Unlinked Brand Mentions → Outreach: A Weekly Link Reclamation Pipeline With Claude

A B2B SaaS (Software as a Service, 软件即服务) client came to me in March with a Domain Rating of 38 and a "we just need more backlinks" problem. I ran an Ahrefs Content Explorer query for their brand name and got 1,400 results. Of those, 917 mentioned them by name and didn't link back. Each one was a near-zero-effort backlink someone had effectively pre-approved by writing about the brand at all. We converted 184 of them into live links in 90 days. I didn't write a single new article to make it happen.

This is the pipeline. It's one afternoon a week, plus a follow-up Friday.

The math that makes the boring work worth it

Before the workflow, the math. Most "link building" agencies sell you DR (Domain Rating, Ahrefs 的 0–100 域名权威分) 50+ guest posts for $400 a pop. The math on this pipeline is different because the cost of each acquired link is roughly 5 minutes of human time, not a piece of content.

The numbers I see on a clean run:

• 200 unlinked mentions / week surfaced by Ahrefs Content Explorer
• ~160 actually unique, non-spam, DR>25 mentions after filtering
• ~32% reply rate on a polite, specific ask (this is high; the industry average for cold link outreach is 4–8%)
• ~55% of replies convert to a live link
• = ~28 new backlinks / week → ~115 / month

It will not be 115. Realistic floor: 18–25% of polite asks get the link. Realistic ceiling: 30%. Either way, this single pipeline produces more referring domains than most sites get from a year of guest posting, and it costs less than one piece of freelance content.

Step 1: Pull unlinked mentions once a week

Two tools do this cleanly. Ahrefs Content Explorer if you already pay for Ahrefs. Mention.com or Brand24 if you don't. Both let you run a "brand name" query and filter for pages that contain the brand string but do not link to your domain.

The Ahrefs filter combo I run every Monday morning:

• "Title or content contains": BrandName (with the actual brand, no quotes)
• "Referring domains" → exclude: yourdomain.com (you don't want to find your own site)
• "DR filter" → 25+ (cut the bottom of the internet; anything under DR 25 won't move your DR and will burn your time on replies that never come)
• "Live / recently seen" → last 90 days (older mentions rarely get edited; editors don't go back to 2023 listicles)
• "Highlight unlinked": ON — Ahrefs' built-in filter that excludes URLs already linking to you

Export as CSV. The first time I ran this for that B2B client it returned 1,400 historical mentions. The weekly refresh, after I cleaned up the historical backlog, returns 150–250. That is the working list.

Step 2: Hand the CSV to Claude — once

This is the trick. Don't write 200 personalized emails. Don't even segment the CSV by hand. Paste the whole weekly export into a single Claude session with a prompt that does three things in one pass: classifies sentiment, classifies context, and drafts the outreach.

The prompt (paste as a single message, attach the CSV):

You are an SEO (Search Engine Optimization, 搜索引擎优化) outreach analyst. I'm giving you a CSV of unlinked brand mentions (DR ≥ 25, last 90 days). For each row, output a CSV with these columns: source_url, source_domain, dr, mention_sentiment, mention_context, suggested_ask, draft_email.

• mention_sentiment: one of positive, neutral, negative. Positive = the article praises or recommends us. Neutral = we are mentioned as one of several examples, or a statistic. Negative = the article is critical or compares us unfavorably.
• mention_context: one of review, listicles, news, comment_or_forum, comparison. (Examples: G2/G2-style review pages → review; "best X tools for Y" articles → listicles; product launches / funding news → news; Reddit / Hacker News / Indie Hackers → comment_or_forum; "X vs Y" pieces → comparison.)
• suggested_ask: short label for what I'd be asking the editor to add (e.g. "add link to our product page", "add link to our original research", "add link to founder bio").
• draft_email: a 3-sentence email. Sentence 1 references a specific thing in the article (not "great piece"). Sentence 2 makes the ask. Sentence 3 closes. No "I hope this finds you well." No exclamation marks. Under 70 words total. Sign off as "Alex" (placeholder — change to your name). Each draft must be visibly tailored to that specific row. Do not repeat the same opener across rows.

Important: for negative rows, set draft_email to "SKIP" — we don't pitch editors who wrote negatively about us.

That last instruction is the important one. The pipeline is about reclaiming good faith mentions. Negative reviews are not the place to ask for a link.

Claude processes 160 rows in about 4 minutes. The output is a CSV with the sentiment, context, and a draft for every row except the negative ones. I paste the whole thing into a Google Sheet.

Step 3: Send via Apollo, Hunter, or by hand — depends on volume

The sending tier depends on the size of the list. Three rules I follow:

• Apollo for the bulk of it. Apollo's free tier finds work emails for ~30% of the domains I export. Above that, you pay. The CSV imports cleanly; you upload source_url and dr as custom fields; the draft email goes in the body field.
• Hunter.io as a fallback. The "find email by domain" feature gets you to the editor or author for the rest. $49/mo gets you 500 searches.
• Manual send for any review context row on a site with DR ≥ 70. The reason: G2-style review pages, Capterra-style comparisons, and high-DR listicles often have a real human editor's name and email visible on the page. A real Hi Sarah, … outperforms a templated find-anywhere pitch by 2x. I send those 10–15 a week by hand. They take an extra 3 minutes each. Worth it.

Don't blast the whole CSV from Apollo on Monday morning. Send in two waves: 80 on Monday, 80 on Wednesday. Sending patterns matter more than people think; a domain that sends 160 cold pitches in 4 hours looks like spam to Gmail and to the recipients.

Step 4: The follow-up (most of the wins live here)

Reply rate on a polite 3-sentence ask is high — but conversion to an actual link is separate. The "I'll get to it" reply is the most common positive response. Most people never follow up on it.

For every positive-but-stalled reply, send one follow-up 7 days later. Two sentences max. The follow-up is also drafted by Claude from the same CSV context — give it the original row plus the editor's reply and ask for a 2-sentence nudge. Then stop. Two touches is the cap. After that, the editor isn't going to do it, and a third email makes you look needy.

This single follow-up step doubles the link conversion in my pipeline. The math goes from ~28 backlinks/week to ~50 in the weeks where the follow-ups pay off.

What this doesn't fix

• Spammy aggregators and listicles you'll never reach. About 30% of the CSV is sites with no contact, no editor name, and an obvious "auto-generated" feel. Skip them. Don't waste Apollo credits.
• Brand names that are also English words. If your brand is "Flux" or "Loop" or "Beacon," Content Explorer returns tens of thousands of false positives. You need a tighter query: "Flux" -flux.circuits.io style negative matches, or you need Brand24's NLP (Natural Language Processing, 自然语言处理) filter to exclude non-brand usage. Budget 2 hours the first week to clean the query.
• Editors who simply don't add links. Some sites cite you as a "source" deliberately without linking — it's a stylistic choice. Reasonable humans disagree on whether you should ask. I ask once, and if they say no, I never re-pitch that domain. Life's too short.

The real win, stated plainly

Link reclamation at scale is one of the few SEO tactics that compounds. A new guest post gets you one link. This pipeline gets you 30–50 a week, every week, for as long as editors keep writing about your brand. The first month is the setup tax. From month two on, it's an afternoon a week and a steady increase in referring domains that you can show up to a client review and say "we added 117 referring domains this quarter, here are the URLs."

That's a line item most SEO agencies can't produce. The work is unglamorous. The math is not.