How backlink opportunity discovery actually works

Link building has a research problem long before it has an outreach problem. Finding pages is trivial — any search engine will give you thousands. Deciding which of those pages your company could plausibly be added to, why an editor would agree, and who to write to is the part that eats a week and produces a spreadsheet nobody trusts.

Dice is built around that research problem. This post walks through the whole pipeline: what happens when you click Find Opportunities, what each phase costs, and why the order matters.

The shape of the problem

Three facts drive the entire design.

• Search is cheap, reading is expensive. A search API call returns fifty results for a fraction of a cent. Fetching one of those pages, extracting its content, and having a language model judge it costs orders of magnitude more. Any pipeline that reads everything it finds is a pipeline that either runs out of money or lowers its standards.
• Domain-level metrics do not answer page-level questions. A site with excellent authority scores publishes pages your product has no business appearing on. The question is never “is this a good site” — it is “would adding us to this exact page make the page better?”
• The reasons matter more than the ranking. A list sorted by a number you cannot interrogate is a list you will second-guess on every row. If the tool cannot say why a page ranked 92, you cannot decide whether to trust it.

A pipeline that reads every page it finds is a pipeline that either runs out of money or lowers its standards. So the order of operations is the product.

Phase 1: Discovery

Discovery starts from your own site. Dice builds a market profile — what you sell, the problems you solve, the terms your audience uses, the adjacent topics you legitimately belong to — and recommends competitors, which you confirm before anything is searched. That profile becomes a set of search queries across several deliberate strategies: category searches, resource and roundup searches, competitor searches, and long-tail searches that reach the smaller publishers a keyword tool never surfaces.

Results are normalized and deduplicated by URL, filtered for the obvious non-candidates — your own domain, file downloads, login pages, search result pages, product listings — and scored. Crucially, every surviving candidate keeps its discovery evidence: which query found it, what type of query that was, and where it ranked. A page found by four different queries across three different strategies is a stronger candidate than one that appeared once at position 40, and the evidence is what makes that visible instead of merely implied.

Phase 1 output is typically a few hundred candidate URLs. None of them have been read yet.

Phase 2: Enrichment

Enrichment is where pages actually get read, so it is where cost control lives. Two gates run before a single page is fetched: a fixed quality floor (a candidate below a discovery score of 45 is not worth reading) and a cost ceiling on how many candidates one pass will process. Between them, they turn an unbounded pool into a bounded, deterministic batch.

What comes back from a read page is not a summary. It is a structured record: the page type, whether it is genuinely article-like, its language, its publication date, its content quality status, a content relevance score with a written reason, a placement suitability score with a written reason, the author if one is identifiable, whether your company is already mentioned or linked and how many times, and an assessment of the publisher behind it.

Phase 3: Qualification

Qualification turns enriched pages into opportunities, and it does two different things in a strict order.

1. 01

   Hard checks first

   Spam, thin content, dead pages, unsupported languages, failed analysis, a prominent existing link to your company, content relevance below 60, and placement suitability below 60 are all disqualifying. These are not weighted inputs — they are gates. A page that fails one is rejected no matter how well it scores elsewhere.

2. 02

   Then a weighted score

   What survives is scored on five weighted components: placement suitability, content relevance, page quality and linkability, audience and commercial intent, and competitor gap. The result is the Opportunity Score, and it lands the page in a tier — exceptional, strong, good, or possible.

Qualified opportunities are persisted with their full evidence trail. So are the rejections, with their reason codes — because "we found nothing" and "we found sixty pages and rejected all of them for these reasons" are very different answers, and only one of them is useful.

After qualification: contacts and ranking

Contact discovery runs only after Phase 3, and only for domains that produced a qualified opportunity. This is a cost decision as much as a design one: looking up contacts for every raw search result would spend provider credits on pages that were never going to be pitched.

Once contacts exist, Dice computes a Reachability Score — how confidently you can actually reach someone at this publisher — and combines it with the Opportunity Score into a Priority Score that ranks every opportunity for the site. Editorial quality carries most of the weight; reachability breaks the ties. An outstanding page you cannot contact still outranks a mediocre page with a verified editor's address.

What the pipeline deliberately does not do

• It does not fill a quota. There is no target number of opportunities. A broad market produces many; a narrow one produces a handful. Lowering the bar to reach a round number would make the list longer and the work worse.
• It does not hide rejections. Not-selected pages stay visible with their score and rejection context, and you can promote one manually when you see an angle the model missed.
• It does not send anything on its own. Outreach is a separate, explicit step: you approve drafts, and they send from your own mailbox.
• It does not treat popularity as quality. Small company blogs and independent niche publishers are included when they are legitimate and relevant. Traffic is not the criterion.

Where to go next

Each phase has its own post with the specifics: how discovery queries are planned, what enrichment extracts from a page, and how the Opportunity Score is calculated. If you want the outreach half of the product instead, start with contact discovery.