The Backlink Gap Play: Find Every Link Your Competitors Have and You Don't

A backlink gap analysis finds the websites that link to your competitors but not to you. Those domains already link to sites in your space, which makes them the warmest link targets you have. Below is the exact process I use to run one in about 45 minutes with Claude Cowork, plus everything I promised in the pack: the setup, the instruction file, my scoring template with worked examples, and the four outreach angles I actually send.

I ran this recently and it returned 41 links worth chasing. That number is what survived my manual pass of the top 60. A few were dead sites or paid placements the model could not detect, so they got cut. That part stays human, and I will show you where it fits.

What you get in this guide:

• A step-by-step Claude Cowork setup: which exports to pull, how to organise them, and what to click
• The exact instruction file I drop in with the exports (copy-paste)
• My scoring spreadsheet template, with worked examples
• Four outreach frameworks, one for each link type

What is a backlink gap analysis?

A backlink gap analysis (sometimes called a link intersect) compares the referring domains of two or more competitors against your own. The output is a list of sites that link to several of your competitors but have never linked to you.

The logic is simple. If a site links to three companies that do what you do, it has already shown it is willing to link to your category. You are not cold pitching. You are asking a site to do again what it has already done, this time with you in the mix.

Why run this in Claude Cowork (and what it cannot do)

Claude Cowork is the Claude desktop app built for non-developers. You drag files in, type what you want in plain language, and it works through the data. No code, no formulas to write. For a job that is really just cross-referencing a few large spreadsheets and reasoning about each row, that fits well.

Here is the honest boundary, because it shapes the whole workflow:

• Cowork reasons over the files you give it. It does not browse the live web unless you connect web tools, so it cannot confirm a page is still online or spot a paid or sponsored link on its own.
• It gives you a ranked, classified starting point in minutes. It does not give you a final send list. You do.
• It outputs a table you can paste into a sheet, and .docx files if you want a written brief. It does not write to Google Docs.

So Cowork does the six hours of sorting, matching, and first-pass judgement. You do the 20-minute quality check at the top of the list. That split is the whole point.

Part 1: The Claude Cowork setup guide

Step 1 - Pull the right exports

You need referring-domain exports, not backlink exports. Referring domains give you one row per linking site, which is what you want to compare. Backlink exports list every single link and will bury you in duplicates.

Pick three competitors who compete for the same readers or buyers as you, not just anyone in the industry. Then export the referring domains for each, plus your own:

• In Ahrefs: Site Explorer, then Referring domains, then Export
• In Semrush: Backlink Analytics, then Referring Domains, then Export
• In Moz: Link Explorer, then Linking Domains, then Export

Export four files in total: competitor-a.csv, competitor-b.csv, competitor-c.csv, and my-site.csv. The fourth file is the one most people forget. Without your own referring domains, Cowork has no way to know which links you already have, so it cannot tell you what is missing. Your file is the subtraction step.

Keep the default columns. You mainly need the referring domain and a strength score (DR, DA, or the tool's equivalent). If anchor text or target URL columns come through, keep them too. They help Cowork guess link type and fit.

Step 2 - Set up a clean folder and Project

Make one folder on your desktop and put all four CSVs in it. Name them clearly, because Cowork refers to them by name. A messy set of export(3).csv files will slow you down.

In Cowork, start a new Project for this analysis. A Project keeps the files and the instructions together, so you can rerun it next quarter by swapping in fresh exports. Give it a name like Backlink Gap - Q3.

Step 3 - Drop the files and the instruction file in

Upload all four CSVs into the Project. Then paste the instruction file from Part 2 as your first message. That single message tells Cowork what each file is, what to compare, how to score, and what to hand back. Send it, and let it work.

Step 4 - Read the output, then rerun once

Cowork returns a ranked table. Read the top rows and sanity-check the logic. If the ranking looks off, ask it to explain how it scored two or three rows. Nine times out of ten you will spot one instruction to tighten (for example, weighting topical fit even harder), fix it, and rerun. Then move to the scoring sheet and the human check.

Part 2: The exact instruction file

This is the message I paste into Cowork with the four exports attached. Copy it as is and change the file names to match yours.

You are my link-building analyst. I have uploaded four referring-domain exports:
- competitor-a.csv, competitor-b.csv, competitor-c.csv (my three competitors)
- my-site.csv (my own referring domains)

Do this:
1. Build one combined list of every referring domain across the three competitor files.
2. For each domain, count how many of the three competitors it links to (1, 2, or 3).
3. Remove any domain that also appears in my-site.csv. I already have those links.
4. Keep only domains that link to 2 or more competitors.
5. For each remaining domain, add these fields:
   - Referring domain
   - Number of competitors it links to
   - DR or DA from the export
   - Topical fit, scored 1 to 5, based on the domain name and any anchor or URL text in the files
   - Likely link type: editorial, guest post, broken link, or digital PR
   - The single page on my site it would most likely link to
   - One line on why that site would care
6. Score each row: Priority = (topical fit x 3) + (authority x 2), where authority is 1 to 5 mapped from DR (70+ = 5, 50 to 69 = 4, 30 to 49 = 3, 15 to 29 = 2, under 15 = 1).
7. Sort by Priority, highest first.
8. Output a table I can paste into a spreadsheet, and flag the 10 you would contact first.

