A link gap analysis is the least glamorous document in this discipline and the only one that turns an opinion into a plan. Done properly it takes about two hours, works with any backlink tool including the free tiers, and ends with something an outreach team can act on the same week.
Done badly — which is most of the time — it produces a spreadsheet of nine thousand backlinks that nobody opens twice. The difference is entirely in what you filter for.
Start with one URL, not a domain
Every mistake in gap analysis descends from analysing the wrong unit. You are not competing with a company. You are competing for a position on one search result, and the thing occupying that position is a URL.
So the first step is to name the page. One page, with a commercial job — the category term, the comparison page, the alternative-to page. Then open the search result and take the four URLs above it. Those four are your comparison set for the whole exercise.
Export the referring domains for each of those four URLs at page level. Every tool has this view; it is usually called "referring domains" inside the URL-level report rather than the domain-level one. Export yours too.
You now have five lists. The work from here is set arithmetic followed by judgement, in that order.
The subtraction
Combine the four competitor lists and de-duplicate. Then remove every domain that already links to your page. What remains is the raw gap: domains that link to at least one competitor and not to you.
Two refinements make the list much more useful before you go any further.
Count how many competitors each domain links to. A publication that links to three of the four is describing the category rather than covering one vendor. Those are the highest-value targets on the list and they belong at the top. A domain linking to exactly one competitor is often a relationship rather than a category signal, and it is harder to convert.
Drop the structural noise. Directories every vendor is in, review platforms, the competitor's own properties, syndication mirrors. These inflate the count and none of them are outreach targets. In a typical B2B SaaS export this removes thirty to fifty percent of the raw list, which is why raw gap numbers are so misleading.
What survives is usually between forty and a hundred and fifty domains. That is a target list you can read in one sitting.
Five filters worth applying, in this order
Now the judgement. Apply these in sequence, because each one is cheaper than the next and killing a candidate early saves the cost of evaluating it properly.
Page-level traffic, not domain traffic. Does the kind of page you would appear on get read? A site with eighty thousand monthly visitors can publish you on a page that gets eleven. This filter alone removes a large share of what a metrics-first list would have kept.
Topical overlap. Does the publication already cover your category, measured rather than asserted? In B2B this is not a tiebreaker; it is most of the value. A trade title read by four thousand of the right people beats a general business site read by four hundred thousand of the wrong ones.
Publishing history. Output that predates any outreach programme. A site that started publishing in the same quarter it started accepting contributions is inventory, not a publication.
A findable author. Named, with a history elsewhere. Anonymous bylines that appear only on one site are the cheapest signal available and one of the most reliable.
Reachability. This is the filter almost nobody applies and the one that decides whether the list is real. Does a named editor exist, do they publish outside contributors, and has anyone had a reply in the last year? A publication can be perfect on every metric and completely closed. Metrics tell you what a placement would be worth; reachability tells you whether it is available.
Scoring what survives
Resist the urge to build a weighted composite score with eleven inputs. It looks rigorous and it hides the tradeoffs. Three bands are enough:
Band A — links to three or four competitors, real page traffic, strong topical overlap, reachable. Pitch these first and pitch them properly. There will be fewer than you want, usually eight to twenty.
Band B — links to one or two competitors, passes the filters, reachable. This is the working body of the programme and where most placements come from.
Band C — passes the filters but reachability is unproven. Worth attempting when Band B thins out, not before.
Anything that fails a filter does not go in a band. It goes in a rejection log with the reason code, because next quarter somebody will suggest it again and the log is what stops the same argument happening twice.
What the finished analysis tells you
Three things, and all three are decisions rather than data.
The size of the job. Bands A and B added together, compared against the gap number, tells you whether the list can even close it. If your gap is ninety and the reachable list is thirty-one deep, the honest conclusion is that outreach alone will not get you there and something has to be built that earns links without being asked for.
The order of work. Band A first is not obvious to everyone — plenty of programmes start with easy wins to show early numbers. That is a defensible choice for morale and a bad one for outcomes, because the hardest targets take longest to convert and starting them late compresses everything.
Whether to start at all. If the analysis shows your page already carries more referring domains than the median above it, you do not have a link problem. You have a page problem or an intent-match problem, and no volume of acquisition will fix either. This is the finding that saves the most money and it appears in roughly a third of the analyses we run.
The two-hour version
If you want to do this yourself before talking to anyone: name the page, take the four URLs above it, export page-level referring domains for all five, subtract, count how many competitors each remaining domain links to, drop the directories and mirrors, then spend most of your time on the reachability check for the top twenty.
That last part is where the hours actually go, and it is the part that separates a target list from a wish list. Everything before it is arithmetic a tool can do. Everything after it is the programme.