Link building platform with autopilot outreach and link monitoring

A link building platform that runs the outreach loop itself

Pressmerit is a link building platform for teams who need editorial placements and do not want to spend their week on spreadsheets. It reads a site, works out what it is about, drafts the keywords and pages worth pointing links at, then goes looking for relevant publishers on the open web. When it finds a good match it identifies the right person there, verifies the address, and writes a pitch about that specific site. The user's remaining job is to read the draft and press send from their own mailbox.

That last part is a deliberate constraint rather than a missing feature. Sending from the user's own inbox means an authentic sender, no cold-email infrastructure to warm up, and no shared sending reputation to protect. It also means the platform cannot see delivery, which is why it never charges for a send.

The second half of the product starts after a link is published. Most tools treat a placement as the finish line and hand you a screenshot. Pressmerit fetches the page, confirms the link actually resolves to the agreed target, checks that the rel attribute and the page's indexability match what the two sides agreed, and then keeps checking on a schedule for as long as the placement matters. Monitoring costs no credits, because a link that quietly disappeared six months later is the failure mode the whole category is bad at.

Publishers get their own side of the product for free, permanently. They receive an outreach email, follow a claim link, and land on the single request that concerns them without creating an account first. If they want to keep the relationship going, the account is there. That asymmetry is the growth engine: every pitch a user sends is also an invitation for a publisher to join the supply side.

Why catalog marketplaces and outreach tools both leave a gap

The market splits into two camps, and neither one covers the whole job.

Catalogue marketplaces have inventory. You browse publishers who registered themselves, pick one, and pay a placement fee with a commission on top. The catalogue is the ceiling. If the right site for your niche never signed up, it does not exist as far as the product is concerned, and the sites that did sign up have usually signed up everywhere else too. That is a supply problem dressed as a convenience.

Outreach tools have workflow. You get sequences, an inbox and a pipeline view, and what that generation of product calls outreach automation is really send automation. It still expects you to arrive with your own prospects, your own metrics subscription, and your own mailbox. The good ones cost a few hundred dollars a month before you have added a single contact. A blogger outreach tool of that shape is a surface for work you have already done elsewhere, and nothing in it knows whether the link you chased for three weeks is still on the page.

The split shows up cleanly when you line the three models up against each other.

Between the two sits the part nobody owns. Link prospecting on the open web is expensive research. Working out who to write to is a second research problem. Writing something a busy editor will actually read is a third, and confirming that the result held up is a fourth that most teams quietly skip.

Two constraints shaped the commercial design. A per-placement commission turns the platform into an intermediary moving money between its own users, which drags in escrow and payout obligations no MVP should be carrying. Seat-based pricing punishes exactly the small teams the product is for. So the money question got settled early: one subscription at $49 a month, 500 credits, three sites, and a printed credit price on every button that spends them. No commission on anything the two parties agree between themselves.

The other constraint shaped the vocabulary as much as the code. Google's spam policies name buying and selling links for ranking purposes, and excessive reciprocal linking, as link spam, and its guidance on qualifying outbound links is where the rel values in the product come from. Pressmerit is therefore built and described as a digital PR platform that brokers editorial collaborations, with the rel attribute recorded as a term the two parties agree. There is no price-per-link catalogue anywhere in it, and there was never going to be one.

What we built: open-web discovery, autopilot outreach, and a proprietary Domain Score

The product is organised around one idea. The user makes the judgment calls that are genuinely theirs, which sites, how much budget, and what to say in a negotiation, and the platform does everything else while spending a budget the user set.

Adding a site triggers an analysis pass: a crawl of up to twenty-five pages, a classification call that assigns a primary industry from a fixed twenty-six-value taxonomy, language and region detection, an authority score, and a generated keyword set split across commercial, branded, and informational intent. Target pages are picked automatically from the crawl by page type and embedding proximity to those keywords. Every one of those outputs is editable, and none of them has to be reviewed for the product to keep moving.

From there autopilot takes the wheel. It keeps a working inventory of scored prospects, tops it up with a discovery run when it drops too low, prepares outreach for the best matches inside the monthly credit budget, and stops at anything that needs a human. A borderline match waits in a short list instead of being decided. A publisher asking for different terms waits for a reply. What actually reaches the user is a queue of drafts, each one needing a single decision.

Authority needed its own answer. Reselling a vendor's branded metric brings licensing exposure and leaves the product describing itself in someone else's vocabulary, so Domain Score is computed in-house from four inputs: link authority at 45 percent, an organic-traffic estimate at 25 percent, referring-domain breadth at 20 percent, and a naturalness signal at 10 percent. The weighting is opinionated on purpose. Link authority dominates because it is closest to what a placement is worth, while traffic and breadth are what separate a real publication from a domain with a bought-up rank.

Architecture: matching engine, credit ledger, and backlink monitoring

The system is a Turborepo monorepo: a Next.js App Router front end, a NestJS API that also hosts the realtime gateway, and a standalone worker service running BullMQ processors. The web app never touches the database. It talks to the API over REST and WebSocket using types shared from one package, which keeps a whole category of drift from ever starting.

Nearly everything expensive happens in the background. Separate queues handle analysis, discovery, contact lookup, copy generation, link checks, billing, and the autopilot sweep, so a large discovery run cannot starve onboarding or a renewal charge. Job IDs are deterministic and every processor re-verifies database state on entry, which makes a duplicate or stale job a no-op rather than a double charge.

