agentlinkopsHow to qualify niche directories
Qualify a niche directory before you submit: editorial versus paid listings, crawlable listing pages, the link attributes they carry and a last-observed date.
A niche directory is a site that lists products, tools, services or organizations for one audience, with a page per listing and usually a link out. The qualification job is to tell an editorially maintained directory from a pay-for-inclusion one, check whether the listing pages can be crawled and indexed, read the link attributes a listing carries, judge the audience fit, and record a last-observed date so the record can decay honestly.
This guide is one of the discovery use cases. It also carries the eligibility rules and freshness cadence we apply to our own directory records, because those rules were the reason we chose to publish methods rather than a directory list.
Two kinds of directory, one policy line
Google's spam policies name "low-quality directory or bookmark site links" as an example of link spam. The policy does not say directories are spam. It says a directory that exists to sell or trade links, with no editorial standard, produces links Google treats as manipulation. A directory with a stated audience, an editor who rejects listings and a reason to exist beyond the links is a different thing.
The tell is usually in the submission page. "Free review within two weeks, or pay for a fast lane with a followed link" is a directory describing a paid link; Google's guidance would have that link marked sponsored. A directory that lets anyone list themselves is user-generated content, and Google recommends rel="ugc" on such links. Neither attribute is bad news for a listing that reaches the right readers. Both are facts your record should hold, and both come from the page, not from the directory's marketing.
Find candidates with the files you already have
A competitor export, grouped by host. Directory listings cluster: one host, many listing pages, the same template. A supplier export of a competitor's backlinks will show those clusters. AgentLinkOps imports seven supplier layouts and a generic CSV through presets, and any other layout through an explicit column map; keep the listing page URL for each row, because the directory home page tells you nothing about how a listing is marked.
A Search Console sample export. The Links report exports sample linking pages for one of your target pages. If you already have directory listings, those rows show which ones exist today and let you check them. The site-level count export is refused by the importer with its reason.
The directories in your own niche. For an agent-facing product the relevant directories are MCP server catalogs, API catalogs, CLI channels and awesome lists. On September 15, 2026 we researched 164 distinct candidates for AgentLinkOps itself. Nothing was submitted; the numbers below are what the research alone established.
Qualify with six eligibility gates
These are the gates a directory record must pass before we would publish it as an opportunity, from the feasibility memo that decided against a standalone library. They work as a submission checklist too.
| Gate | What passes | What fails |
|---|---|---|
| Owned observation | You fetched the listing page or the submission page and stored the date | A vendor's row, an inferred route |
| Redistribution-safe fields | Domain, platform class, observed submission route, observed link attributes, cost basis, last observed | Vendor metrics, contact data, private notes |
| Route evidence | The page that says submissions are accepted, saved with its date | A contact form taken as an invitation |
| Re-verified within the window | A platform-class record checked within a quarter; an editorial-route record within 90 days | An older record silently reused |
| Policy screen | Editorially neutral platforms with a stated audience | Pay-for-placement markets and link-seller directories |
| No contact enrichment | Deliverability not_checked by default | An address you did not see published |
The product's campaign templates return three verdicts and no score: satisfied, not_satisfied and agent_judgement. Gates one, three and four are observations. Gates two, five and six are judgement, and the software hands them over with the evidence attached. The contact rule applies as everywhere else: a route counts when the directory explicitly published it, and nothing checks or implies deliverability.
Two checks are specific to directories. First, can the listing page be crawled and indexed? Google's robots documentation says "a page that's disallowed in robots.txt can still be indexed if linked to from other sites", and its noindex documentation says the rule only works when the page "must not be blocked by a robots.txt file". So read both: the robots file for the listing path, and the listing page's meta robots or X-Robots-Tag. A listing on a noindex page still exists; it just reaches readers who arrive another way. Second, what rel tokens does an existing listing's outbound link carry? Open a competitor's listing and read the anchor. The verifier records per-occurrence raw rel tokens, so one selected check on a competitor's listing page against the competitor's site answers this for you.
What our own records showed
The feasibility memo examined a private registry of 658 directory-class and platform-class records that sibling projects had collected. Every row was status=new in this repository: zero verified, zero with a last-verified date, zero with a live listing URL. 373 rows were directory-class. 620 of 658 carried a third-party authority score of unstated vendor, method and date, which is why a public library may not carry vendor-metric columns. 602 of 658 carried operational notes that would need field-level sanitization before publication.