Rules:
- Do not invent domains that are not in the files.
- If a field is unknown, write "unknown". Do not guess a number.
- You cannot see live pages, so label link type as "likely" and note that I will verify each one.

Why it is built this way: step 3 is the subtraction that makes it a gap analysis. Step 6 puts topical fit ahead of raw authority on purpose, because a smaller site that speaks to your exact reader is worth more than a big site that does not. The rules stop the model from filling blank cells with confident guesses, which is the failure mode that wastes your time later.

If you run this often, save the instruction file as a custom skill in Cowork so you can trigger it in one line instead of pasting the whole thing each time.

Part 3: The scoring spreadsheet template

Cowork gives you a first-pass score. The sheet is where you make the final call and track outreach. Here are the columns:

The scoring model

Priority = (Topical fit x 3) + (Authority x 2), out of 25.

Fit is scored 1 to 5, where 5 is a site that writes for your exact audience. Authority is 1 to 5, mapped from DR so it stays on the same scale. Fit carries more weight because a link from a site your readers actually visit does more for you than a high-DR link from an unrelated page. When two rows tie, break the tie with the number of competitors linked (more is a stronger category signal) and then with link type (broken-link and editorial wins are usually faster than guest posts).

Worked examples

Read it top down. founderstack.io and the-cmo-brief.com tie at 23, so the tiebreak (links to 3 vs 2) puts founderstack first. martechweekly.com has the highest DR in the list but drops to third because its fit is a notch lower, which is the model working as intended. bizdirectory.biz scores 7 and reads like a paid directory, so it gets marked skip during the human check. That last row is exactly the kind of thing the model cannot rule out on its own.

Part 4: The four outreach frameworks

One angle does not fit every link. Match the framework to the link type in your sheet.

1. Editorial

When it fits: the site publishes articles in your space and links out to sources inside them.

The angle: you are a better or more current source for something they already write about. You are not asking for a favour, you are improving their piece.

Template:

Subject: a fresher source for your [topic] piece

Hi [name],

I was reading your piece on [topic] and noticed you link out to sources on [subtopic]. I recently published [your page], which covers [specific angle or fresh data]. If it is useful, it might be a good fit alongside what you already link to there.

Either way, good piece. [Specific detail you liked.]

[You]

2. Guest post

When it fits: the site runs contributed articles and links to several competitors from author bylines or body copy.

The angle: you can write something their readers want that they do not have yet. Lead with the idea, not the ask.

Template:

Subject: guest piece idea for [site]

Hi [name],

I write about [area] and read [site] often. One gap I noticed: you have covered [topic A] and [topic B], but not [specific idea]. I could write that for you, drawing on [your credibility in one line].

Here are two angles: [angle 1] or [angle 2]. Happy to send an outline if either fits.

[You]

3. Broken link

When it fits: the site links to a resource that is now dead, and you have a live page that covers the same ground.

The angle: you are doing them a small favour first. Point out the broken link, then offer yours as a replacement.

Template:

Subject: broken link on your [page name] page

Hi [name],

Quick heads up: on [URL], the link to [dead resource] no longer works (it returns a [404 / error]).

If you want a live replacement, [your page] covers the same thing. Only if it fits, of course. Either way I wanted to flag the dead link.

[You]

4. Digital PR

When it fits: the site covers news, trends, or data in your space, and links to studies or expert takes.

The angle: give them something citable. Original data or a sharp expert quote, not a plea for coverage.

Template:

Subject: data on [topic] you can cite

Hi [name],

You cover [beat] closely. We pulled together [original data or finding], and one number stood out: [specific stat].

Happy to share the full breakdown if it is useful for a piece, or to give a short quote on [angle]. No pressure either way.

[You]

The human check (do not skip this)

Before you send anything, work down the top 40 to 60 rows by hand. This is the 20 minutes that turned my list into 41 real targets. Three things to look for:

1. Is the page still live? Cowork works from an export that may be weeks old. Open the linking page. If it is gone, cut the row.
2. Is the link paid? Look for sponsored labels, a "partners" directory, or a page that links to fifty companies in your category. If it reads like a paid placement, skip it. You do not want it, and you cannot usually buy your way in cleanly anyway.
3. Does the fit hold up? The model guesses fit from a domain name. Sometimes it is wrong. A 10-second look at the site confirms or kills it.

Whatever survives that pass is your outreach list. Everything above it was the machine doing the heavy sorting so you only spend judgement where it counts.

Backlink Scoring Template

Grab the template:

Backlink-Gap-Scoring-Template.xlsx

Google Docs

FAQ

What is a backlink gap analysis? A method that finds websites linking to your competitors but not to you, so you can target link sources that already link to sites like yours.

Can Claude Cowork do backlink analysis? Yes. Cowork reads the referring-domain CSV exports you drop in and cross-references them. It cannot browse live pages unless you connect web tools, so it works from the data you give it.

Which export should I pull? Referring domains, not backlinks. You want one row per linking site. Export three competitors plus your own domain.

What can AI not detect here? Whether a page is still live, and whether a link is paid or sponsored. Both need a human check before you reach out.

How long does it take? About 45 minutes in Cowork, plus a manual pass of the top results.

Want the workflows I run every week, written up like this one? Subscribe to The AI Driven Marketer newsletter, and follow along as I turn slow marketing jobs into short ones.