Hiring is the buying signal most people take on trust. A contact database says a company posted a job in the last thirty days, the row goes into a campaign, and a first message goes out saying "saw you are hiring". The company's own careers page stays unopened, because the database already said yes.
We run outbound as an operator, on our own profiles and on client campaigns, and we check that box for a living. This article covers what the company's own board adds, how often that board exists at all, and the three checks that keep a hiring fact from turning into a fabricated one.
The numbers come from our work between 1 July and 6 September 2026, when we sourced 4,488 people, 193 of them from a company's own job board. Each coverage figure carries its own scan and denominator, because the average across them is meaningless: coverage swings by a factor of ten depending on which slice of the market you point it at.
The filter and the board answer different questions
The hiring filter inside a large contact database is a selection tool. It narrows a company list down to the ones that look active. That is genuinely useful, and it is where we start.
It becomes a problem the moment its output turns into a sentence in a message. We checked it against the company's own source seven times across three clients and three geographies, and seven times the company's own page disagreed. Two had a careers page stating plainly that there were no open roles. One had roles open, published three and five months earlier. One was hiring at volume and still had nothing quotable, because its careers page redirected to an enterprise system that returns an empty page without JavaScript.
There is a second, quieter gap. A database usually gives you the date it crawled, and that reads like the date the role went up. In July we sent a message describing a role as opened on 14 July. The company's own board carried 19 January for the same requisition. Six months of difference, inside a sentence built to prove we were paying attention.
So we split the two jobs. The filter picks candidates. The fact in the message comes from the company's own board, with the date that board prints. When the board is silent, the message names the role and stays quiet about timing, or it uses a different fact entirely.
Coverage is the whole game
The public board endpoints are easy to read. Greenhouse, Lever, Ashby, SmartRecruiters, Recruitee and Workable all publish an open JSON feed per company, with title, location, department and publish date. They open in a browser and cost nothing to read.
The hard part is that a large share of companies sit somewhere else. Here is what a dozen scans looked like over six weeks, each on a different slice:
| Slice scanned | Domains | Boards found |
|---|---|---|
| US SaaS and tech, central, south and west | 39 | 21 |
| US mid-market east coast: finance, media, biotech, consulting | 14 | 2 |
| IT outsourcing and nearshore dev, fresh slice | 60 | 14 |
| Continental Europe, mixed | 22 | 2 |
| DACH, Nordics and Ireland, second pass | 20 | 0 |
| Southern Europe and Poland | 39 | 4 |
| Classic European employers: banks, insurers, telecom, energy | 13 | 0 |
| Law firms of 11 to 50 people, Benelux | 19 | 0 |
| Creative and video agencies, 11 to 200 people | 54 | 3 |
Read the first two rows together: same tool, same day, 54% coverage in one slice and 14% in the next one over. The split follows the hiring stack, and the hiring stack follows company type more than country or size. American venture-backed companies buy the American set. European mid-market buys Personio, Teamtailor, Join and Factorial. Large enterprises sit on systems that render through JavaScript and stay closed to a plain reader. Professional firms under fifty people hire through their own site and LinkedIn, because an applicant tracking system is more machinery than they need.
Two practical consequences, and both are about planning.
First, hiring can only carry volume where coverage supports it. One of our campaign plans gave 40% of daily volume to the hiring line as the one signal reproducible every day. In a European territory that line produced zero usable events for three days running, because the boards sat on systems our collector cannot read. That is a planning error, and it looks like a sourcing failure for a week before anyone names it correctly.
Second, coverage is only half the multiplier. On one European territory a scan found boards at 11 of 39 companies, and all eleven had already been contacted under our one-person-per-company rule. A later pass on ten untouched companies from the same portrait found boards at zero of ten. The apparent coverage was sitting entirely on companies we had already worked. Count the overlap of "board exists" and "company still untouched", because that intersection is your real pool.
Guessing the careers URL is a poor substitute, by the way. We tried the obvious patterns on eight companies, and six returned a 404 or bounced to a portal carrying no roles of its own.
