A backlink gap report tells you who links to your competitors and not to you. It doesn't tell you which of those pages an AI assistant actually pulls from when it answers a question.
TL;DR: Miloš Radić, writing for Search Engine Land, built a small local tool that overlays backlink gap data with AI citation data to pick outreach targets. His first campaign drew strong engagement, but he's still measuring how many replies turned into placements.
🔗 Backlink Gaps Miss the Pages AI Keeps Citing
A backlink gap analysis finds sites that link to your competitors but not to you. It's a foundational tactic, and it still works. What it can miss is a second group of targets: third-party pages that show up again and again as sources in AI-generated answers.
Radić ranks outreach targets by how often a page appears as an AI citation, instead of by Domain Rating (DR), a score for a site's backlink strength. DR tells you how strong a site is. Citation count is closer to what you want if the goal is getting mentioned where AI looks for answers.
He's upfront that this is a heuristic. He isn't claiming a mention on a cited page causes AI visibility, and he's still measuring whether it improves outreach results.
He sorts every prospect into one of three groups:
Both
Links to your competitors and gets cited by AI. This is the top priority.
AI-only
Shows up as an AI source but isn't a typical link target. Traditional backlink audits miss these completely.
Backlink-only
The classic link building play. It links to your rivals but doesn't show up in AI citations.
His example is ZenPilot's ClickUp vs Wrike article. It appears in the AI answers he monitors and mentions competing products, but not Productive. A page like that could get skipped on a list sorted by DR, yet its citation frequency makes it a strong target.
🧪 Two Dead Ends Came Before the Tool That Worked
Radić first tried Make.com, connecting Ahrefs, Google Sheets, and Lemlist. Merging the data, ranking prospects on custom signals, and finding contacts with fallbacks got clunky in a visual builder. He spent more time tinkering with the tool than on SEO strategy.
Next he tried live Ahrefs API queries. His first competitor intersection returned 311,000 domains, and refreshing that much data live used up a lot of API units.
What worked was a small local app. He filtered the 311,000 down to about 39,000 in Ahrefs and exported a CSV. Then he cut it to about 24,000 domains on his own machine, using filters like DR 40+, 500+ monthly organic visits, English-language sites, and spam patterns.
Because the sorting and filtering happen locally, they don't cost extra Ahrefs API units.
📬 The Early Numbers Look Good. Placements Are the Real Test
His first campaign targeted 760 prospects. It saw a 58.6% open rate, a 15.3% reply rate, and a 31.9% LinkedIn invitation acceptance rate.
Those numbers show people are engaging. They don't show how many replies became links or mentions, and Radić says that's still being measured. Treat this as a promising method, not a proven one.
The app itself is one dashboard: an opportunities board grouped into the three tiers, contact discovery, outreach tracking through Lemlist, and reply alerts sent to Slack. Each alert includes the exact article he's trying to get into, so he has context right away.
Citation coverage is a real limit. The results depend on the specific AI prompts he monitors, so a different set of prompts could surface different pages. I covered how to build a prompt set worth trusting in September.
🧰 A One-Person Workbench Is Not a SaaS Product
Radić works at a SaaS company, and he says building a product in a week is harder than it sounds. His tool works because it's one user on one laptop. He still reviews targets and approves messages himself.
A real product needs login, permissions, data isolation, databases, backups, and reliability. A filter bug that wrongly marked live backlinks as dead cost him an annoying afternoon. In a commercial product, he says, that same bug would be a support problem that erodes trust.
🛠️ What to Do With This
Rank Your Targets by Citations, Not Just DR
If you already track AI answers, check which third-party pages get cited most often for your category. Then compare that list to your competitor backlink gap. Pages that show up on both are where to start.
Spend AI Tokens on the Last Mile
Radić doesn't generate thousands of pitches up front. He writes one personalized pitch for one real, actionable target, which keeps AI costs low. He focuses on roughly 200 of the most actionable prospects instead of all 24,000.
Build Small, Then Add Guardrails
He didn't ask AI to build an SEO tool. He asked for small pieces, like a local app that loads a CSV into a sortable board, then added a second data source. When his contact finder guessed email addresses that didn't exist, one added sentence telling it never to invent an email address fixed the problem.
He also keeps API keys out of the code, using a single launch script that never gets committed.
Hit reply
Pull up your last backlink gap report. How many of those pages do you know are cited by AI answers in your category, and how would you find out?
Did You Know? One of the outreach sequences Radić describes runs four touches over eight days, mixing LinkedIn and email.
💭 One thing I'm thinking about
The most useful part of this article is what it admits it doesn't know yet. A 15.3% reply rate means people are answering, but it isn't a link, and Radić says so.
A new signal is easy to get excited about because it gives you a new way to sort a long list. Sorting is the cheap part. I'd treat AI citation counts like any early metric: a reason to look closer, not proof.
If you try this, track placements from day one. That's the number that will tell you whether the sorting was any good.
Till next time,