The costs were measured, and they are lopsided. Re-fetching all 658 rows costs about one cent and ten minutes at the measured fetch rate from an owned crawl. Reviewing them to publication grade costs an estimated 22 to 44 hours per quarter at two to four minutes of judgement per record. Fetch money is noise; review time is the price. That is the reason this page holds methods and gates rather than a list of directories.
The September 15, 2026 inventory for AgentLinkOps's own niche put the same gates to work on 164 candidates across MCP, API, CLI and general product directories:
| Measure | Count |
|---|---|
| Distinct candidates | 164 |
| Submission route fetched and confirmed | 140 |
| Route page exists but the form renders client-side | 2 |
| Route unverified because of a bot block, rate limit or origin error | 22 |
| Eligible now with no product-side blocker | 71 |
| Ineligible for the product | 13 |
| Ineligible while access is invitation-only | 4 |
Two limits from that record belong on this page. The 22 unverified routes carry a best-known URL and a note of what the fetch returned, and their fees and requirements come from search snippets or third-party guides that must be rechecked in a browser before anyone submits or pays. And every "dofollow" or "nofollow" claim on a launch platform in that inventory is the platform's or a guide's statement, never inspected page source. Directories describe their own links generously. Read the anchor.
What your agent does and what you decide
Agent work, with your tools: find the submission page and save it with the date; read the robots file and the listing template's indexing rules; open a competitor's listing and read the outbound link's rel tokens; note the fee, the review promise and any followed-link offer; find the published contact route; fill in the six gates; prepare the listing copy for your review.
Your decisions: whether a paid tier is an advertisement you want; whether the audience is yours; whether a ugc or nofollow listing is still worth having for the readers it reaches; and whether to submit. Submission is browser work with your account, on the directory's form. AgentLinkOps sends nothing and fills in no forms. The candidate evaluation guide covers the general review; the nofollow, sponsored and UGC guide covers what the attributes do and do not mean.
How AgentLinkOps records the result
When a listing goes live, register the listing page and your destination as a source-and-target pair. The candidate starts 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, the HTTP status and the standing note that JavaScript execution and visual visibility were not checked. The rel tokens on the listing's link are in that observation, so your record says ugc, nofollow, sponsored or nothing, from the page. The recall, verify and monitor pipeline carries a whole shortlist through that check in one bounded batch and converts a present result into a watch that keeps its candidate lineage.
Directories decay. A listing page that later returns 404, or that starts redirecting to the directory's home page, shows up in the history. A redirect to your own target is refused as evidence of a link. Confirmed loss needs two complete absent observations at least 30 minutes apart, and an unknown between them does not advance that clock. The monitoring guide covers cadence and events.
What can go wrong
The listing page is rendered client-side. Many directories are single-page applications. A plain fetch returns unknown with possible_render_required. Unknown never fails a run and never means the listing was removed. Have your agent open it in a browser and save the date.
The directory blocks fetches. Of 164 submission routes researched on September 15, 2026, 22 could not be verified because of a bot block, a rate limit or an origin error. The record keeps the best-known URL and says why. That is the pattern to copy: an unverified route is a route you have not read, and it stays labelled that way.
The robots file cannot be read. Under RFC 9309 a 4xx robots response other than 429 lets a crawler proceed, while 429, 5xx and a robots file served as HTML are refusals. The verifier follows that reading and reports which case it hit.
A followed link that is not. The directory says "dofollow"; the anchor says rel="nofollow ugc". The anchor is the fact.
A record older than its window. A platform-class record older than a quarter, or an editorial-route record older than 90 days, needs a fresh read before you rely on it. Show the age or take the record down; never let it persist silently.
Where this sits in the product today
Imports, selected candidate checks and placement monitoring run in the invitation-only pilot. Live supplier discovery is disabled in this build and stops with PROVIDER_NOT_CONFIGURED. No directory record is published. The opportunity library counts what is held per niche with its verification state and dates, and its agent-tooling directories niche holds no record yet and carries the disclosure that AgentLinkOps intends to list itself on those directories. The private registry stays private; a public record would be a re-verified observation of a page we read, never an export of that registry. AgentLinkOps sends nothing and submits nothing.
The listicle and review guide covers ranked lists that an editor writes, where you ask rather than submit.
Sources
- Spam policies for Google web search · Google Search Central · 2026-08-28
- Introduction to robots.txt · Google Search Central · 2025-12-10
- Block Search indexing with noindex · Google Search Central · 2025-12-10
- Qualify your outbound links to Google · Google Search Central · 2025-12-10
- RFC 9309: Robots Exclusion Protocol · IETF · 2022-09-01