A board found automatically can belong to someone else
Board discovery works by matching a piece of the domain to a board name, and short brand words collide.
In September a domain belonging to a hundred person AI company resolved to a board owned by a global consultancy sharing the same brand word. Fifteen roles came back: insolvency, corporate reputation advisory, financial advisory, in Sydney, Dublin and Hong Kong. Had that gone out untouched, a founder in Stockholm would have received a message about hiring that belonged to a different company entirely. That is the most expensive class of outbound error, because it burns the single first attempt and proves to the reader that a machine wrote it. A second collision put a job aggregator's board behind a small design studio's domain and returned 162 roles from all over the world.
There is a date trap as well. At least one system returns records with an empty publish date, and a thirty day filter passes those rows straight through, because there is nothing in them to compare against. One scan brought back nineteen dateless roles from a single company and treated them as recent.
Three checks close all of it, and together they take about a minute per company:
- Subject match. Do the roles describe the business you think you are looking at? Consulting roles on the board of an AI startup means a different company.
- Geography match. Do the cities match the company's footprint? A different continent means a different company.
- Date present. A role with an empty publish date is a role you can describe as open and nothing more. Keep the timing out of the sentence.
The role you can quote is rarely the role of your buyer
We scanned 37 domains on the US east coast for a campaign whose buyer owns workplace and office decisions. 19 of those domains had a live board, 51%, among the best coverage of any slice we measured. We then filtered those boards for roles in the buyer's own function: workplace, office, facilities, receptionist, people operations. Across all 19 boards the count was zero. Without the filter, the same 19 boards held 247 open roles.
The boards were full. The buyer's own function appears in them once every few years, because that is how often such a person gets hired. A search built on "they are hiring someone like you" reports a dry source and gets abandoned, while 247 live requisitions sit one query away.
The event was in the shape of the requisitions. One company posted a single role across four cities in one line. Another put up to seven sites in a single requisition. A third opened the same role in three cities on the same day. For a buyer who has to seat people where the company holds no lease, that beats any headcount number, and it comes with a link and a date.
So scan without the role filter, then read the distribution. Which departments are growing, in how many cities, how recently. The fact you quote should be one the reader recognises as their own week.
What it is actually worth
Honest numbers, with their samples attached, from 1 July to 6 September 2026.
People sourced from a company's own board: 193. Of those, 125 received a first touch, 19 accepted the connection, 3 replied. From the database hiring filter: 158 sourced, 95 touched, 20 accepted, 3 replied. From our largest source, plain portrait search with a fact taken from the person's own profile: 2,273 sourced, 1,480 touched, 290 accepted, 33 replied.
Reply rates of 2.4%, 3.2% and 2.2%. At three replies per group those numbers sit on top of each other, and a winner declared from that sample is a winner read out of noise. Where the source of the fact shows up in reply rate, we have yet to measure it, and we will say so when the volumes get there.
What the board buys sits upstream of reply rate: it decides whether the sentence you send is checkable by the person reading it. In the same period our enrichment step dropped candidates under three separate reasons that all mean the same thing, a fact that was missing, unverifiable, or too thin to check. 25, 21 and 16 people. Each of them would otherwise have received a message built on a claim we could not source, and a share of those claims would have been wrong in front of the buyer.
What to do with this on Monday
Take twenty target domains and run them through the six public endpoints. Count how many return a board. That count is the ceiling on your hiring line, and it takes twenty minutes to establish. Under a third, plan hiring as a top-up and give the volume to a signal with a live source.
Where a board answers, read the requisition rather than the count: cities, departments, and the date the board prints. Check subject and geography before you quote anything. When the board stays quiet, take the fact from something the person published themselves, which exists for every company in every country and costs nothing to check.
We build these lines for a living, one signal at a time, and we publish the shape of them in our outbound playbooks. If you would rather point your own model at the sourcing stack and run the scans yourself, that is what the MCP level is for.