How the match score ranks link prospects

Link prospecting starts wide and narrows fast, because the cost per candidate climbs sharply at each step. Up to ten approved keywords are expanded into four query templates each and run against up to two locales, returning several hundred candidate domains per run. Those get deduped to registrable domains, then filtered against a platform blocklist, the account's own exclusions, and any domain already in an open thread. A cosine-similarity check drops near-identical commercial competitors, since a rival SaaS is not going to link to you.

Only what survives gets enriched, and enrichment is cache-first. A short crawl, a classification pass, an authority fetch, a deterministic spam score, and a 1024-dimension embedding stored in pgvector. Anything scoring 50 or above on spam risk is rejected outright and remembered, so the platform never pays to evaluate it twice.

The surviving candidates are ranked by a match score weighted 40 percent topical similarity, 20 percent authority fit, 15 percent industry, 15 percent region, and 10 percent language, then multiplied down by whatever spam risk remains. Only prospects clearing 55 are delivered. Because the score components are stored as structured data, the interface can show the user why a domain ranked where it did instead of asking them to trust a number.

The donor cache is what makes this economics work. Domains are stored platform-wide with per-facet freshness: authority refreshes on a 30-day cycle, classification on 90, contacts on 60. A domain another user's run already enriched costs nothing to reuse. The cache compounds with every run anyone makes.

Link verification and 30-day backlink monitoring

Verification parses server-rendered HTML with cheerio and no headless browser at all. A link that only appears after client-side script has dubious value, and the publisher-facing error says exactly that. The check normalizes both URLs before comparing them, stripping tracking parameters, fragments, trailing slashes, and the www prefix, then confirms an anchor genuinely resolves to the agreed target. It reads the rel attribute against the policy the two parties recorded, treating nofollow, sponsored and ugc as the same class of signal rather than checking only for nofollow, and it looks for a page-level noindex in the meta robots tag or the X-Robots-Tag header, because a clean anchor on a page nobody indexes passes nothing.

Most of the hard work here went into not crying wolf, which is the first thing anyone who has run a monitoring tool asks about. A false "your link is gone" costs a publisher relationship, so the checker is deliberately reluctant to declare one. It follows up to five redirects and judges the final URL rather than the one submitted. A canonical pointing elsewhere is recorded as a warning, not a failure, since that is an editorial decision and not a broken placement. Network errors, timeouts, rate limiting and bot blocks are all classed as transient: they retry twice within a few hours, roll into the next day, and only count against the placement after three consecutive days of the same result. A content failure on a weekly check triggers two daily rechecks before anything is declared removed, and both sides get the last known good snapshot when it is.

The rest of the failure grading follows the same logic. A missing link is conclusive. Anchor text that drifted from the agreement is a warning to the requester, not a failure, because publishers edit for editorial reasons and hard-failing them burns the relationship the product exists to build.

Credits run through an append-only ledger with two buckets and a hold-then-capture model, the metered half of the hybrid we describe in our guide to SaaS pricing models. A discovery run holds its full price up front and captures per prospect actually delivered against a run minimum, releasing the rest automatically. A contact lookup that finds nobody releases in full. Billing runs on Stripe with card capture and off-session charges rather than Stripe Billing, so the renewal schedule, the dunning attempts, and the grace window are all database state we control and can reason about.

Results: link building automation with published unit economics

Pressmerit went live at pressmerit.com as a single-plan link building platform, with a credit economy sized so the included allowance covers real work rather than a demo. Two sites analysed, forty prospects found, and forty outreach requests prepared, all inside the monthly grant.

The number the product is built to move is cost per verified live link. At benchmark reply rates, the published estimate runs $12 to $30 against the $200 to $500 a placement typically costs through an agency. Pressmerit publishes those as estimates from its own unit economics with the assumptions labelled, which is a rarer thing in this category than it should be.

Two design decisions carry most of that gap. Contact lookups happen lazily, at the moment a user decides to pursue a prospect rather than eagerly for every search result, which is the difference between paying cents and paying fractions of a cent per candidate. And the platform-wide donor cache means the expensive part of evaluating a domain is paid once for everyone, not once per customer.

Monitoring is included rather than sold as link insurance, which several competitors treat as an upsell. Every placement is verified on publication, re-checked weekly through its first month, and re-checked every thirty days after that, at no credit cost. When a link disappears, both sides hear about it and asking for it back is one click.

The publisher side stays free permanently, which is not generosity. Supply is the constraint in this market, and friction on the supply side is what kills a two-sided product before it starts.

Backlink management software for agencies, SaaS teams, and publishers

One link building platform, four working patterns. The matching engine, the credit ledger, and the monitoring loop stay the same underneath while the shape of the work on top changes with who is doing it.

The engineering decisions behind a link building platform of this shape, how a metered credit economy stays auditable, where the tenancy boundaries go, and which half of a two-sided product to launch first, are the ones we work through in our guide to how to build a SaaS and its companion on multi-tenant SaaS architecture. Scoping a product like it is SaaS development work with a heavy AI software development component sitting in the matching layer, and carrying a working MVP through to something that holds under real load is product development. The scoring model at the centre of it is the same class of problem we solved for Ugreator.