Discovery use cases: five guides, five jobs

Name the job before the source list

Guest posts, insertions in existing articles, listicle and review inclusions, directory listings and AI-citation sources are different requests to different people, and they need different evidence. Pick one, write down the destination page it should send readers to, and only then open the export. A source list sorted by a domain metric cannot tell you which of the five jobs a page fits.

Reuse the inputs you already hold

Every guide starts from a file you can already produce: a supplier backlink export, a Search Console page-pair export, a competitor inventory you compared earlier, or a saved page. AgentLinkOps imports seven supplier layouts and a generic CSV through presets, and any other layout through an explicit column map. It previews without writing anything, and refuses a Search Console export that only counts links per site, because reading it as rows would invent placements.

Qualify with three words and no score

The qualification vocabulary is the one the product's campaign templates use: satisfied, not satisfied, or agent judgement. An observation decides the first two. Nothing observable decides the third, so it is handed to you with the evidence attached and no recommendation. A contact route counts only when the publisher explicitly published it, and the product never checks or implies that a message would arrive.

Let the record say what a check proved

A candidate you register starts as not checked. One explicit selection creates a paused watch and a single metered check. The observation carries a hash of the fetched bytes, the robots posture, the redirect chain and the HTTP status, with a standing note that JavaScript execution and visual visibility were not checked. A result that cannot conclude says unknown with a reason, and unknown never means the link was removed.

One pipeline after the job is chosen

Every guide ends at the same place: a source page and a destination you want checked. From there the stages are the same for all five jobs. Recall reads the candidates a run holds; a candidate starts as not checked. Verify admits 1 to 50 selected candidates under one workspace budget reservation and returns an outcome for every item, including the ones refused for budget, so exhaustion is visible instead of silent. Monitor converts a completed present verification into a placement watch that carries the run, candidate, job and observation it came from. A check never starts monitoring by itself, and an unknown observation never counts as a removal.

Read the recall, verify and monitor pipeline walkthrough for the command names, the per-item outcomes and a request your agent can run.

A practical example

Suppose a competitor export lists a tools roundup, a long tutorial on your subject, a software directory and a page that names your product without linking to it. Those are four jobs: a listicle inclusion, an insertion into an existing article, a directory listing and a reclamation request. The roundup needs a freshness date and an inclusion basis. The tutorial needs the paragraph that lacks a source. The directory needs a crawlable listing page and its link attributes. The mention needs a complete read proving that no link exists, because in a September 2026 scan of 140 pages, four of the seven pages that named a target already linked to it.

Use these guides when you know what kind of placement you want and need a repeatable way to judge candidates. Use the backlink discovery workflow while you are still building the candidate list, and the link-building workflow when a qualified page is ready for a handoff. A guide can tell you what evidence a request needs. It cannot promise that a publisher, a directory editor or an assistant will act on it. The opportunity library applies the same six types and three verdicts to whole niches: it counts the candidate surfaces an owned crawl found, states each one’s verification state and age, and lists nothing until a person publishes a record.

What you can use in AgentLinkOps

AgentLinkOps imports exports, verifies selected candidates one at a time or in a bounded batch of up to 50, converts a present result into a monitored placement with its lineage, monitors supplied placements and records evidence in the private pilot. Live supplier discovery is disabled in this build and stops with PROVIDER_NOT_CONFIGURED before anything is saved. Your agent does the reading, the judgement, the browser work and any authorized outreach with your own tools. The service sends nothing.

Read the tool guide and connection requirements before setting up a workflow, or browse the tool catalog for every registered tool with its state.