The short version
- Enrichment fetches and analyzes the actual article, not the search snippet.
- Candidates below a discovery score of 45 are never read, and one pass processes a bounded batch — the quality floor and the cost ceiling are both fixed, not negotiable per run.
- Each read produces a structured record: page type, article-likeness, language, publication date, HTTP status, content quality, relevance, placement suitability, author, existing mentions of your company, and publisher classification.
- Relevance and placement suitability are scored separately, each with a written reason, because a page can be perfectly on-topic with nowhere to add you.
- Existing mentions and links to your company are counted, which is what later prevents you pitching a page that already links to you.
Discovery produces a few hundred URLs with titles and snippets attached. That is enough to guess with and not nearly enough to decide with. A title like “The 12 Best Tools for Remote Teams” could be a carefully maintained editorial roundup, a two-year-old affiliate page that has not been touched since, or a listicle that quietly closed to submissions in 2024. Only one of those is worth an email.
Enrichment is the phase that finds out.
Two gates before anything is fetched
Reading a page costs real money — a fetch, an extraction, and a language model call per candidate. So enrichment applies two independent limits before it starts, in this order:
- 01
A fixed quality floor
Candidates below a discovery score of 45 are not enriched at all. This is a fixed constant rather than a per-run setting, so a thin run cannot quietly lower the bar to look productive.
- 02
A deterministic cost ceiling
What clears the floor is ordered by eligibility and capped. The cap bounds spend per pass; the ordering makes sure the pages that do get read are the ones with the strongest evidence behind them.
Eligibility ordering is where discovery evidence pays off. A candidate earns bonuses for being found by several distinct queries, for being found across several different strategies, for being found by a competitor query, for sitting on an article-shaped URL path, and for ranking near the top of the results it appeared in. None of these are large individually. Together they reliably float the pages worth reading above the pages that merely matched a keyword.
What a read page produces
The output of enrichment is not prose. It is a structured record, because the next phase has to make deterministic decisions from it:
| Field | What it answers |
|---|---|
| Page type | Is this a tool roundup, comparison, alternatives page, review, guide, tutorial, resource list, directory, case study, research piece, news item, opinion column, company blog article, agency article, landing page, or category page? |
| Article-like | Is this editorial content at all, or a page that merely reads like one in search results? |
| Content quality | Acceptable, thin, spam, low quality, or in a language the pipeline does not support. |
| Content relevance + reason | How close is this page to your market, and why — stated in a sentence you can disagree with. |
| Placement suitability + reason | Is there a natural place on this page to add a company like yours, and where. |
| Author | Who wrote it, when that is identifiable from the page. |
| Existing mentions | Whether your company is already mentioned or linked, and how many times. |
| Publication date, HTTP status, language, canonical URL | Whether the page is alive, current, and the one you think it is. |
| Publisher | Who runs the site, what kind of operation it is, and whether it is plausibly reachable. |
47 opportunities for acme.com
Ranked by priority
Best AI visibility platforms for SaaS
growthlab.example · Comparison roundup
0
- Directly compares products in your category
- Three competitors are listed while your company is absent
- The roundup structure supports a new entry without an edit war
Who to email
Sarah Jones · Author · Senior editor
sarah@growthlab.exampleVerified
Example data. Every field shown comes with its source in the product.
Relevance and placement are not the same question
This is the distinction that most link tools flatten, and flattening it is why so many prospect lists feel plausible and convert badly.
Content relevance asks whether the page is about your world. Placement suitability asks whether this specific page has somewhere a company like yours could be added without the addition feeling forced.
A well-researched industry retrospective can score extremely high on relevance and near zero on placement: it is about your market, it is read by your buyers, and there is nowhere in it that a product mention belongs. Pitching it wastes your credibility and the editor's time. Conversely, a modest guide with an explicit “tools we recommend” section may be only moderately relevant and an excellent placement.
Scoring them separately means you can see which one a page failed, and qualification can require a minimum on both rather than letting a strong relevance score paper over a placement of zero.
Relevance tells you the page is about your world. Placement tells you there is a door. You need both.
Counting what is already there
Enrichment records whether your company is already mentioned on the page, whether it is already linked, and how often. This sounds minor and is not.
A prominent existing link is normally a reason not to pitch — you already have the placement, and writing to ask for one makes you look like you did not read the page. It is one of the hard rejections in qualification. But the page is not deleted: it stays visible with that context, because “you are already here” is genuinely useful information, and occasionally a passing mention is worth converting into a proper link.
Publisher analysis, resolved once
Every enriched page belongs to a publisher, and several candidates often belong to the same one. Analyzing that publisher once per domain and reusing the result — rather than re-deriving it for every article — is what keeps a run over forty articles from ten sites affordable. The details are in publisher intelligence.