those numbers with How to audit a competitor's backlink profile for benchmark: if your top 3 competitors run 7% exact and you sit at 31%, you are the outlier, not the smart one.

The classic mistake comes from poorly coordinated Honest link building: what replaces guest posts in 2026 campaigns. Three outsourced writers publish guest posts in the same month, all anchored "best CRM for startups" because the brief said so. Within 60 days the cluster shows as an anomaly in Ahrefs. The fix is not to stop targeting the keyword, it is to diversify the anchor pool per campaign: set quotas (max 1 exact every 8 links), share a tracking sheet with the writers. Anyone running Editorial partnerships: the scalable off-page model knows anchor control belongs in the contract, not the footer.

Topical anchors ("CRM tool", "sales automation") are the warm middle zone few exploit, and that is where the gains live. They carry semantic context without the heavy-handed signal of exact match. In tests I ran with A/B testing in SEO: methodology that resists noise as method, pages receiving 70% partial+topical climbed 2.3 median positions in 90 days, against 0.8 for the exact-heavy group. Google uses BERT to understand "platform that organizes pipeline" equals "CRM" intent, so you keep relevance and gain safety at the same time.

There are four operational traps I see often. First: 301 redirects carry the original anchor, so consolidating old domains can inject unwanted exact match, read 301 vs 302 Redirects: The Real Ranking Impact before migrating. Second: link reclamation tends to bring branded, great, but check in Link reclamation: recovering lost equity in 30 days that it is not concentrating on a single page. Third: digital PR generates plenty of naked URL, and that is good, Digital PR for SEO: how to measure the real ROI of mentions has the detail. Fourth: internal interlinking counts. If you repeat the same anchor across 80 internal pages, you built the artificial signal yourself.

To audit, I use BigQuery joining GSC export with Ahrefs API, clustering anchors by embedding similarity (sentence-transformers, paraphrase-multilingual model). That catches variations that look different but Google reads as the same concept: "best CRM 2026", "top CRM 2026", "#1 CRM 2026". If most of those clusters cross 15% of total, that is already bad. The base query lives in BigQuery + GSC: queries your agency won't run. For interlinking, run a Screaming Frog crawl and export inlinks by anchor, then pivot.

Practical takeaway: open your anchor profile today, split into four buckets (branded, naked, partial, exact), and compute the percentage per target page, not per domain. If any page sits above 10% exact incoming, freeze campaigns to that URL for 60 days and redirect new links to related cluster pages, using Topical authority: how to build clusters that rank as the guide. Natural anchor work is not about avoiding the keyword, it is about diluting it in context. Get that, and you stop fighting the algorithm and start riding it.